4 # Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011, 2012 Google Inc.
6 # This program is free software; you can redistribute it and/or modify
7 # it under the terms of the GNU General Public License as published by
8 # the Free Software Foundation; either version 2 of the License, or
9 # (at your option) any later version.
11 # This program is distributed in the hope that it will be useful, but
12 # WITHOUT ANY WARRANTY; without even the implied warranty of
13 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 # General Public License for more details.
16 # You should have received a copy of the GNU General Public License
17 # along with this program; if not, write to the Free Software
18 # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
22 """Module implementing the master-side code."""
24 # pylint: disable=W0201,C0302
26 # W0201 since most LU attributes are defined in CheckPrereq or similar
29 # C0302: since we have waaaay too many lines in this module
44 from ganeti import ssh
45 from ganeti import utils
46 from ganeti import errors
47 from ganeti import hypervisor
48 from ganeti import locking
49 from ganeti import constants
50 from ganeti import objects
51 from ganeti import serializer
52 from ganeti import ssconf
53 from ganeti import uidpool
54 from ganeti import compat
55 from ganeti import masterd
56 from ganeti import netutils
57 from ganeti import query
58 from ganeti import qlang
59 from ganeti import opcodes
61 from ganeti import rpc
62 from ganeti import runtime
64 import ganeti.masterd.instance # pylint: disable=W0611
67 #: Size of DRBD meta block device
71 INSTANCE_DOWN = [constants.ADMINST_DOWN]
72 INSTANCE_ONLINE = [constants.ADMINST_DOWN, constants.ADMINST_UP]
73 INSTANCE_NOT_RUNNING = [constants.ADMINST_DOWN, constants.ADMINST_OFFLINE]
75 #: Instance status in which an instance can be marked as offline/online
76 CAN_CHANGE_INSTANCE_OFFLINE = (frozenset(INSTANCE_DOWN) | frozenset([
77 constants.ADMINST_OFFLINE,
82 """Data container for LU results with jobs.
84 Instances of this class returned from L{LogicalUnit.Exec} will be recognized
85 by L{mcpu._ProcessResult}. The latter will then submit the jobs
86 contained in the C{jobs} attribute and include the job IDs in the opcode
90 def __init__(self, jobs, **kwargs):
91 """Initializes this class.
93 Additional return values can be specified as keyword arguments.
95 @type jobs: list of lists of L{opcode.OpCode}
96 @param jobs: A list of lists of opcode objects
103 class LogicalUnit(object):
104 """Logical Unit base class.
106 Subclasses must follow these rules:
107 - implement ExpandNames
108 - implement CheckPrereq (except when tasklets are used)
109 - implement Exec (except when tasklets are used)
110 - implement BuildHooksEnv
111 - implement BuildHooksNodes
112 - redefine HPATH and HTYPE
113 - optionally redefine their run requirements:
114 REQ_BGL: the LU needs to hold the Big Ganeti Lock exclusively
116 Note that all commands require root permissions.
118 @ivar dry_run_result: the value (if any) that will be returned to the caller
119 in dry-run mode (signalled by opcode dry_run parameter)
126 def __init__(self, processor, op, context, rpc_runner):
127 """Constructor for LogicalUnit.
129 This needs to be overridden in derived classes in order to check op
133 self.proc = processor
135 self.cfg = context.cfg
136 self.glm = context.glm
138 self.owned_locks = context.glm.list_owned
139 self.context = context
140 self.rpc = rpc_runner
141 # Dicts used to declare locking needs to mcpu
142 self.needed_locks = None
143 self.share_locks = dict.fromkeys(locking.LEVELS, 0)
145 self.remove_locks = {}
146 # Used to force good behavior when calling helper functions
147 self.recalculate_locks = {}
149 self.Log = processor.Log # pylint: disable=C0103
150 self.LogWarning = processor.LogWarning # pylint: disable=C0103
151 self.LogInfo = processor.LogInfo # pylint: disable=C0103
152 self.LogStep = processor.LogStep # pylint: disable=C0103
153 # support for dry-run
154 self.dry_run_result = None
155 # support for generic debug attribute
156 if (not hasattr(self.op, "debug_level") or
157 not isinstance(self.op.debug_level, int)):
158 self.op.debug_level = 0
163 # Validate opcode parameters and set defaults
164 self.op.Validate(True)
166 self.CheckArguments()
168 def CheckArguments(self):
169 """Check syntactic validity for the opcode arguments.
171 This method is for doing a simple syntactic check and ensure
172 validity of opcode parameters, without any cluster-related
173 checks. While the same can be accomplished in ExpandNames and/or
174 CheckPrereq, doing these separate is better because:
176 - ExpandNames is left as as purely a lock-related function
177 - CheckPrereq is run after we have acquired locks (and possible
180 The function is allowed to change the self.op attribute so that
181 later methods can no longer worry about missing parameters.
186 def ExpandNames(self):
187 """Expand names for this LU.
189 This method is called before starting to execute the opcode, and it should
190 update all the parameters of the opcode to their canonical form (e.g. a
191 short node name must be fully expanded after this method has successfully
192 completed). This way locking, hooks, logging, etc. can work correctly.
194 LUs which implement this method must also populate the self.needed_locks
195 member, as a dict with lock levels as keys, and a list of needed lock names
198 - use an empty dict if you don't need any lock
199 - if you don't need any lock at a particular level omit that
200 level (note that in this case C{DeclareLocks} won't be called
201 at all for that level)
202 - if you need locks at a level, but you can't calculate it in
203 this function, initialise that level with an empty list and do
204 further processing in L{LogicalUnit.DeclareLocks} (see that
205 function's docstring)
206 - don't put anything for the BGL level
207 - if you want all locks at a level use L{locking.ALL_SET} as a value
209 If you need to share locks (rather than acquire them exclusively) at one
210 level you can modify self.share_locks, setting a true value (usually 1) for
211 that level. By default locks are not shared.
213 This function can also define a list of tasklets, which then will be
214 executed in order instead of the usual LU-level CheckPrereq and Exec
215 functions, if those are not defined by the LU.
219 # Acquire all nodes and one instance
220 self.needed_locks = {
221 locking.LEVEL_NODE: locking.ALL_SET,
222 locking.LEVEL_INSTANCE: ['instance1.example.com'],
224 # Acquire just two nodes
225 self.needed_locks = {
226 locking.LEVEL_NODE: ['node1.example.com', 'node2.example.com'],
229 self.needed_locks = {} # No, you can't leave it to the default value None
232 # The implementation of this method is mandatory only if the new LU is
233 # concurrent, so that old LUs don't need to be changed all at the same
236 self.needed_locks = {} # Exclusive LUs don't need locks.
238 raise NotImplementedError
240 def DeclareLocks(self, level):
241 """Declare LU locking needs for a level
243 While most LUs can just declare their locking needs at ExpandNames time,
244 sometimes there's the need to calculate some locks after having acquired
245 the ones before. This function is called just before acquiring locks at a
246 particular level, but after acquiring the ones at lower levels, and permits
247 such calculations. It can be used to modify self.needed_locks, and by
248 default it does nothing.
250 This function is only called if you have something already set in
251 self.needed_locks for the level.
253 @param level: Locking level which is going to be locked
254 @type level: member of L{ganeti.locking.LEVELS}
258 def CheckPrereq(self):
259 """Check prerequisites for this LU.
261 This method should check that the prerequisites for the execution
262 of this LU are fulfilled. It can do internode communication, but
263 it should be idempotent - no cluster or system changes are
266 The method should raise errors.OpPrereqError in case something is
267 not fulfilled. Its return value is ignored.
269 This method should also update all the parameters of the opcode to
270 their canonical form if it hasn't been done by ExpandNames before.
273 if self.tasklets is not None:
274 for (idx, tl) in enumerate(self.tasklets):
275 logging.debug("Checking prerequisites for tasklet %s/%s",
276 idx + 1, len(self.tasklets))
281 def Exec(self, feedback_fn):
284 This method should implement the actual work. It should raise
285 errors.OpExecError for failures that are somewhat dealt with in
289 if self.tasklets is not None:
290 for (idx, tl) in enumerate(self.tasklets):
291 logging.debug("Executing tasklet %s/%s", idx + 1, len(self.tasklets))
294 raise NotImplementedError
296 def BuildHooksEnv(self):
297 """Build hooks environment for this LU.
300 @return: Dictionary containing the environment that will be used for
301 running the hooks for this LU. The keys of the dict must not be prefixed
302 with "GANETI_"--that'll be added by the hooks runner. The hooks runner
303 will extend the environment with additional variables. If no environment
304 should be defined, an empty dictionary should be returned (not C{None}).
305 @note: If the C{HPATH} attribute of the LU class is C{None}, this function
309 raise NotImplementedError
311 def BuildHooksNodes(self):
312 """Build list of nodes to run LU's hooks.
314 @rtype: tuple; (list, list)
315 @return: Tuple containing a list of node names on which the hook
316 should run before the execution and a list of node names on which the
317 hook should run after the execution. No nodes should be returned as an
318 empty list (and not None).
319 @note: If the C{HPATH} attribute of the LU class is C{None}, this function
323 raise NotImplementedError
325 def HooksCallBack(self, phase, hook_results, feedback_fn, lu_result):
326 """Notify the LU about the results of its hooks.
328 This method is called every time a hooks phase is executed, and notifies
329 the Logical Unit about the hooks' result. The LU can then use it to alter
330 its result based on the hooks. By default the method does nothing and the
331 previous result is passed back unchanged but any LU can define it if it
332 wants to use the local cluster hook-scripts somehow.
334 @param phase: one of L{constants.HOOKS_PHASE_POST} or
335 L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
336 @param hook_results: the results of the multi-node hooks rpc call
337 @param feedback_fn: function used send feedback back to the caller
338 @param lu_result: the previous Exec result this LU had, or None
340 @return: the new Exec result, based on the previous result
344 # API must be kept, thus we ignore the unused argument and could
345 # be a function warnings
346 # pylint: disable=W0613,R0201
349 def _ExpandAndLockInstance(self):
350 """Helper function to expand and lock an instance.
352 Many LUs that work on an instance take its name in self.op.instance_name
353 and need to expand it and then declare the expanded name for locking. This
354 function does it, and then updates self.op.instance_name to the expanded
355 name. It also initializes needed_locks as a dict, if this hasn't been done
359 if self.needed_locks is None:
360 self.needed_locks = {}
362 assert locking.LEVEL_INSTANCE not in self.needed_locks, \
363 "_ExpandAndLockInstance called with instance-level locks set"
364 self.op.instance_name = _ExpandInstanceName(self.cfg,
365 self.op.instance_name)
366 self.needed_locks[locking.LEVEL_INSTANCE] = self.op.instance_name
368 def _LockInstancesNodes(self, primary_only=False,
369 level=locking.LEVEL_NODE):
370 """Helper function to declare instances' nodes for locking.
372 This function should be called after locking one or more instances to lock
373 their nodes. Its effect is populating self.needed_locks[locking.LEVEL_NODE]
374 with all primary or secondary nodes for instances already locked and
375 present in self.needed_locks[locking.LEVEL_INSTANCE].
377 It should be called from DeclareLocks, and for safety only works if
378 self.recalculate_locks[locking.LEVEL_NODE] is set.
380 In the future it may grow parameters to just lock some instance's nodes, or
381 to just lock primaries or secondary nodes, if needed.
383 If should be called in DeclareLocks in a way similar to::
385 if level == locking.LEVEL_NODE:
386 self._LockInstancesNodes()
388 @type primary_only: boolean
389 @param primary_only: only lock primary nodes of locked instances
390 @param level: Which lock level to use for locking nodes
393 assert level in self.recalculate_locks, \
394 "_LockInstancesNodes helper function called with no nodes to recalculate"
396 # TODO: check if we're really been called with the instance locks held
398 # For now we'll replace self.needed_locks[locking.LEVEL_NODE], but in the
399 # future we might want to have different behaviors depending on the value
400 # of self.recalculate_locks[locking.LEVEL_NODE]
402 locked_i = self.owned_locks(locking.LEVEL_INSTANCE)
403 for _, instance in self.cfg.GetMultiInstanceInfo(locked_i):
404 wanted_nodes.append(instance.primary_node)
406 wanted_nodes.extend(instance.secondary_nodes)
408 if self.recalculate_locks[level] == constants.LOCKS_REPLACE:
409 self.needed_locks[level] = wanted_nodes
410 elif self.recalculate_locks[level] == constants.LOCKS_APPEND:
411 self.needed_locks[level].extend(wanted_nodes)
413 raise errors.ProgrammerError("Unknown recalculation mode")
415 del self.recalculate_locks[level]
418 class NoHooksLU(LogicalUnit): # pylint: disable=W0223
419 """Simple LU which runs no hooks.
421 This LU is intended as a parent for other LogicalUnits which will
422 run no hooks, in order to reduce duplicate code.
428 def BuildHooksEnv(self):
429 """Empty BuildHooksEnv for NoHooksLu.
431 This just raises an error.
434 raise AssertionError("BuildHooksEnv called for NoHooksLUs")
436 def BuildHooksNodes(self):
437 """Empty BuildHooksNodes for NoHooksLU.
440 raise AssertionError("BuildHooksNodes called for NoHooksLU")
444 """Tasklet base class.
446 Tasklets are subcomponents for LUs. LUs can consist entirely of tasklets or
447 they can mix legacy code with tasklets. Locking needs to be done in the LU,
448 tasklets know nothing about locks.
450 Subclasses must follow these rules:
451 - Implement CheckPrereq
455 def __init__(self, lu):
462 def CheckPrereq(self):
463 """Check prerequisites for this tasklets.
465 This method should check whether the prerequisites for the execution of
466 this tasklet are fulfilled. It can do internode communication, but it
467 should be idempotent - no cluster or system changes are allowed.
469 The method should raise errors.OpPrereqError in case something is not
470 fulfilled. Its return value is ignored.
472 This method should also update all parameters to their canonical form if it
473 hasn't been done before.
478 def Exec(self, feedback_fn):
479 """Execute the tasklet.
481 This method should implement the actual work. It should raise
482 errors.OpExecError for failures that are somewhat dealt with in code, or
486 raise NotImplementedError
490 """Base for query utility classes.
493 #: Attribute holding field definitions
499 def __init__(self, qfilter, fields, use_locking):
500 """Initializes this class.
503 self.use_locking = use_locking
505 self.query = query.Query(self.FIELDS, fields, qfilter=qfilter,
506 namefield=self.SORT_FIELD)
507 self.requested_data = self.query.RequestedData()
508 self.names = self.query.RequestedNames()
510 # Sort only if no names were requested
511 self.sort_by_name = not self.names
513 self.do_locking = None
516 def _GetNames(self, lu, all_names, lock_level):
517 """Helper function to determine names asked for in the query.
521 names = lu.owned_locks(lock_level)
525 if self.wanted == locking.ALL_SET:
526 assert not self.names
527 # caller didn't specify names, so ordering is not important
528 return utils.NiceSort(names)
530 # caller specified names and we must keep the same order
532 assert not self.do_locking or lu.glm.is_owned(lock_level)
534 missing = set(self.wanted).difference(names)
536 raise errors.OpExecError("Some items were removed before retrieving"
537 " their data: %s" % missing)
539 # Return expanded names
542 def ExpandNames(self, lu):
543 """Expand names for this query.
545 See L{LogicalUnit.ExpandNames}.
548 raise NotImplementedError()
550 def DeclareLocks(self, lu, level):
551 """Declare locks for this query.
553 See L{LogicalUnit.DeclareLocks}.
556 raise NotImplementedError()
558 def _GetQueryData(self, lu):
559 """Collects all data for this query.
561 @return: Query data object
564 raise NotImplementedError()
566 def NewStyleQuery(self, lu):
567 """Collect data and execute query.
570 return query.GetQueryResponse(self.query, self._GetQueryData(lu),
571 sort_by_name=self.sort_by_name)
573 def OldStyleQuery(self, lu):
574 """Collect data and execute query.
577 return self.query.OldStyleQuery(self._GetQueryData(lu),
578 sort_by_name=self.sort_by_name)
582 """Returns a dict declaring all lock levels shared.
585 return dict.fromkeys(locking.LEVELS, 1)
588 def _MakeLegacyNodeInfo(data):
589 """Formats the data returned by L{rpc.RpcRunner.call_node_info}.
591 Converts the data into a single dictionary. This is fine for most use cases,
592 but some require information from more than one volume group or hypervisor.
595 (bootid, (vg_info, ), (hv_info, )) = data
597 return utils.JoinDisjointDicts(utils.JoinDisjointDicts(vg_info, hv_info), {
602 def _AnnotateDiskParams(instance, devs, cfg):
603 """Little helper wrapper to the rpc annotation method.
605 @param instance: The instance object
606 @type devs: List of L{objects.Disk}
607 @param devs: The root devices (not any of its children!)
608 @param cfg: The config object
609 @returns The annotated disk copies
610 @see L{rpc.AnnotateDiskParams}
613 return rpc.AnnotateDiskParams(instance.disk_template, devs,
614 cfg.GetInstanceDiskParams(instance))
617 def _CheckInstancesNodeGroups(cfg, instances, owned_groups, owned_nodes,
619 """Checks if node groups for locked instances are still correct.
621 @type cfg: L{config.ConfigWriter}
622 @param cfg: Cluster configuration
623 @type instances: dict; string as key, L{objects.Instance} as value
624 @param instances: Dictionary, instance name as key, instance object as value
625 @type owned_groups: iterable of string
626 @param owned_groups: List of owned groups
627 @type owned_nodes: iterable of string
628 @param owned_nodes: List of owned nodes
629 @type cur_group_uuid: string or None
630 @param cur_group_uuid: Optional group UUID to check against instance's groups
633 for (name, inst) in instances.items():
634 assert owned_nodes.issuperset(inst.all_nodes), \
635 "Instance %s's nodes changed while we kept the lock" % name
637 inst_groups = _CheckInstanceNodeGroups(cfg, name, owned_groups)
639 assert cur_group_uuid is None or cur_group_uuid in inst_groups, \
640 "Instance %s has no node in group %s" % (name, cur_group_uuid)
643 def _CheckInstanceNodeGroups(cfg, instance_name, owned_groups):
644 """Checks if the owned node groups are still correct for an instance.
646 @type cfg: L{config.ConfigWriter}
647 @param cfg: The cluster configuration
648 @type instance_name: string
649 @param instance_name: Instance name
650 @type owned_groups: set or frozenset
651 @param owned_groups: List of currently owned node groups
654 inst_groups = cfg.GetInstanceNodeGroups(instance_name)
656 if not owned_groups.issuperset(inst_groups):
657 raise errors.OpPrereqError("Instance %s's node groups changed since"
658 " locks were acquired, current groups are"
659 " are '%s', owning groups '%s'; retry the"
662 utils.CommaJoin(inst_groups),
663 utils.CommaJoin(owned_groups)),
669 def _CheckNodeGroupInstances(cfg, group_uuid, owned_instances):
670 """Checks if the instances in a node group are still correct.
672 @type cfg: L{config.ConfigWriter}
673 @param cfg: The cluster configuration
674 @type group_uuid: string
675 @param group_uuid: Node group UUID
676 @type owned_instances: set or frozenset
677 @param owned_instances: List of currently owned instances
680 wanted_instances = cfg.GetNodeGroupInstances(group_uuid)
681 if owned_instances != wanted_instances:
682 raise errors.OpPrereqError("Instances in node group '%s' changed since"
683 " locks were acquired, wanted '%s', have '%s';"
684 " retry the operation" %
686 utils.CommaJoin(wanted_instances),
687 utils.CommaJoin(owned_instances)),
690 return wanted_instances
693 def _SupportsOob(cfg, node):
694 """Tells if node supports OOB.
696 @type cfg: L{config.ConfigWriter}
697 @param cfg: The cluster configuration
698 @type node: L{objects.Node}
699 @param node: The node
700 @return: The OOB script if supported or an empty string otherwise
703 return cfg.GetNdParams(node)[constants.ND_OOB_PROGRAM]
706 def _GetWantedNodes(lu, nodes):
707 """Returns list of checked and expanded node names.
709 @type lu: L{LogicalUnit}
710 @param lu: the logical unit on whose behalf we execute
712 @param nodes: list of node names or None for all nodes
714 @return: the list of nodes, sorted
715 @raise errors.ProgrammerError: if the nodes parameter is wrong type
719 return [_ExpandNodeName(lu.cfg, name) for name in nodes]
721 return utils.NiceSort(lu.cfg.GetNodeList())
724 def _GetWantedInstances(lu, instances):
725 """Returns list of checked and expanded instance names.
727 @type lu: L{LogicalUnit}
728 @param lu: the logical unit on whose behalf we execute
729 @type instances: list
730 @param instances: list of instance names or None for all instances
732 @return: the list of instances, sorted
733 @raise errors.OpPrereqError: if the instances parameter is wrong type
734 @raise errors.OpPrereqError: if any of the passed instances is not found
738 wanted = [_ExpandInstanceName(lu.cfg, name) for name in instances]
740 wanted = utils.NiceSort(lu.cfg.GetInstanceList())
744 def _GetUpdatedParams(old_params, update_dict,
745 use_default=True, use_none=False):
746 """Return the new version of a parameter dictionary.
748 @type old_params: dict
749 @param old_params: old parameters
750 @type update_dict: dict
751 @param update_dict: dict containing new parameter values, or
752 constants.VALUE_DEFAULT to reset the parameter to its default
754 @param use_default: boolean
755 @type use_default: whether to recognise L{constants.VALUE_DEFAULT}
756 values as 'to be deleted' values
757 @param use_none: boolean
758 @type use_none: whether to recognise C{None} values as 'to be
761 @return: the new parameter dictionary
764 params_copy = copy.deepcopy(old_params)
765 for key, val in update_dict.iteritems():
766 if ((use_default and val == constants.VALUE_DEFAULT) or
767 (use_none and val is None)):
773 params_copy[key] = val
777 def _GetUpdatedIPolicy(old_ipolicy, new_ipolicy, group_policy=False):
778 """Return the new version of a instance policy.
780 @param group_policy: whether this policy applies to a group and thus
781 we should support removal of policy entries
784 use_none = use_default = group_policy
785 ipolicy = copy.deepcopy(old_ipolicy)
786 for key, value in new_ipolicy.items():
787 if key not in constants.IPOLICY_ALL_KEYS:
788 raise errors.OpPrereqError("Invalid key in new ipolicy: %s" % key,
790 if key in constants.IPOLICY_ISPECS:
791 utils.ForceDictType(value, constants.ISPECS_PARAMETER_TYPES)
792 ipolicy[key] = _GetUpdatedParams(old_ipolicy.get(key, {}), value,
794 use_default=use_default)
796 if (not value or value == [constants.VALUE_DEFAULT] or
797 value == constants.VALUE_DEFAULT):
801 raise errors.OpPrereqError("Can't unset ipolicy attribute '%s'"
802 " on the cluster'" % key,
805 if key in constants.IPOLICY_PARAMETERS:
806 # FIXME: we assume all such values are float
808 ipolicy[key] = float(value)
809 except (TypeError, ValueError), err:
810 raise errors.OpPrereqError("Invalid value for attribute"
811 " '%s': '%s', error: %s" %
812 (key, value, err), errors.ECODE_INVAL)
814 # FIXME: we assume all others are lists; this should be redone
816 ipolicy[key] = list(value)
818 objects.InstancePolicy.CheckParameterSyntax(ipolicy)
819 except errors.ConfigurationError, err:
820 raise errors.OpPrereqError("Invalid instance policy: %s" % err,
825 def _UpdateAndVerifySubDict(base, updates, type_check):
826 """Updates and verifies a dict with sub dicts of the same type.
828 @param base: The dict with the old data
829 @param updates: The dict with the new data
830 @param type_check: Dict suitable to ForceDictType to verify correct types
831 @returns: A new dict with updated and verified values
835 new = _GetUpdatedParams(old, value)
836 utils.ForceDictType(new, type_check)
839 ret = copy.deepcopy(base)
840 ret.update(dict((key, fn(base.get(key, {}), value))
841 for key, value in updates.items()))
845 def _MergeAndVerifyHvState(op_input, obj_input):
846 """Combines the hv state from an opcode with the one of the object
848 @param op_input: The input dict from the opcode
849 @param obj_input: The input dict from the objects
850 @return: The verified and updated dict
854 invalid_hvs = set(op_input) - constants.HYPER_TYPES
856 raise errors.OpPrereqError("Invalid hypervisor(s) in hypervisor state:"
857 " %s" % utils.CommaJoin(invalid_hvs),
859 if obj_input is None:
861 type_check = constants.HVSTS_PARAMETER_TYPES
862 return _UpdateAndVerifySubDict(obj_input, op_input, type_check)
867 def _MergeAndVerifyDiskState(op_input, obj_input):
868 """Combines the disk state from an opcode with the one of the object
870 @param op_input: The input dict from the opcode
871 @param obj_input: The input dict from the objects
872 @return: The verified and updated dict
875 invalid_dst = set(op_input) - constants.DS_VALID_TYPES
877 raise errors.OpPrereqError("Invalid storage type(s) in disk state: %s" %
878 utils.CommaJoin(invalid_dst),
880 type_check = constants.DSS_PARAMETER_TYPES
881 if obj_input is None:
883 return dict((key, _UpdateAndVerifySubDict(obj_input.get(key, {}), value,
885 for key, value in op_input.items())
890 def _ReleaseLocks(lu, level, names=None, keep=None):
891 """Releases locks owned by an LU.
893 @type lu: L{LogicalUnit}
894 @param level: Lock level
895 @type names: list or None
896 @param names: Names of locks to release
897 @type keep: list or None
898 @param keep: Names of locks to retain
901 assert not (keep is not None and names is not None), \
902 "Only one of the 'names' and the 'keep' parameters can be given"
904 if names is not None:
905 should_release = names.__contains__
907 should_release = lambda name: name not in keep
909 should_release = None
911 owned = lu.owned_locks(level)
913 # Not owning any lock at this level, do nothing
920 # Determine which locks to release
922 if should_release(name):
927 assert len(lu.owned_locks(level)) == (len(retain) + len(release))
929 # Release just some locks
930 lu.glm.release(level, names=release)
932 assert frozenset(lu.owned_locks(level)) == frozenset(retain)
935 lu.glm.release(level)
937 assert not lu.glm.is_owned(level), "No locks should be owned"
940 def _MapInstanceDisksToNodes(instances):
941 """Creates a map from (node, volume) to instance name.
943 @type instances: list of L{objects.Instance}
944 @rtype: dict; tuple of (node name, volume name) as key, instance name as value
947 return dict(((node, vol), inst.name)
948 for inst in instances
949 for (node, vols) in inst.MapLVsByNode().items()
953 def _RunPostHook(lu, node_name):
954 """Runs the post-hook for an opcode on a single node.
957 hm = lu.proc.BuildHooksManager(lu)
959 hm.RunPhase(constants.HOOKS_PHASE_POST, nodes=[node_name])
961 # pylint: disable=W0702
962 lu.LogWarning("Errors occurred running hooks on %s" % node_name)
965 def _CheckOutputFields(static, dynamic, selected):
966 """Checks whether all selected fields are valid.
968 @type static: L{utils.FieldSet}
969 @param static: static fields set
970 @type dynamic: L{utils.FieldSet}
971 @param dynamic: dynamic fields set
978 delta = f.NonMatching(selected)
980 raise errors.OpPrereqError("Unknown output fields selected: %s"
981 % ",".join(delta), errors.ECODE_INVAL)
984 def _CheckGlobalHvParams(params):
985 """Validates that given hypervisor params are not global ones.
987 This will ensure that instances don't get customised versions of
991 used_globals = constants.HVC_GLOBALS.intersection(params)
993 msg = ("The following hypervisor parameters are global and cannot"
994 " be customized at instance level, please modify them at"
995 " cluster level: %s" % utils.CommaJoin(used_globals))
996 raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
999 def _CheckNodeOnline(lu, node, msg=None):
1000 """Ensure that a given node is online.
1002 @param lu: the LU on behalf of which we make the check
1003 @param node: the node to check
1004 @param msg: if passed, should be a message to replace the default one
1005 @raise errors.OpPrereqError: if the node is offline
1009 msg = "Can't use offline node"
1010 if lu.cfg.GetNodeInfo(node).offline:
1011 raise errors.OpPrereqError("%s: %s" % (msg, node), errors.ECODE_STATE)
1014 def _CheckNodeNotDrained(lu, node):
1015 """Ensure that a given node is not drained.
1017 @param lu: the LU on behalf of which we make the check
1018 @param node: the node to check
1019 @raise errors.OpPrereqError: if the node is drained
1022 if lu.cfg.GetNodeInfo(node).drained:
1023 raise errors.OpPrereqError("Can't use drained node %s" % node,
1027 def _CheckNodeVmCapable(lu, node):
1028 """Ensure that a given node is vm capable.
1030 @param lu: the LU on behalf of which we make the check
1031 @param node: the node to check
1032 @raise errors.OpPrereqError: if the node is not vm capable
1035 if not lu.cfg.GetNodeInfo(node).vm_capable:
1036 raise errors.OpPrereqError("Can't use non-vm_capable node %s" % node,
1040 def _CheckNodeHasOS(lu, node, os_name, force_variant):
1041 """Ensure that a node supports a given OS.
1043 @param lu: the LU on behalf of which we make the check
1044 @param node: the node to check
1045 @param os_name: the OS to query about
1046 @param force_variant: whether to ignore variant errors
1047 @raise errors.OpPrereqError: if the node is not supporting the OS
1050 result = lu.rpc.call_os_get(node, os_name)
1051 result.Raise("OS '%s' not in supported OS list for node %s" %
1053 prereq=True, ecode=errors.ECODE_INVAL)
1054 if not force_variant:
1055 _CheckOSVariant(result.payload, os_name)
1058 def _CheckNodeHasSecondaryIP(lu, node, secondary_ip, prereq):
1059 """Ensure that a node has the given secondary ip.
1061 @type lu: L{LogicalUnit}
1062 @param lu: the LU on behalf of which we make the check
1064 @param node: the node to check
1065 @type secondary_ip: string
1066 @param secondary_ip: the ip to check
1067 @type prereq: boolean
1068 @param prereq: whether to throw a prerequisite or an execute error
1069 @raise errors.OpPrereqError: if the node doesn't have the ip, and prereq=True
1070 @raise errors.OpExecError: if the node doesn't have the ip, and prereq=False
1073 result = lu.rpc.call_node_has_ip_address(node, secondary_ip)
1074 result.Raise("Failure checking secondary ip on node %s" % node,
1075 prereq=prereq, ecode=errors.ECODE_ENVIRON)
1076 if not result.payload:
1077 msg = ("Node claims it doesn't have the secondary ip you gave (%s),"
1078 " please fix and re-run this command" % secondary_ip)
1080 raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
1082 raise errors.OpExecError(msg)
1085 def _GetClusterDomainSecret():
1086 """Reads the cluster domain secret.
1089 return utils.ReadOneLineFile(constants.CLUSTER_DOMAIN_SECRET_FILE,
1093 def _CheckInstanceState(lu, instance, req_states, msg=None):
1094 """Ensure that an instance is in one of the required states.
1096 @param lu: the LU on behalf of which we make the check
1097 @param instance: the instance to check
1098 @param msg: if passed, should be a message to replace the default one
1099 @raise errors.OpPrereqError: if the instance is not in the required state
1103 msg = "can't use instance from outside %s states" % ", ".join(req_states)
1104 if instance.admin_state not in req_states:
1105 raise errors.OpPrereqError("Instance '%s' is marked to be %s, %s" %
1106 (instance.name, instance.admin_state, msg),
1109 if constants.ADMINST_UP not in req_states:
1110 pnode = instance.primary_node
1111 ins_l = lu.rpc.call_instance_list([pnode], [instance.hypervisor])[pnode]
1112 ins_l.Raise("Can't contact node %s for instance information" % pnode,
1113 prereq=True, ecode=errors.ECODE_ENVIRON)
1115 if instance.name in ins_l.payload:
1116 raise errors.OpPrereqError("Instance %s is running, %s" %
1117 (instance.name, msg), errors.ECODE_STATE)
1120 def _ComputeMinMaxSpec(name, qualifier, ipolicy, value):
1121 """Computes if value is in the desired range.
1123 @param name: name of the parameter for which we perform the check
1124 @param qualifier: a qualifier used in the error message (e.g. 'disk/1',
1126 @param ipolicy: dictionary containing min, max and std values
1127 @param value: actual value that we want to use
1128 @return: None or element not meeting the criteria
1132 if value in [None, constants.VALUE_AUTO]:
1134 max_v = ipolicy[constants.ISPECS_MAX].get(name, value)
1135 min_v = ipolicy[constants.ISPECS_MIN].get(name, value)
1136 if value > max_v or min_v > value:
1138 fqn = "%s/%s" % (name, qualifier)
1141 return ("%s value %s is not in range [%s, %s]" %
1142 (fqn, value, min_v, max_v))
1146 def _ComputeIPolicySpecViolation(ipolicy, mem_size, cpu_count, disk_count,
1147 nic_count, disk_sizes, spindle_use,
1148 _compute_fn=_ComputeMinMaxSpec):
1149 """Verifies ipolicy against provided specs.
1152 @param ipolicy: The ipolicy
1154 @param mem_size: The memory size
1155 @type cpu_count: int
1156 @param cpu_count: Used cpu cores
1157 @type disk_count: int
1158 @param disk_count: Number of disks used
1159 @type nic_count: int
1160 @param nic_count: Number of nics used
1161 @type disk_sizes: list of ints
1162 @param disk_sizes: Disk sizes of used disk (len must match C{disk_count})
1163 @type spindle_use: int
1164 @param spindle_use: The number of spindles this instance uses
1165 @param _compute_fn: The compute function (unittest only)
1166 @return: A list of violations, or an empty list of no violations are found
1169 assert disk_count == len(disk_sizes)
1172 (constants.ISPEC_MEM_SIZE, "", mem_size),
1173 (constants.ISPEC_CPU_COUNT, "", cpu_count),
1174 (constants.ISPEC_DISK_COUNT, "", disk_count),
1175 (constants.ISPEC_NIC_COUNT, "", nic_count),
1176 (constants.ISPEC_SPINDLE_USE, "", spindle_use),
1177 ] + [(constants.ISPEC_DISK_SIZE, str(idx), d)
1178 for idx, d in enumerate(disk_sizes)]
1181 (_compute_fn(name, qualifier, ipolicy, value)
1182 for (name, qualifier, value) in test_settings))
1185 def _ComputeIPolicyInstanceViolation(ipolicy, instance,
1186 _compute_fn=_ComputeIPolicySpecViolation):
1187 """Compute if instance meets the specs of ipolicy.
1190 @param ipolicy: The ipolicy to verify against
1191 @type instance: L{objects.Instance}
1192 @param instance: The instance to verify
1193 @param _compute_fn: The function to verify ipolicy (unittest only)
1194 @see: L{_ComputeIPolicySpecViolation}
1197 mem_size = instance.beparams.get(constants.BE_MAXMEM, None)
1198 cpu_count = instance.beparams.get(constants.BE_VCPUS, None)
1199 spindle_use = instance.beparams.get(constants.BE_SPINDLE_USE, None)
1200 disk_count = len(instance.disks)
1201 disk_sizes = [disk.size for disk in instance.disks]
1202 nic_count = len(instance.nics)
1204 return _compute_fn(ipolicy, mem_size, cpu_count, disk_count, nic_count,
1205 disk_sizes, spindle_use)
1208 def _ComputeIPolicyInstanceSpecViolation(ipolicy, instance_spec,
1209 _compute_fn=_ComputeIPolicySpecViolation):
1210 """Compute if instance specs meets the specs of ipolicy.
1213 @param ipolicy: The ipolicy to verify against
1214 @param instance_spec: dict
1215 @param instance_spec: The instance spec to verify
1216 @param _compute_fn: The function to verify ipolicy (unittest only)
1217 @see: L{_ComputeIPolicySpecViolation}
1220 mem_size = instance_spec.get(constants.ISPEC_MEM_SIZE, None)
1221 cpu_count = instance_spec.get(constants.ISPEC_CPU_COUNT, None)
1222 disk_count = instance_spec.get(constants.ISPEC_DISK_COUNT, 0)
1223 disk_sizes = instance_spec.get(constants.ISPEC_DISK_SIZE, [])
1224 nic_count = instance_spec.get(constants.ISPEC_NIC_COUNT, 0)
1225 spindle_use = instance_spec.get(constants.ISPEC_SPINDLE_USE, None)
1227 return _compute_fn(ipolicy, mem_size, cpu_count, disk_count, nic_count,
1228 disk_sizes, spindle_use)
1231 def _ComputeIPolicyNodeViolation(ipolicy, instance, current_group,
1233 _compute_fn=_ComputeIPolicyInstanceViolation):
1234 """Compute if instance meets the specs of the new target group.
1236 @param ipolicy: The ipolicy to verify
1237 @param instance: The instance object to verify
1238 @param current_group: The current group of the instance
1239 @param target_group: The new group of the instance
1240 @param _compute_fn: The function to verify ipolicy (unittest only)
1241 @see: L{_ComputeIPolicySpecViolation}
1244 if current_group == target_group:
1247 return _compute_fn(ipolicy, instance)
1250 def _CheckTargetNodeIPolicy(lu, ipolicy, instance, node, ignore=False,
1251 _compute_fn=_ComputeIPolicyNodeViolation):
1252 """Checks that the target node is correct in terms of instance policy.
1254 @param ipolicy: The ipolicy to verify
1255 @param instance: The instance object to verify
1256 @param node: The new node to relocate
1257 @param ignore: Ignore violations of the ipolicy
1258 @param _compute_fn: The function to verify ipolicy (unittest only)
1259 @see: L{_ComputeIPolicySpecViolation}
1262 primary_node = lu.cfg.GetNodeInfo(instance.primary_node)
1263 res = _compute_fn(ipolicy, instance, primary_node.group, node.group)
1266 msg = ("Instance does not meet target node group's (%s) instance"
1267 " policy: %s") % (node.group, utils.CommaJoin(res))
1271 raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
1274 def _ComputeNewInstanceViolations(old_ipolicy, new_ipolicy, instances):
1275 """Computes a set of any instances that would violate the new ipolicy.
1277 @param old_ipolicy: The current (still in-place) ipolicy
1278 @param new_ipolicy: The new (to become) ipolicy
1279 @param instances: List of instances to verify
1280 @return: A list of instances which violates the new ipolicy but
1284 return (_ComputeViolatingInstances(new_ipolicy, instances) -
1285 _ComputeViolatingInstances(old_ipolicy, instances))
1288 def _ExpandItemName(fn, name, kind):
1289 """Expand an item name.
1291 @param fn: the function to use for expansion
1292 @param name: requested item name
1293 @param kind: text description ('Node' or 'Instance')
1294 @return: the resolved (full) name
1295 @raise errors.OpPrereqError: if the item is not found
1298 full_name = fn(name)
1299 if full_name is None:
1300 raise errors.OpPrereqError("%s '%s' not known" % (kind, name),
1305 def _ExpandNodeName(cfg, name):
1306 """Wrapper over L{_ExpandItemName} for nodes."""
1307 return _ExpandItemName(cfg.ExpandNodeName, name, "Node")
1310 def _ExpandInstanceName(cfg, name):
1311 """Wrapper over L{_ExpandItemName} for instance."""
1312 return _ExpandItemName(cfg.ExpandInstanceName, name, "Instance")
1315 def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
1316 minmem, maxmem, vcpus, nics, disk_template, disks,
1317 bep, hvp, hypervisor_name, tags):
1318 """Builds instance related env variables for hooks
1320 This builds the hook environment from individual variables.
1323 @param name: the name of the instance
1324 @type primary_node: string
1325 @param primary_node: the name of the instance's primary node
1326 @type secondary_nodes: list
1327 @param secondary_nodes: list of secondary nodes as strings
1328 @type os_type: string
1329 @param os_type: the name of the instance's OS
1330 @type status: string
1331 @param status: the desired status of the instance
1332 @type minmem: string
1333 @param minmem: the minimum memory size of the instance
1334 @type maxmem: string
1335 @param maxmem: the maximum memory size of the instance
1337 @param vcpus: the count of VCPUs the instance has
1339 @param nics: list of tuples (ip, mac, mode, link) representing
1340 the NICs the instance has
1341 @type disk_template: string
1342 @param disk_template: the disk template of the instance
1344 @param disks: the list of (size, mode) pairs
1346 @param bep: the backend parameters for the instance
1348 @param hvp: the hypervisor parameters for the instance
1349 @type hypervisor_name: string
1350 @param hypervisor_name: the hypervisor for the instance
1352 @param tags: list of instance tags as strings
1354 @return: the hook environment for this instance
1359 "INSTANCE_NAME": name,
1360 "INSTANCE_PRIMARY": primary_node,
1361 "INSTANCE_SECONDARIES": " ".join(secondary_nodes),
1362 "INSTANCE_OS_TYPE": os_type,
1363 "INSTANCE_STATUS": status,
1364 "INSTANCE_MINMEM": minmem,
1365 "INSTANCE_MAXMEM": maxmem,
1366 # TODO(2.7) remove deprecated "memory" value
1367 "INSTANCE_MEMORY": maxmem,
1368 "INSTANCE_VCPUS": vcpus,
1369 "INSTANCE_DISK_TEMPLATE": disk_template,
1370 "INSTANCE_HYPERVISOR": hypervisor_name,
1373 nic_count = len(nics)
1374 for idx, (ip, mac, mode, link) in enumerate(nics):
1377 env["INSTANCE_NIC%d_IP" % idx] = ip
1378 env["INSTANCE_NIC%d_MAC" % idx] = mac
1379 env["INSTANCE_NIC%d_MODE" % idx] = mode
1380 env["INSTANCE_NIC%d_LINK" % idx] = link
1381 if mode == constants.NIC_MODE_BRIDGED:
1382 env["INSTANCE_NIC%d_BRIDGE" % idx] = link
1386 env["INSTANCE_NIC_COUNT"] = nic_count
1389 disk_count = len(disks)
1390 for idx, (size, mode) in enumerate(disks):
1391 env["INSTANCE_DISK%d_SIZE" % idx] = size
1392 env["INSTANCE_DISK%d_MODE" % idx] = mode
1396 env["INSTANCE_DISK_COUNT"] = disk_count
1401 env["INSTANCE_TAGS"] = " ".join(tags)
1403 for source, kind in [(bep, "BE"), (hvp, "HV")]:
1404 for key, value in source.items():
1405 env["INSTANCE_%s_%s" % (kind, key)] = value
1410 def _NICListToTuple(lu, nics):
1411 """Build a list of nic information tuples.
1413 This list is suitable to be passed to _BuildInstanceHookEnv or as a return
1414 value in LUInstanceQueryData.
1416 @type lu: L{LogicalUnit}
1417 @param lu: the logical unit on whose behalf we execute
1418 @type nics: list of L{objects.NIC}
1419 @param nics: list of nics to convert to hooks tuples
1423 cluster = lu.cfg.GetClusterInfo()
1427 filled_params = cluster.SimpleFillNIC(nic.nicparams)
1428 mode = filled_params[constants.NIC_MODE]
1429 link = filled_params[constants.NIC_LINK]
1430 hooks_nics.append((ip, mac, mode, link))
1434 def _BuildInstanceHookEnvByObject(lu, instance, override=None):
1435 """Builds instance related env variables for hooks from an object.
1437 @type lu: L{LogicalUnit}
1438 @param lu: the logical unit on whose behalf we execute
1439 @type instance: L{objects.Instance}
1440 @param instance: the instance for which we should build the
1442 @type override: dict
1443 @param override: dictionary with key/values that will override
1446 @return: the hook environment dictionary
1449 cluster = lu.cfg.GetClusterInfo()
1450 bep = cluster.FillBE(instance)
1451 hvp = cluster.FillHV(instance)
1453 "name": instance.name,
1454 "primary_node": instance.primary_node,
1455 "secondary_nodes": instance.secondary_nodes,
1456 "os_type": instance.os,
1457 "status": instance.admin_state,
1458 "maxmem": bep[constants.BE_MAXMEM],
1459 "minmem": bep[constants.BE_MINMEM],
1460 "vcpus": bep[constants.BE_VCPUS],
1461 "nics": _NICListToTuple(lu, instance.nics),
1462 "disk_template": instance.disk_template,
1463 "disks": [(disk.size, disk.mode) for disk in instance.disks],
1466 "hypervisor_name": instance.hypervisor,
1467 "tags": instance.tags,
1470 args.update(override)
1471 return _BuildInstanceHookEnv(**args) # pylint: disable=W0142
1474 def _AdjustCandidatePool(lu, exceptions):
1475 """Adjust the candidate pool after node operations.
1478 mod_list = lu.cfg.MaintainCandidatePool(exceptions)
1480 lu.LogInfo("Promoted nodes to master candidate role: %s",
1481 utils.CommaJoin(node.name for node in mod_list))
1482 for name in mod_list:
1483 lu.context.ReaddNode(name)
1484 mc_now, mc_max, _ = lu.cfg.GetMasterCandidateStats(exceptions)
1486 lu.LogInfo("Note: more nodes are candidates (%d) than desired (%d)" %
1490 def _DecideSelfPromotion(lu, exceptions=None):
1491 """Decide whether I should promote myself as a master candidate.
1494 cp_size = lu.cfg.GetClusterInfo().candidate_pool_size
1495 mc_now, mc_should, _ = lu.cfg.GetMasterCandidateStats(exceptions)
1496 # the new node will increase mc_max with one, so:
1497 mc_should = min(mc_should + 1, cp_size)
1498 return mc_now < mc_should
1501 def _CalculateGroupIPolicy(cluster, group):
1502 """Calculate instance policy for group.
1505 return cluster.SimpleFillIPolicy(group.ipolicy)
1508 def _ComputeViolatingInstances(ipolicy, instances):
1509 """Computes a set of instances who violates given ipolicy.
1511 @param ipolicy: The ipolicy to verify
1512 @type instances: object.Instance
1513 @param instances: List of instances to verify
1514 @return: A frozenset of instance names violating the ipolicy
1517 return frozenset([inst.name for inst in instances
1518 if _ComputeIPolicyInstanceViolation(ipolicy, inst)])
1521 def _CheckNicsBridgesExist(lu, target_nics, target_node):
1522 """Check that the brigdes needed by a list of nics exist.
1525 cluster = lu.cfg.GetClusterInfo()
1526 paramslist = [cluster.SimpleFillNIC(nic.nicparams) for nic in target_nics]
1527 brlist = [params[constants.NIC_LINK] for params in paramslist
1528 if params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED]
1530 result = lu.rpc.call_bridges_exist(target_node, brlist)
1531 result.Raise("Error checking bridges on destination node '%s'" %
1532 target_node, prereq=True, ecode=errors.ECODE_ENVIRON)
1535 def _CheckInstanceBridgesExist(lu, instance, node=None):
1536 """Check that the brigdes needed by an instance exist.
1540 node = instance.primary_node
1541 _CheckNicsBridgesExist(lu, instance.nics, node)
1544 def _CheckOSVariant(os_obj, name):
1545 """Check whether an OS name conforms to the os variants specification.
1547 @type os_obj: L{objects.OS}
1548 @param os_obj: OS object to check
1550 @param name: OS name passed by the user, to check for validity
1553 variant = objects.OS.GetVariant(name)
1554 if not os_obj.supported_variants:
1556 raise errors.OpPrereqError("OS '%s' doesn't support variants ('%s'"
1557 " passed)" % (os_obj.name, variant),
1561 raise errors.OpPrereqError("OS name must include a variant",
1564 if variant not in os_obj.supported_variants:
1565 raise errors.OpPrereqError("Unsupported OS variant", errors.ECODE_INVAL)
1568 def _GetNodeInstancesInner(cfg, fn):
1569 return [i for i in cfg.GetAllInstancesInfo().values() if fn(i)]
1572 def _GetNodeInstances(cfg, node_name):
1573 """Returns a list of all primary and secondary instances on a node.
1577 return _GetNodeInstancesInner(cfg, lambda inst: node_name in inst.all_nodes)
1580 def _GetNodePrimaryInstances(cfg, node_name):
1581 """Returns primary instances on a node.
1584 return _GetNodeInstancesInner(cfg,
1585 lambda inst: node_name == inst.primary_node)
1588 def _GetNodeSecondaryInstances(cfg, node_name):
1589 """Returns secondary instances on a node.
1592 return _GetNodeInstancesInner(cfg,
1593 lambda inst: node_name in inst.secondary_nodes)
1596 def _GetStorageTypeArgs(cfg, storage_type):
1597 """Returns the arguments for a storage type.
1600 # Special case for file storage
1601 if storage_type == constants.ST_FILE:
1602 # storage.FileStorage wants a list of storage directories
1603 return [[cfg.GetFileStorageDir(), cfg.GetSharedFileStorageDir()]]
1608 def _FindFaultyInstanceDisks(cfg, rpc_runner, instance, node_name, prereq):
1611 for dev in instance.disks:
1612 cfg.SetDiskID(dev, node_name)
1614 result = rpc_runner.call_blockdev_getmirrorstatus(node_name, (instance.disks,
1616 result.Raise("Failed to get disk status from node %s" % node_name,
1617 prereq=prereq, ecode=errors.ECODE_ENVIRON)
1619 for idx, bdev_status in enumerate(result.payload):
1620 if bdev_status and bdev_status.ldisk_status == constants.LDS_FAULTY:
1626 def _CheckIAllocatorOrNode(lu, iallocator_slot, node_slot):
1627 """Check the sanity of iallocator and node arguments and use the
1628 cluster-wide iallocator if appropriate.
1630 Check that at most one of (iallocator, node) is specified. If none is
1631 specified, then the LU's opcode's iallocator slot is filled with the
1632 cluster-wide default iallocator.
1634 @type iallocator_slot: string
1635 @param iallocator_slot: the name of the opcode iallocator slot
1636 @type node_slot: string
1637 @param node_slot: the name of the opcode target node slot
1640 node = getattr(lu.op, node_slot, None)
1641 iallocator = getattr(lu.op, iallocator_slot, None)
1643 if node is not None and iallocator is not None:
1644 raise errors.OpPrereqError("Do not specify both, iallocator and node",
1646 elif node is None and iallocator is None:
1647 default_iallocator = lu.cfg.GetDefaultIAllocator()
1648 if default_iallocator:
1649 setattr(lu.op, iallocator_slot, default_iallocator)
1651 raise errors.OpPrereqError("No iallocator or node given and no"
1652 " cluster-wide default iallocator found;"
1653 " please specify either an iallocator or a"
1654 " node, or set a cluster-wide default"
1658 def _GetDefaultIAllocator(cfg, iallocator):
1659 """Decides on which iallocator to use.
1661 @type cfg: L{config.ConfigWriter}
1662 @param cfg: Cluster configuration object
1663 @type iallocator: string or None
1664 @param iallocator: Iallocator specified in opcode
1666 @return: Iallocator name
1670 # Use default iallocator
1671 iallocator = cfg.GetDefaultIAllocator()
1674 raise errors.OpPrereqError("No iallocator was specified, neither in the"
1675 " opcode nor as a cluster-wide default",
1681 class LUClusterPostInit(LogicalUnit):
1682 """Logical unit for running hooks after cluster initialization.
1685 HPATH = "cluster-init"
1686 HTYPE = constants.HTYPE_CLUSTER
1688 def BuildHooksEnv(self):
1693 "OP_TARGET": self.cfg.GetClusterName(),
1696 def BuildHooksNodes(self):
1697 """Build hooks nodes.
1700 return ([], [self.cfg.GetMasterNode()])
1702 def Exec(self, feedback_fn):
1709 class LUClusterDestroy(LogicalUnit):
1710 """Logical unit for destroying the cluster.
1713 HPATH = "cluster-destroy"
1714 HTYPE = constants.HTYPE_CLUSTER
1716 def BuildHooksEnv(self):
1721 "OP_TARGET": self.cfg.GetClusterName(),
1724 def BuildHooksNodes(self):
1725 """Build hooks nodes.
1730 def CheckPrereq(self):
1731 """Check prerequisites.
1733 This checks whether the cluster is empty.
1735 Any errors are signaled by raising errors.OpPrereqError.
1738 master = self.cfg.GetMasterNode()
1740 nodelist = self.cfg.GetNodeList()
1741 if len(nodelist) != 1 or nodelist[0] != master:
1742 raise errors.OpPrereqError("There are still %d node(s) in"
1743 " this cluster." % (len(nodelist) - 1),
1745 instancelist = self.cfg.GetInstanceList()
1747 raise errors.OpPrereqError("There are still %d instance(s) in"
1748 " this cluster." % len(instancelist),
1751 def Exec(self, feedback_fn):
1752 """Destroys the cluster.
1755 master_params = self.cfg.GetMasterNetworkParameters()
1757 # Run post hooks on master node before it's removed
1758 _RunPostHook(self, master_params.name)
1760 ems = self.cfg.GetUseExternalMipScript()
1761 result = self.rpc.call_node_deactivate_master_ip(master_params.name,
1764 self.LogWarning("Error disabling the master IP address: %s",
1767 return master_params.name
1770 def _VerifyCertificate(filename):
1771 """Verifies a certificate for L{LUClusterVerifyConfig}.
1773 @type filename: string
1774 @param filename: Path to PEM file
1778 cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
1779 utils.ReadFile(filename))
1780 except Exception, err: # pylint: disable=W0703
1781 return (LUClusterVerifyConfig.ETYPE_ERROR,
1782 "Failed to load X509 certificate %s: %s" % (filename, err))
1785 utils.VerifyX509Certificate(cert, constants.SSL_CERT_EXPIRATION_WARN,
1786 constants.SSL_CERT_EXPIRATION_ERROR)
1789 fnamemsg = "While verifying %s: %s" % (filename, msg)
1794 return (None, fnamemsg)
1795 elif errcode == utils.CERT_WARNING:
1796 return (LUClusterVerifyConfig.ETYPE_WARNING, fnamemsg)
1797 elif errcode == utils.CERT_ERROR:
1798 return (LUClusterVerifyConfig.ETYPE_ERROR, fnamemsg)
1800 raise errors.ProgrammerError("Unhandled certificate error code %r" % errcode)
1803 def _GetAllHypervisorParameters(cluster, instances):
1804 """Compute the set of all hypervisor parameters.
1806 @type cluster: L{objects.Cluster}
1807 @param cluster: the cluster object
1808 @param instances: list of L{objects.Instance}
1809 @param instances: additional instances from which to obtain parameters
1810 @rtype: list of (origin, hypervisor, parameters)
1811 @return: a list with all parameters found, indicating the hypervisor they
1812 apply to, and the origin (can be "cluster", "os X", or "instance Y")
1817 for hv_name in cluster.enabled_hypervisors:
1818 hvp_data.append(("cluster", hv_name, cluster.GetHVDefaults(hv_name)))
1820 for os_name, os_hvp in cluster.os_hvp.items():
1821 for hv_name, hv_params in os_hvp.items():
1823 full_params = cluster.GetHVDefaults(hv_name, os_name=os_name)
1824 hvp_data.append(("os %s" % os_name, hv_name, full_params))
1826 # TODO: collapse identical parameter values in a single one
1827 for instance in instances:
1828 if instance.hvparams:
1829 hvp_data.append(("instance %s" % instance.name, instance.hypervisor,
1830 cluster.FillHV(instance)))
1835 class _VerifyErrors(object):
1836 """Mix-in for cluster/group verify LUs.
1838 It provides _Error and _ErrorIf, and updates the self.bad boolean. (Expects
1839 self.op and self._feedback_fn to be available.)
1843 ETYPE_FIELD = "code"
1844 ETYPE_ERROR = "ERROR"
1845 ETYPE_WARNING = "WARNING"
1847 def _Error(self, ecode, item, msg, *args, **kwargs):
1848 """Format an error message.
1850 Based on the opcode's error_codes parameter, either format a
1851 parseable error code, or a simpler error string.
1853 This must be called only from Exec and functions called from Exec.
1856 ltype = kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR)
1857 itype, etxt, _ = ecode
1858 # first complete the msg
1861 # then format the whole message
1862 if self.op.error_codes: # This is a mix-in. pylint: disable=E1101
1863 msg = "%s:%s:%s:%s:%s" % (ltype, etxt, itype, item, msg)
1869 msg = "%s: %s%s: %s" % (ltype, itype, item, msg)
1870 # and finally report it via the feedback_fn
1871 self._feedback_fn(" - %s" % msg) # Mix-in. pylint: disable=E1101
1873 def _ErrorIf(self, cond, ecode, *args, **kwargs):
1874 """Log an error message if the passed condition is True.
1878 or self.op.debug_simulate_errors) # pylint: disable=E1101
1880 # If the error code is in the list of ignored errors, demote the error to a
1882 (_, etxt, _) = ecode
1883 if etxt in self.op.ignore_errors: # pylint: disable=E1101
1884 kwargs[self.ETYPE_FIELD] = self.ETYPE_WARNING
1887 self._Error(ecode, *args, **kwargs)
1889 # do not mark the operation as failed for WARN cases only
1890 if kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR) == self.ETYPE_ERROR:
1891 self.bad = self.bad or cond
1894 class LUClusterVerify(NoHooksLU):
1895 """Submits all jobs necessary to verify the cluster.
1900 def ExpandNames(self):
1901 self.needed_locks = {}
1903 def Exec(self, feedback_fn):
1906 if self.op.group_name:
1907 groups = [self.op.group_name]
1908 depends_fn = lambda: None
1910 groups = self.cfg.GetNodeGroupList()
1912 # Verify global configuration
1914 opcodes.OpClusterVerifyConfig(ignore_errors=self.op.ignore_errors)
1917 # Always depend on global verification
1918 depends_fn = lambda: [(-len(jobs), [])]
1920 jobs.extend([opcodes.OpClusterVerifyGroup(group_name=group,
1921 ignore_errors=self.op.ignore_errors,
1922 depends=depends_fn())]
1923 for group in groups)
1925 # Fix up all parameters
1926 for op in itertools.chain(*jobs): # pylint: disable=W0142
1927 op.debug_simulate_errors = self.op.debug_simulate_errors
1928 op.verbose = self.op.verbose
1929 op.error_codes = self.op.error_codes
1931 op.skip_checks = self.op.skip_checks
1932 except AttributeError:
1933 assert not isinstance(op, opcodes.OpClusterVerifyGroup)
1935 return ResultWithJobs(jobs)
1938 class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
1939 """Verifies the cluster config.
1944 def _VerifyHVP(self, hvp_data):
1945 """Verifies locally the syntax of the hypervisor parameters.
1948 for item, hv_name, hv_params in hvp_data:
1949 msg = ("hypervisor %s parameters syntax check (source %s): %%s" %
1952 hv_class = hypervisor.GetHypervisor(hv_name)
1953 utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
1954 hv_class.CheckParameterSyntax(hv_params)
1955 except errors.GenericError, err:
1956 self._ErrorIf(True, constants.CV_ECLUSTERCFG, None, msg % str(err))
1958 def ExpandNames(self):
1959 self.needed_locks = dict.fromkeys(locking.LEVELS, locking.ALL_SET)
1960 self.share_locks = _ShareAll()
1962 def CheckPrereq(self):
1963 """Check prerequisites.
1966 # Retrieve all information
1967 self.all_group_info = self.cfg.GetAllNodeGroupsInfo()
1968 self.all_node_info = self.cfg.GetAllNodesInfo()
1969 self.all_inst_info = self.cfg.GetAllInstancesInfo()
1971 def Exec(self, feedback_fn):
1972 """Verify integrity of cluster, performing various test on nodes.
1976 self._feedback_fn = feedback_fn
1978 feedback_fn("* Verifying cluster config")
1980 for msg in self.cfg.VerifyConfig():
1981 self._ErrorIf(True, constants.CV_ECLUSTERCFG, None, msg)
1983 feedback_fn("* Verifying cluster certificate files")
1985 for cert_filename in constants.ALL_CERT_FILES:
1986 (errcode, msg) = _VerifyCertificate(cert_filename)
1987 self._ErrorIf(errcode, constants.CV_ECLUSTERCERT, None, msg, code=errcode)
1989 feedback_fn("* Verifying hypervisor parameters")
1991 self._VerifyHVP(_GetAllHypervisorParameters(self.cfg.GetClusterInfo(),
1992 self.all_inst_info.values()))
1994 feedback_fn("* Verifying all nodes belong to an existing group")
1996 # We do this verification here because, should this bogus circumstance
1997 # occur, it would never be caught by VerifyGroup, which only acts on
1998 # nodes/instances reachable from existing node groups.
2000 dangling_nodes = set(node.name for node in self.all_node_info.values()
2001 if node.group not in self.all_group_info)
2003 dangling_instances = {}
2004 no_node_instances = []
2006 for inst in self.all_inst_info.values():
2007 if inst.primary_node in dangling_nodes:
2008 dangling_instances.setdefault(inst.primary_node, []).append(inst.name)
2009 elif inst.primary_node not in self.all_node_info:
2010 no_node_instances.append(inst.name)
2015 utils.CommaJoin(dangling_instances.get(node.name,
2017 for node in dangling_nodes]
2019 self._ErrorIf(bool(dangling_nodes), constants.CV_ECLUSTERDANGLINGNODES,
2021 "the following nodes (and their instances) belong to a non"
2022 " existing group: %s", utils.CommaJoin(pretty_dangling))
2024 self._ErrorIf(bool(no_node_instances), constants.CV_ECLUSTERDANGLINGINST,
2026 "the following instances have a non-existing primary-node:"
2027 " %s", utils.CommaJoin(no_node_instances))
2032 class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
2033 """Verifies the status of a node group.
2036 HPATH = "cluster-verify"
2037 HTYPE = constants.HTYPE_CLUSTER
2040 _HOOKS_INDENT_RE = re.compile("^", re.M)
2042 class NodeImage(object):
2043 """A class representing the logical and physical status of a node.
2046 @ivar name: the node name to which this object refers
2047 @ivar volumes: a structure as returned from
2048 L{ganeti.backend.GetVolumeList} (runtime)
2049 @ivar instances: a list of running instances (runtime)
2050 @ivar pinst: list of configured primary instances (config)
2051 @ivar sinst: list of configured secondary instances (config)
2052 @ivar sbp: dictionary of {primary-node: list of instances} for all
2053 instances for which this node is secondary (config)
2054 @ivar mfree: free memory, as reported by hypervisor (runtime)
2055 @ivar dfree: free disk, as reported by the node (runtime)
2056 @ivar offline: the offline status (config)
2057 @type rpc_fail: boolean
2058 @ivar rpc_fail: whether the RPC verify call was successfull (overall,
2059 not whether the individual keys were correct) (runtime)
2060 @type lvm_fail: boolean
2061 @ivar lvm_fail: whether the RPC call didn't return valid LVM data
2062 @type hyp_fail: boolean
2063 @ivar hyp_fail: whether the RPC call didn't return the instance list
2064 @type ghost: boolean
2065 @ivar ghost: whether this is a known node or not (config)
2066 @type os_fail: boolean
2067 @ivar os_fail: whether the RPC call didn't return valid OS data
2069 @ivar oslist: list of OSes as diagnosed by DiagnoseOS
2070 @type vm_capable: boolean
2071 @ivar vm_capable: whether the node can host instances
2074 def __init__(self, offline=False, name=None, vm_capable=True):
2083 self.offline = offline
2084 self.vm_capable = vm_capable
2085 self.rpc_fail = False
2086 self.lvm_fail = False
2087 self.hyp_fail = False
2089 self.os_fail = False
2092 def ExpandNames(self):
2093 # This raises errors.OpPrereqError on its own:
2094 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
2096 # Get instances in node group; this is unsafe and needs verification later
2098 self.cfg.GetNodeGroupInstances(self.group_uuid, primary_only=True)
2100 self.needed_locks = {
2101 locking.LEVEL_INSTANCE: inst_names,
2102 locking.LEVEL_NODEGROUP: [self.group_uuid],
2103 locking.LEVEL_NODE: [],
2106 self.share_locks = _ShareAll()
2108 def DeclareLocks(self, level):
2109 if level == locking.LEVEL_NODE:
2110 # Get members of node group; this is unsafe and needs verification later
2111 nodes = set(self.cfg.GetNodeGroup(self.group_uuid).members)
2113 all_inst_info = self.cfg.GetAllInstancesInfo()
2115 # In Exec(), we warn about mirrored instances that have primary and
2116 # secondary living in separate node groups. To fully verify that
2117 # volumes for these instances are healthy, we will need to do an
2118 # extra call to their secondaries. We ensure here those nodes will
2120 for inst in self.owned_locks(locking.LEVEL_INSTANCE):
2121 # Important: access only the instances whose lock is owned
2122 if all_inst_info[inst].disk_template in constants.DTS_INT_MIRROR:
2123 nodes.update(all_inst_info[inst].secondary_nodes)
2125 self.needed_locks[locking.LEVEL_NODE] = nodes
2127 def CheckPrereq(self):
2128 assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
2129 self.group_info = self.cfg.GetNodeGroup(self.group_uuid)
2131 group_nodes = set(self.group_info.members)
2133 self.cfg.GetNodeGroupInstances(self.group_uuid, primary_only=True)
2136 group_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
2138 unlocked_instances = \
2139 group_instances.difference(self.owned_locks(locking.LEVEL_INSTANCE))
2142 raise errors.OpPrereqError("Missing lock for nodes: %s" %
2143 utils.CommaJoin(unlocked_nodes),
2146 if unlocked_instances:
2147 raise errors.OpPrereqError("Missing lock for instances: %s" %
2148 utils.CommaJoin(unlocked_instances),
2151 self.all_node_info = self.cfg.GetAllNodesInfo()
2152 self.all_inst_info = self.cfg.GetAllInstancesInfo()
2154 self.my_node_names = utils.NiceSort(group_nodes)
2155 self.my_inst_names = utils.NiceSort(group_instances)
2157 self.my_node_info = dict((name, self.all_node_info[name])
2158 for name in self.my_node_names)
2160 self.my_inst_info = dict((name, self.all_inst_info[name])
2161 for name in self.my_inst_names)
2163 # We detect here the nodes that will need the extra RPC calls for verifying
2164 # split LV volumes; they should be locked.
2165 extra_lv_nodes = set()
2167 for inst in self.my_inst_info.values():
2168 if inst.disk_template in constants.DTS_INT_MIRROR:
2169 for nname in inst.all_nodes:
2170 if self.all_node_info[nname].group != self.group_uuid:
2171 extra_lv_nodes.add(nname)
2173 unlocked_lv_nodes = \
2174 extra_lv_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
2176 if unlocked_lv_nodes:
2177 raise errors.OpPrereqError("Missing node locks for LV check: %s" %
2178 utils.CommaJoin(unlocked_lv_nodes),
2180 self.extra_lv_nodes = list(extra_lv_nodes)
2182 def _VerifyNode(self, ninfo, nresult):
2183 """Perform some basic validation on data returned from a node.
2185 - check the result data structure is well formed and has all the
2187 - check ganeti version
2189 @type ninfo: L{objects.Node}
2190 @param ninfo: the node to check
2191 @param nresult: the results from the node
2193 @return: whether overall this call was successful (and we can expect
2194 reasonable values in the respose)
2198 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2200 # main result, nresult should be a non-empty dict
2201 test = not nresult or not isinstance(nresult, dict)
2202 _ErrorIf(test, constants.CV_ENODERPC, node,
2203 "unable to verify node: no data returned")
2207 # compares ganeti version
2208 local_version = constants.PROTOCOL_VERSION
2209 remote_version = nresult.get("version", None)
2210 test = not (remote_version and
2211 isinstance(remote_version, (list, tuple)) and
2212 len(remote_version) == 2)
2213 _ErrorIf(test, constants.CV_ENODERPC, node,
2214 "connection to node returned invalid data")
2218 test = local_version != remote_version[0]
2219 _ErrorIf(test, constants.CV_ENODEVERSION, node,
2220 "incompatible protocol versions: master %s,"
2221 " node %s", local_version, remote_version[0])
2225 # node seems compatible, we can actually try to look into its results
2227 # full package version
2228 self._ErrorIf(constants.RELEASE_VERSION != remote_version[1],
2229 constants.CV_ENODEVERSION, node,
2230 "software version mismatch: master %s, node %s",
2231 constants.RELEASE_VERSION, remote_version[1],
2232 code=self.ETYPE_WARNING)
2234 hyp_result = nresult.get(constants.NV_HYPERVISOR, None)
2235 if ninfo.vm_capable and isinstance(hyp_result, dict):
2236 for hv_name, hv_result in hyp_result.iteritems():
2237 test = hv_result is not None
2238 _ErrorIf(test, constants.CV_ENODEHV, node,
2239 "hypervisor %s verify failure: '%s'", hv_name, hv_result)
2241 hvp_result = nresult.get(constants.NV_HVPARAMS, None)
2242 if ninfo.vm_capable and isinstance(hvp_result, list):
2243 for item, hv_name, hv_result in hvp_result:
2244 _ErrorIf(True, constants.CV_ENODEHV, node,
2245 "hypervisor %s parameter verify failure (source %s): %s",
2246 hv_name, item, hv_result)
2248 test = nresult.get(constants.NV_NODESETUP,
2249 ["Missing NODESETUP results"])
2250 _ErrorIf(test, constants.CV_ENODESETUP, node, "node setup error: %s",
2255 def _VerifyNodeTime(self, ninfo, nresult,
2256 nvinfo_starttime, nvinfo_endtime):
2257 """Check the node time.
2259 @type ninfo: L{objects.Node}
2260 @param ninfo: the node to check
2261 @param nresult: the remote results for the node
2262 @param nvinfo_starttime: the start time of the RPC call
2263 @param nvinfo_endtime: the end time of the RPC call
2267 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2269 ntime = nresult.get(constants.NV_TIME, None)
2271 ntime_merged = utils.MergeTime(ntime)
2272 except (ValueError, TypeError):
2273 _ErrorIf(True, constants.CV_ENODETIME, node, "Node returned invalid time")
2276 if ntime_merged < (nvinfo_starttime - constants.NODE_MAX_CLOCK_SKEW):
2277 ntime_diff = "%.01fs" % abs(nvinfo_starttime - ntime_merged)
2278 elif ntime_merged > (nvinfo_endtime + constants.NODE_MAX_CLOCK_SKEW):
2279 ntime_diff = "%.01fs" % abs(ntime_merged - nvinfo_endtime)
2283 _ErrorIf(ntime_diff is not None, constants.CV_ENODETIME, node,
2284 "Node time diverges by at least %s from master node time",
2287 def _VerifyNodeLVM(self, ninfo, nresult, vg_name):
2288 """Check the node LVM results.
2290 @type ninfo: L{objects.Node}
2291 @param ninfo: the node to check
2292 @param nresult: the remote results for the node
2293 @param vg_name: the configured VG name
2300 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2302 # checks vg existence and size > 20G
2303 vglist = nresult.get(constants.NV_VGLIST, None)
2305 _ErrorIf(test, constants.CV_ENODELVM, node, "unable to check volume groups")
2307 vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
2308 constants.MIN_VG_SIZE)
2309 _ErrorIf(vgstatus, constants.CV_ENODELVM, node, vgstatus)
2312 pvlist = nresult.get(constants.NV_PVLIST, None)
2313 test = pvlist is None
2314 _ErrorIf(test, constants.CV_ENODELVM, node, "Can't get PV list from node")
2316 # check that ':' is not present in PV names, since it's a
2317 # special character for lvcreate (denotes the range of PEs to
2319 for _, pvname, owner_vg in pvlist:
2320 test = ":" in pvname
2321 _ErrorIf(test, constants.CV_ENODELVM, node,
2322 "Invalid character ':' in PV '%s' of VG '%s'",
2325 def _VerifyNodeBridges(self, ninfo, nresult, bridges):
2326 """Check the node bridges.
2328 @type ninfo: L{objects.Node}
2329 @param ninfo: the node to check
2330 @param nresult: the remote results for the node
2331 @param bridges: the expected list of bridges
2338 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2340 missing = nresult.get(constants.NV_BRIDGES, None)
2341 test = not isinstance(missing, list)
2342 _ErrorIf(test, constants.CV_ENODENET, node,
2343 "did not return valid bridge information")
2345 _ErrorIf(bool(missing), constants.CV_ENODENET, node,
2346 "missing bridges: %s" % utils.CommaJoin(sorted(missing)))
2348 def _VerifyNodeUserScripts(self, ninfo, nresult):
2349 """Check the results of user scripts presence and executability on the node
2351 @type ninfo: L{objects.Node}
2352 @param ninfo: the node to check
2353 @param nresult: the remote results for the node
2358 test = not constants.NV_USERSCRIPTS in nresult
2359 self._ErrorIf(test, constants.CV_ENODEUSERSCRIPTS, node,
2360 "did not return user scripts information")
2362 broken_scripts = nresult.get(constants.NV_USERSCRIPTS, None)
2364 self._ErrorIf(broken_scripts, constants.CV_ENODEUSERSCRIPTS, node,
2365 "user scripts not present or not executable: %s" %
2366 utils.CommaJoin(sorted(broken_scripts)))
2368 def _VerifyNodeNetwork(self, ninfo, nresult):
2369 """Check the node network connectivity results.
2371 @type ninfo: L{objects.Node}
2372 @param ninfo: the node to check
2373 @param nresult: the remote results for the node
2377 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2379 test = constants.NV_NODELIST not in nresult
2380 _ErrorIf(test, constants.CV_ENODESSH, node,
2381 "node hasn't returned node ssh connectivity data")
2383 if nresult[constants.NV_NODELIST]:
2384 for a_node, a_msg in nresult[constants.NV_NODELIST].items():
2385 _ErrorIf(True, constants.CV_ENODESSH, node,
2386 "ssh communication with node '%s': %s", a_node, a_msg)
2388 test = constants.NV_NODENETTEST not in nresult
2389 _ErrorIf(test, constants.CV_ENODENET, node,
2390 "node hasn't returned node tcp connectivity data")
2392 if nresult[constants.NV_NODENETTEST]:
2393 nlist = utils.NiceSort(nresult[constants.NV_NODENETTEST].keys())
2395 _ErrorIf(True, constants.CV_ENODENET, node,
2396 "tcp communication with node '%s': %s",
2397 anode, nresult[constants.NV_NODENETTEST][anode])
2399 test = constants.NV_MASTERIP not in nresult
2400 _ErrorIf(test, constants.CV_ENODENET, node,
2401 "node hasn't returned node master IP reachability data")
2403 if not nresult[constants.NV_MASTERIP]:
2404 if node == self.master_node:
2405 msg = "the master node cannot reach the master IP (not configured?)"
2407 msg = "cannot reach the master IP"
2408 _ErrorIf(True, constants.CV_ENODENET, node, msg)
2410 def _VerifyInstance(self, instance, instanceconfig, node_image,
2412 """Verify an instance.
2414 This function checks to see if the required block devices are
2415 available on the instance's node.
2418 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2419 node_current = instanceconfig.primary_node
2421 node_vol_should = {}
2422 instanceconfig.MapLVsByNode(node_vol_should)
2424 ipolicy = _CalculateGroupIPolicy(self.cfg.GetClusterInfo(), self.group_info)
2425 err = _ComputeIPolicyInstanceViolation(ipolicy, instanceconfig)
2426 _ErrorIf(err, constants.CV_EINSTANCEPOLICY, instance, utils.CommaJoin(err))
2428 for node in node_vol_should:
2429 n_img = node_image[node]
2430 if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
2431 # ignore missing volumes on offline or broken nodes
2433 for volume in node_vol_should[node]:
2434 test = volume not in n_img.volumes
2435 _ErrorIf(test, constants.CV_EINSTANCEMISSINGDISK, instance,
2436 "volume %s missing on node %s", volume, node)
2438 if instanceconfig.admin_state == constants.ADMINST_UP:
2439 pri_img = node_image[node_current]
2440 test = instance not in pri_img.instances and not pri_img.offline
2441 _ErrorIf(test, constants.CV_EINSTANCEDOWN, instance,
2442 "instance not running on its primary node %s",
2445 diskdata = [(nname, success, status, idx)
2446 for (nname, disks) in diskstatus.items()
2447 for idx, (success, status) in enumerate(disks)]
2449 for nname, success, bdev_status, idx in diskdata:
2450 # the 'ghost node' construction in Exec() ensures that we have a
2452 snode = node_image[nname]
2453 bad_snode = snode.ghost or snode.offline
2454 _ErrorIf(instanceconfig.admin_state == constants.ADMINST_UP and
2455 not success and not bad_snode,
2456 constants.CV_EINSTANCEFAULTYDISK, instance,
2457 "couldn't retrieve status for disk/%s on %s: %s",
2458 idx, nname, bdev_status)
2459 _ErrorIf((instanceconfig.admin_state == constants.ADMINST_UP and
2460 success and bdev_status.ldisk_status == constants.LDS_FAULTY),
2461 constants.CV_EINSTANCEFAULTYDISK, instance,
2462 "disk/%s on %s is faulty", idx, nname)
2464 def _VerifyOrphanVolumes(self, node_vol_should, node_image, reserved):
2465 """Verify if there are any unknown volumes in the cluster.
2467 The .os, .swap and backup volumes are ignored. All other volumes are
2468 reported as unknown.
2470 @type reserved: L{ganeti.utils.FieldSet}
2471 @param reserved: a FieldSet of reserved volume names
2474 for node, n_img in node_image.items():
2475 if (n_img.offline or n_img.rpc_fail or n_img.lvm_fail or
2476 self.all_node_info[node].group != self.group_uuid):
2477 # skip non-healthy nodes
2479 for volume in n_img.volumes:
2480 test = ((node not in node_vol_should or
2481 volume not in node_vol_should[node]) and
2482 not reserved.Matches(volume))
2483 self._ErrorIf(test, constants.CV_ENODEORPHANLV, node,
2484 "volume %s is unknown", volume)
2486 def _VerifyNPlusOneMemory(self, node_image, instance_cfg):
2487 """Verify N+1 Memory Resilience.
2489 Check that if one single node dies we can still start all the
2490 instances it was primary for.
2493 cluster_info = self.cfg.GetClusterInfo()
2494 for node, n_img in node_image.items():
2495 # This code checks that every node which is now listed as
2496 # secondary has enough memory to host all instances it is
2497 # supposed to should a single other node in the cluster fail.
2498 # FIXME: not ready for failover to an arbitrary node
2499 # FIXME: does not support file-backed instances
2500 # WARNING: we currently take into account down instances as well
2501 # as up ones, considering that even if they're down someone
2502 # might want to start them even in the event of a node failure.
2503 if n_img.offline or self.all_node_info[node].group != self.group_uuid:
2504 # we're skipping nodes marked offline and nodes in other groups from
2505 # the N+1 warning, since most likely we don't have good memory
2506 # infromation from them; we already list instances living on such
2507 # nodes, and that's enough warning
2509 #TODO(dynmem): also consider ballooning out other instances
2510 for prinode, instances in n_img.sbp.items():
2512 for instance in instances:
2513 bep = cluster_info.FillBE(instance_cfg[instance])
2514 if bep[constants.BE_AUTO_BALANCE]:
2515 needed_mem += bep[constants.BE_MINMEM]
2516 test = n_img.mfree < needed_mem
2517 self._ErrorIf(test, constants.CV_ENODEN1, node,
2518 "not enough memory to accomodate instance failovers"
2519 " should node %s fail (%dMiB needed, %dMiB available)",
2520 prinode, needed_mem, n_img.mfree)
2523 def _VerifyFiles(cls, errorif, nodeinfo, master_node, all_nvinfo,
2524 (files_all, files_opt, files_mc, files_vm)):
2525 """Verifies file checksums collected from all nodes.
2527 @param errorif: Callback for reporting errors
2528 @param nodeinfo: List of L{objects.Node} objects
2529 @param master_node: Name of master node
2530 @param all_nvinfo: RPC results
2533 # Define functions determining which nodes to consider for a file
2536 (files_mc, lambda node: (node.master_candidate or
2537 node.name == master_node)),
2538 (files_vm, lambda node: node.vm_capable),
2541 # Build mapping from filename to list of nodes which should have the file
2543 for (files, fn) in files2nodefn:
2545 filenodes = nodeinfo
2547 filenodes = filter(fn, nodeinfo)
2548 nodefiles.update((filename,
2549 frozenset(map(operator.attrgetter("name"), filenodes)))
2550 for filename in files)
2552 assert set(nodefiles) == (files_all | files_mc | files_vm)
2554 fileinfo = dict((filename, {}) for filename in nodefiles)
2555 ignore_nodes = set()
2557 for node in nodeinfo:
2559 ignore_nodes.add(node.name)
2562 nresult = all_nvinfo[node.name]
2564 if nresult.fail_msg or not nresult.payload:
2567 node_files = nresult.payload.get(constants.NV_FILELIST, None)
2569 test = not (node_files and isinstance(node_files, dict))
2570 errorif(test, constants.CV_ENODEFILECHECK, node.name,
2571 "Node did not return file checksum data")
2573 ignore_nodes.add(node.name)
2576 # Build per-checksum mapping from filename to nodes having it
2577 for (filename, checksum) in node_files.items():
2578 assert filename in nodefiles
2579 fileinfo[filename].setdefault(checksum, set()).add(node.name)
2581 for (filename, checksums) in fileinfo.items():
2582 assert compat.all(len(i) > 10 for i in checksums), "Invalid checksum"
2584 # Nodes having the file
2585 with_file = frozenset(node_name
2586 for nodes in fileinfo[filename].values()
2587 for node_name in nodes) - ignore_nodes
2589 expected_nodes = nodefiles[filename] - ignore_nodes
2591 # Nodes missing file
2592 missing_file = expected_nodes - with_file
2594 if filename in files_opt:
2596 errorif(missing_file and missing_file != expected_nodes,
2597 constants.CV_ECLUSTERFILECHECK, None,
2598 "File %s is optional, but it must exist on all or no"
2599 " nodes (not found on %s)",
2600 filename, utils.CommaJoin(utils.NiceSort(missing_file)))
2602 errorif(missing_file, constants.CV_ECLUSTERFILECHECK, None,
2603 "File %s is missing from node(s) %s", filename,
2604 utils.CommaJoin(utils.NiceSort(missing_file)))
2606 # Warn if a node has a file it shouldn't
2607 unexpected = with_file - expected_nodes
2609 constants.CV_ECLUSTERFILECHECK, None,
2610 "File %s should not exist on node(s) %s",
2611 filename, utils.CommaJoin(utils.NiceSort(unexpected)))
2613 # See if there are multiple versions of the file
2614 test = len(checksums) > 1
2616 variants = ["variant %s on %s" %
2617 (idx + 1, utils.CommaJoin(utils.NiceSort(nodes)))
2618 for (idx, (checksum, nodes)) in
2619 enumerate(sorted(checksums.items()))]
2623 errorif(test, constants.CV_ECLUSTERFILECHECK, None,
2624 "File %s found with %s different checksums (%s)",
2625 filename, len(checksums), "; ".join(variants))
2627 def _VerifyNodeDrbd(self, ninfo, nresult, instanceinfo, drbd_helper,
2629 """Verifies and the node DRBD status.
2631 @type ninfo: L{objects.Node}
2632 @param ninfo: the node to check
2633 @param nresult: the remote results for the node
2634 @param instanceinfo: the dict of instances
2635 @param drbd_helper: the configured DRBD usermode helper
2636 @param drbd_map: the DRBD map as returned by
2637 L{ganeti.config.ConfigWriter.ComputeDRBDMap}
2641 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2644 helper_result = nresult.get(constants.NV_DRBDHELPER, None)
2645 test = (helper_result == None)
2646 _ErrorIf(test, constants.CV_ENODEDRBDHELPER, node,
2647 "no drbd usermode helper returned")
2649 status, payload = helper_result
2651 _ErrorIf(test, constants.CV_ENODEDRBDHELPER, node,
2652 "drbd usermode helper check unsuccessful: %s", payload)
2653 test = status and (payload != drbd_helper)
2654 _ErrorIf(test, constants.CV_ENODEDRBDHELPER, node,
2655 "wrong drbd usermode helper: %s", payload)
2657 # compute the DRBD minors
2659 for minor, instance in drbd_map[node].items():
2660 test = instance not in instanceinfo
2661 _ErrorIf(test, constants.CV_ECLUSTERCFG, None,
2662 "ghost instance '%s' in temporary DRBD map", instance)
2663 # ghost instance should not be running, but otherwise we
2664 # don't give double warnings (both ghost instance and
2665 # unallocated minor in use)
2667 node_drbd[minor] = (instance, False)
2669 instance = instanceinfo[instance]
2670 node_drbd[minor] = (instance.name,
2671 instance.admin_state == constants.ADMINST_UP)
2673 # and now check them
2674 used_minors = nresult.get(constants.NV_DRBDLIST, [])
2675 test = not isinstance(used_minors, (tuple, list))
2676 _ErrorIf(test, constants.CV_ENODEDRBD, node,
2677 "cannot parse drbd status file: %s", str(used_minors))
2679 # we cannot check drbd status
2682 for minor, (iname, must_exist) in node_drbd.items():
2683 test = minor not in used_minors and must_exist
2684 _ErrorIf(test, constants.CV_ENODEDRBD, node,
2685 "drbd minor %d of instance %s is not active", minor, iname)
2686 for minor in used_minors:
2687 test = minor not in node_drbd
2688 _ErrorIf(test, constants.CV_ENODEDRBD, node,
2689 "unallocated drbd minor %d is in use", minor)
2691 def _UpdateNodeOS(self, ninfo, nresult, nimg):
2692 """Builds the node OS structures.
2694 @type ninfo: L{objects.Node}
2695 @param ninfo: the node to check
2696 @param nresult: the remote results for the node
2697 @param nimg: the node image object
2701 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2703 remote_os = nresult.get(constants.NV_OSLIST, None)
2704 test = (not isinstance(remote_os, list) or
2705 not compat.all(isinstance(v, list) and len(v) == 7
2706 for v in remote_os))
2708 _ErrorIf(test, constants.CV_ENODEOS, node,
2709 "node hasn't returned valid OS data")
2718 for (name, os_path, status, diagnose,
2719 variants, parameters, api_ver) in nresult[constants.NV_OSLIST]:
2721 if name not in os_dict:
2724 # parameters is a list of lists instead of list of tuples due to
2725 # JSON lacking a real tuple type, fix it:
2726 parameters = [tuple(v) for v in parameters]
2727 os_dict[name].append((os_path, status, diagnose,
2728 set(variants), set(parameters), set(api_ver)))
2730 nimg.oslist = os_dict
2732 def _VerifyNodeOS(self, ninfo, nimg, base):
2733 """Verifies the node OS list.
2735 @type ninfo: L{objects.Node}
2736 @param ninfo: the node to check
2737 @param nimg: the node image object
2738 @param base: the 'template' node we match against (e.g. from the master)
2742 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2744 assert not nimg.os_fail, "Entered _VerifyNodeOS with failed OS rpc?"
2746 beautify_params = lambda l: ["%s: %s" % (k, v) for (k, v) in l]
2747 for os_name, os_data in nimg.oslist.items():
2748 assert os_data, "Empty OS status for OS %s?!" % os_name
2749 f_path, f_status, f_diag, f_var, f_param, f_api = os_data[0]
2750 _ErrorIf(not f_status, constants.CV_ENODEOS, node,
2751 "Invalid OS %s (located at %s): %s", os_name, f_path, f_diag)
2752 _ErrorIf(len(os_data) > 1, constants.CV_ENODEOS, node,
2753 "OS '%s' has multiple entries (first one shadows the rest): %s",
2754 os_name, utils.CommaJoin([v[0] for v in os_data]))
2755 # comparisons with the 'base' image
2756 test = os_name not in base.oslist
2757 _ErrorIf(test, constants.CV_ENODEOS, node,
2758 "Extra OS %s not present on reference node (%s)",
2762 assert base.oslist[os_name], "Base node has empty OS status?"
2763 _, b_status, _, b_var, b_param, b_api = base.oslist[os_name][0]
2765 # base OS is invalid, skipping
2767 for kind, a, b in [("API version", f_api, b_api),
2768 ("variants list", f_var, b_var),
2769 ("parameters", beautify_params(f_param),
2770 beautify_params(b_param))]:
2771 _ErrorIf(a != b, constants.CV_ENODEOS, node,
2772 "OS %s for %s differs from reference node %s: [%s] vs. [%s]",
2773 kind, os_name, base.name,
2774 utils.CommaJoin(sorted(a)), utils.CommaJoin(sorted(b)))
2776 # check any missing OSes
2777 missing = set(base.oslist.keys()).difference(nimg.oslist.keys())
2778 _ErrorIf(missing, constants.CV_ENODEOS, node,
2779 "OSes present on reference node %s but missing on this node: %s",
2780 base.name, utils.CommaJoin(missing))
2782 def _VerifyOob(self, ninfo, nresult):
2783 """Verifies out of band functionality of a node.
2785 @type ninfo: L{objects.Node}
2786 @param ninfo: the node to check
2787 @param nresult: the remote results for the node
2791 # We just have to verify the paths on master and/or master candidates
2792 # as the oob helper is invoked on the master
2793 if ((ninfo.master_candidate or ninfo.master_capable) and
2794 constants.NV_OOB_PATHS in nresult):
2795 for path_result in nresult[constants.NV_OOB_PATHS]:
2796 self._ErrorIf(path_result, constants.CV_ENODEOOBPATH, node, path_result)
2798 def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
2799 """Verifies and updates the node volume data.
2801 This function will update a L{NodeImage}'s internal structures
2802 with data from the remote call.
2804 @type ninfo: L{objects.Node}
2805 @param ninfo: the node to check
2806 @param nresult: the remote results for the node
2807 @param nimg: the node image object
2808 @param vg_name: the configured VG name
2812 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2814 nimg.lvm_fail = True
2815 lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
2818 elif isinstance(lvdata, basestring):
2819 _ErrorIf(True, constants.CV_ENODELVM, node, "LVM problem on node: %s",
2820 utils.SafeEncode(lvdata))
2821 elif not isinstance(lvdata, dict):
2822 _ErrorIf(True, constants.CV_ENODELVM, node,
2823 "rpc call to node failed (lvlist)")
2825 nimg.volumes = lvdata
2826 nimg.lvm_fail = False
2828 def _UpdateNodeInstances(self, ninfo, nresult, nimg):
2829 """Verifies and updates the node instance list.
2831 If the listing was successful, then updates this node's instance
2832 list. Otherwise, it marks the RPC call as failed for the instance
2835 @type ninfo: L{objects.Node}
2836 @param ninfo: the node to check
2837 @param nresult: the remote results for the node
2838 @param nimg: the node image object
2841 idata = nresult.get(constants.NV_INSTANCELIST, None)
2842 test = not isinstance(idata, list)
2843 self._ErrorIf(test, constants.CV_ENODEHV, ninfo.name,
2844 "rpc call to node failed (instancelist): %s",
2845 utils.SafeEncode(str(idata)))
2847 nimg.hyp_fail = True
2849 nimg.instances = idata
2851 def _UpdateNodeInfo(self, ninfo, nresult, nimg, vg_name):
2852 """Verifies and computes a node information map
2854 @type ninfo: L{objects.Node}
2855 @param ninfo: the node to check
2856 @param nresult: the remote results for the node
2857 @param nimg: the node image object
2858 @param vg_name: the configured VG name
2862 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2864 # try to read free memory (from the hypervisor)
2865 hv_info = nresult.get(constants.NV_HVINFO, None)
2866 test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
2867 _ErrorIf(test, constants.CV_ENODEHV, node,
2868 "rpc call to node failed (hvinfo)")
2871 nimg.mfree = int(hv_info["memory_free"])
2872 except (ValueError, TypeError):
2873 _ErrorIf(True, constants.CV_ENODERPC, node,
2874 "node returned invalid nodeinfo, check hypervisor")
2876 # FIXME: devise a free space model for file based instances as well
2877 if vg_name is not None:
2878 test = (constants.NV_VGLIST not in nresult or
2879 vg_name not in nresult[constants.NV_VGLIST])
2880 _ErrorIf(test, constants.CV_ENODELVM, node,
2881 "node didn't return data for the volume group '%s'"
2882 " - it is either missing or broken", vg_name)
2885 nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
2886 except (ValueError, TypeError):
2887 _ErrorIf(True, constants.CV_ENODERPC, node,
2888 "node returned invalid LVM info, check LVM status")
2890 def _CollectDiskInfo(self, nodelist, node_image, instanceinfo):
2891 """Gets per-disk status information for all instances.
2893 @type nodelist: list of strings
2894 @param nodelist: Node names
2895 @type node_image: dict of (name, L{objects.Node})
2896 @param node_image: Node objects
2897 @type instanceinfo: dict of (name, L{objects.Instance})
2898 @param instanceinfo: Instance objects
2899 @rtype: {instance: {node: [(succes, payload)]}}
2900 @return: a dictionary of per-instance dictionaries with nodes as
2901 keys and disk information as values; the disk information is a
2902 list of tuples (success, payload)
2905 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2908 node_disks_devonly = {}
2909 diskless_instances = set()
2910 diskless = constants.DT_DISKLESS
2912 for nname in nodelist:
2913 node_instances = list(itertools.chain(node_image[nname].pinst,
2914 node_image[nname].sinst))
2915 diskless_instances.update(inst for inst in node_instances
2916 if instanceinfo[inst].disk_template == diskless)
2917 disks = [(inst, disk)
2918 for inst in node_instances
2919 for disk in instanceinfo[inst].disks]
2922 # No need to collect data
2925 node_disks[nname] = disks
2927 # _AnnotateDiskParams makes already copies of the disks
2929 for (inst, dev) in disks:
2930 (anno_disk,) = _AnnotateDiskParams(instanceinfo[inst], [dev], self.cfg)
2931 self.cfg.SetDiskID(anno_disk, nname)
2932 devonly.append(anno_disk)
2934 node_disks_devonly[nname] = devonly
2936 assert len(node_disks) == len(node_disks_devonly)
2938 # Collect data from all nodes with disks
2939 result = self.rpc.call_blockdev_getmirrorstatus_multi(node_disks.keys(),
2942 assert len(result) == len(node_disks)
2946 for (nname, nres) in result.items():
2947 disks = node_disks[nname]
2950 # No data from this node
2951 data = len(disks) * [(False, "node offline")]
2954 _ErrorIf(msg, constants.CV_ENODERPC, nname,
2955 "while getting disk information: %s", msg)
2957 # No data from this node
2958 data = len(disks) * [(False, msg)]
2961 for idx, i in enumerate(nres.payload):
2962 if isinstance(i, (tuple, list)) and len(i) == 2:
2965 logging.warning("Invalid result from node %s, entry %d: %s",
2967 data.append((False, "Invalid result from the remote node"))
2969 for ((inst, _), status) in zip(disks, data):
2970 instdisk.setdefault(inst, {}).setdefault(nname, []).append(status)
2972 # Add empty entries for diskless instances.
2973 for inst in diskless_instances:
2974 assert inst not in instdisk
2977 assert compat.all(len(statuses) == len(instanceinfo[inst].disks) and
2978 len(nnames) <= len(instanceinfo[inst].all_nodes) and
2979 compat.all(isinstance(s, (tuple, list)) and
2980 len(s) == 2 for s in statuses)
2981 for inst, nnames in instdisk.items()
2982 for nname, statuses in nnames.items())
2983 assert set(instdisk) == set(instanceinfo), "instdisk consistency failure"
2988 def _SshNodeSelector(group_uuid, all_nodes):
2989 """Create endless iterators for all potential SSH check hosts.
2992 nodes = [node for node in all_nodes
2993 if (node.group != group_uuid and
2995 keyfunc = operator.attrgetter("group")
2997 return map(itertools.cycle,
2998 [sorted(map(operator.attrgetter("name"), names))
2999 for _, names in itertools.groupby(sorted(nodes, key=keyfunc),
3003 def _SelectSshCheckNodes(cls, group_nodes, group_uuid, all_nodes):
3004 """Choose which nodes should talk to which other nodes.
3006 We will make nodes contact all nodes in their group, and one node from
3009 @warning: This algorithm has a known issue if one node group is much
3010 smaller than others (e.g. just one node). In such a case all other
3011 nodes will talk to the single node.
3014 online_nodes = sorted(node.name for node in group_nodes if not node.offline)
3015 sel = cls._SshNodeSelector(group_uuid, all_nodes)
3017 return (online_nodes,
3018 dict((name, sorted([i.next() for i in sel]))
3019 for name in online_nodes))
3021 def BuildHooksEnv(self):
3024 Cluster-Verify hooks just ran in the post phase and their failure makes
3025 the output be logged in the verify output and the verification to fail.
3029 "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
3032 env.update(("NODE_TAGS_%s" % node.name, " ".join(node.GetTags()))
3033 for node in self.my_node_info.values())
3037 def BuildHooksNodes(self):
3038 """Build hooks nodes.
3041 return ([], self.my_node_names)
3043 def Exec(self, feedback_fn):
3044 """Verify integrity of the node group, performing various test on nodes.
3047 # This method has too many local variables. pylint: disable=R0914
3048 feedback_fn("* Verifying group '%s'" % self.group_info.name)
3050 if not self.my_node_names:
3052 feedback_fn("* Empty node group, skipping verification")
3056 _ErrorIf = self._ErrorIf # pylint: disable=C0103
3057 verbose = self.op.verbose
3058 self._feedback_fn = feedback_fn
3060 vg_name = self.cfg.GetVGName()
3061 drbd_helper = self.cfg.GetDRBDHelper()
3062 cluster = self.cfg.GetClusterInfo()
3063 groupinfo = self.cfg.GetAllNodeGroupsInfo()
3064 hypervisors = cluster.enabled_hypervisors
3065 node_data_list = [self.my_node_info[name] for name in self.my_node_names]
3067 i_non_redundant = [] # Non redundant instances
3068 i_non_a_balanced = [] # Non auto-balanced instances
3069 i_offline = 0 # Count of offline instances
3070 n_offline = 0 # Count of offline nodes
3071 n_drained = 0 # Count of nodes being drained
3072 node_vol_should = {}
3074 # FIXME: verify OS list
3077 filemap = _ComputeAncillaryFiles(cluster, False)
3079 # do local checksums
3080 master_node = self.master_node = self.cfg.GetMasterNode()
3081 master_ip = self.cfg.GetMasterIP()
3083 feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
3086 if self.cfg.GetUseExternalMipScript():
3087 user_scripts.append(constants.EXTERNAL_MASTER_SETUP_SCRIPT)
3089 node_verify_param = {
3090 constants.NV_FILELIST:
3091 utils.UniqueSequence(filename
3092 for files in filemap
3093 for filename in files),
3094 constants.NV_NODELIST:
3095 self._SelectSshCheckNodes(node_data_list, self.group_uuid,
3096 self.all_node_info.values()),
3097 constants.NV_HYPERVISOR: hypervisors,
3098 constants.NV_HVPARAMS:
3099 _GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
3100 constants.NV_NODENETTEST: [(node.name, node.primary_ip, node.secondary_ip)
3101 for node in node_data_list
3102 if not node.offline],
3103 constants.NV_INSTANCELIST: hypervisors,
3104 constants.NV_VERSION: None,
3105 constants.NV_HVINFO: self.cfg.GetHypervisorType(),
3106 constants.NV_NODESETUP: None,
3107 constants.NV_TIME: None,
3108 constants.NV_MASTERIP: (master_node, master_ip),
3109 constants.NV_OSLIST: None,
3110 constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
3111 constants.NV_USERSCRIPTS: user_scripts,
3114 if vg_name is not None:
3115 node_verify_param[constants.NV_VGLIST] = None
3116 node_verify_param[constants.NV_LVLIST] = vg_name
3117 node_verify_param[constants.NV_PVLIST] = [vg_name]
3118 node_verify_param[constants.NV_DRBDLIST] = None
3121 node_verify_param[constants.NV_DRBDHELPER] = drbd_helper
3124 # FIXME: this needs to be changed per node-group, not cluster-wide
3126 default_nicpp = cluster.nicparams[constants.PP_DEFAULT]
3127 if default_nicpp[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
3128 bridges.add(default_nicpp[constants.NIC_LINK])
3129 for instance in self.my_inst_info.values():
3130 for nic in instance.nics:
3131 full_nic = cluster.SimpleFillNIC(nic.nicparams)
3132 if full_nic[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
3133 bridges.add(full_nic[constants.NIC_LINK])
3136 node_verify_param[constants.NV_BRIDGES] = list(bridges)
3138 # Build our expected cluster state
3139 node_image = dict((node.name, self.NodeImage(offline=node.offline,
3141 vm_capable=node.vm_capable))
3142 for node in node_data_list)
3146 for node in self.all_node_info.values():
3147 path = _SupportsOob(self.cfg, node)
3148 if path and path not in oob_paths:
3149 oob_paths.append(path)
3152 node_verify_param[constants.NV_OOB_PATHS] = oob_paths
3154 for instance in self.my_inst_names:
3155 inst_config = self.my_inst_info[instance]
3156 if inst_config.admin_state == constants.ADMINST_OFFLINE:
3159 for nname in inst_config.all_nodes:
3160 if nname not in node_image:
3161 gnode = self.NodeImage(name=nname)
3162 gnode.ghost = (nname not in self.all_node_info)
3163 node_image[nname] = gnode
3165 inst_config.MapLVsByNode(node_vol_should)
3167 pnode = inst_config.primary_node
3168 node_image[pnode].pinst.append(instance)
3170 for snode in inst_config.secondary_nodes:
3171 nimg = node_image[snode]
3172 nimg.sinst.append(instance)
3173 if pnode not in nimg.sbp:
3174 nimg.sbp[pnode] = []
3175 nimg.sbp[pnode].append(instance)
3177 # At this point, we have the in-memory data structures complete,
3178 # except for the runtime information, which we'll gather next
3180 # Due to the way our RPC system works, exact response times cannot be
3181 # guaranteed (e.g. a broken node could run into a timeout). By keeping the
3182 # time before and after executing the request, we can at least have a time
3184 nvinfo_starttime = time.time()
3185 all_nvinfo = self.rpc.call_node_verify(self.my_node_names,
3187 self.cfg.GetClusterName())
3188 nvinfo_endtime = time.time()
3190 if self.extra_lv_nodes and vg_name is not None:
3192 self.rpc.call_node_verify(self.extra_lv_nodes,
3193 {constants.NV_LVLIST: vg_name},
3194 self.cfg.GetClusterName())
3196 extra_lv_nvinfo = {}
3198 all_drbd_map = self.cfg.ComputeDRBDMap()
3200 feedback_fn("* Gathering disk information (%s nodes)" %
3201 len(self.my_node_names))
3202 instdisk = self._CollectDiskInfo(self.my_node_names, node_image,
3205 feedback_fn("* Verifying configuration file consistency")
3207 # If not all nodes are being checked, we need to make sure the master node
3208 # and a non-checked vm_capable node are in the list.
3209 absent_nodes = set(self.all_node_info).difference(self.my_node_info)
3211 vf_nvinfo = all_nvinfo.copy()
3212 vf_node_info = list(self.my_node_info.values())
3213 additional_nodes = []
3214 if master_node not in self.my_node_info:
3215 additional_nodes.append(master_node)
3216 vf_node_info.append(self.all_node_info[master_node])
3217 # Add the first vm_capable node we find which is not included
3218 for node in absent_nodes:
3219 nodeinfo = self.all_node_info[node]
3220 if nodeinfo.vm_capable and not nodeinfo.offline:
3221 additional_nodes.append(node)
3222 vf_node_info.append(self.all_node_info[node])
3224 key = constants.NV_FILELIST
3225 vf_nvinfo.update(self.rpc.call_node_verify(additional_nodes,
3226 {key: node_verify_param[key]},
3227 self.cfg.GetClusterName()))
3229 vf_nvinfo = all_nvinfo
3230 vf_node_info = self.my_node_info.values()
3232 self._VerifyFiles(_ErrorIf, vf_node_info, master_node, vf_nvinfo, filemap)
3234 feedback_fn("* Verifying node status")
3238 for node_i in node_data_list:
3240 nimg = node_image[node]
3244 feedback_fn("* Skipping offline node %s" % (node,))
3248 if node == master_node:
3250 elif node_i.master_candidate:
3251 ntype = "master candidate"
3252 elif node_i.drained:
3258 feedback_fn("* Verifying node %s (%s)" % (node, ntype))
3260 msg = all_nvinfo[node].fail_msg
3261 _ErrorIf(msg, constants.CV_ENODERPC, node, "while contacting node: %s",
3264 nimg.rpc_fail = True
3267 nresult = all_nvinfo[node].payload
3269 nimg.call_ok = self._VerifyNode(node_i, nresult)
3270 self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
3271 self._VerifyNodeNetwork(node_i, nresult)
3272 self._VerifyNodeUserScripts(node_i, nresult)
3273 self._VerifyOob(node_i, nresult)
3276 self._VerifyNodeLVM(node_i, nresult, vg_name)
3277 self._VerifyNodeDrbd(node_i, nresult, self.all_inst_info, drbd_helper,
3280 self._UpdateNodeVolumes(node_i, nresult, nimg, vg_name)
3281 self._UpdateNodeInstances(node_i, nresult, nimg)
3282 self._UpdateNodeInfo(node_i, nresult, nimg, vg_name)
3283 self._UpdateNodeOS(node_i, nresult, nimg)
3285 if not nimg.os_fail:
3286 if refos_img is None:
3288 self._VerifyNodeOS(node_i, nimg, refos_img)
3289 self._VerifyNodeBridges(node_i, nresult, bridges)
3291 # Check whether all running instancies are primary for the node. (This
3292 # can no longer be done from _VerifyInstance below, since some of the
3293 # wrong instances could be from other node groups.)
3294 non_primary_inst = set(nimg.instances).difference(nimg.pinst)
3296 for inst in non_primary_inst:
3297 test = inst in self.all_inst_info
3298 _ErrorIf(test, constants.CV_EINSTANCEWRONGNODE, inst,
3299 "instance should not run on node %s", node_i.name)
3300 _ErrorIf(not test, constants.CV_ENODEORPHANINSTANCE, node_i.name,
3301 "node is running unknown instance %s", inst)
3303 for node, result in extra_lv_nvinfo.items():
3304 self._UpdateNodeVolumes(self.all_node_info[node], result.payload,
3305 node_image[node], vg_name)
3307 feedback_fn("* Verifying instance status")
3308 for instance in self.my_inst_names:
3310 feedback_fn("* Verifying instance %s" % instance)
3311 inst_config = self.my_inst_info[instance]
3312 self._VerifyInstance(instance, inst_config, node_image,
3314 inst_nodes_offline = []
3316 pnode = inst_config.primary_node
3317 pnode_img = node_image[pnode]
3318 _ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
3319 constants.CV_ENODERPC, pnode, "instance %s, connection to"
3320 " primary node failed", instance)
3322 _ErrorIf(inst_config.admin_state == constants.ADMINST_UP and
3324 constants.CV_EINSTANCEBADNODE, instance,
3325 "instance is marked as running and lives on offline node %s",
3326 inst_config.primary_node)
3328 # If the instance is non-redundant we cannot survive losing its primary
3329 # node, so we are not N+1 compliant. On the other hand we have no disk
3330 # templates with more than one secondary so that situation is not well
3332 # FIXME: does not support file-backed instances
3333 if not inst_config.secondary_nodes:
3334 i_non_redundant.append(instance)
3336 _ErrorIf(len(inst_config.secondary_nodes) > 1,
3337 constants.CV_EINSTANCELAYOUT,
3338 instance, "instance has multiple secondary nodes: %s",
3339 utils.CommaJoin(inst_config.secondary_nodes),
3340 code=self.ETYPE_WARNING)
3342 if inst_config.disk_template in constants.DTS_INT_MIRROR:
3343 pnode = inst_config.primary_node
3344 instance_nodes = utils.NiceSort(inst_config.all_nodes)
3345 instance_groups = {}
3347 for node in instance_nodes:
3348 instance_groups.setdefault(self.all_node_info[node].group,
3352 "%s (group %s)" % (utils.CommaJoin(nodes), groupinfo[group].name)
3353 # Sort so that we always list the primary node first.
3354 for group, nodes in sorted(instance_groups.items(),
3355 key=lambda (_, nodes): pnode in nodes,
3358 self._ErrorIf(len(instance_groups) > 1,
3359 constants.CV_EINSTANCESPLITGROUPS,
3360 instance, "instance has primary and secondary nodes in"
3361 " different groups: %s", utils.CommaJoin(pretty_list),
3362 code=self.ETYPE_WARNING)
3364 if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
3365 i_non_a_balanced.append(instance)
3367 for snode in inst_config.secondary_nodes:
3368 s_img = node_image[snode]
3369 _ErrorIf(s_img.rpc_fail and not s_img.offline, constants.CV_ENODERPC,
3370 snode, "instance %s, connection to secondary node failed",
3374 inst_nodes_offline.append(snode)
3376 # warn that the instance lives on offline nodes
3377 _ErrorIf(inst_nodes_offline, constants.CV_EINSTANCEBADNODE, instance,
3378 "instance has offline secondary node(s) %s",
3379 utils.CommaJoin(inst_nodes_offline))
3380 # ... or ghost/non-vm_capable nodes
3381 for node in inst_config.all_nodes:
3382 _ErrorIf(node_image[node].ghost, constants.CV_EINSTANCEBADNODE,
3383 instance, "instance lives on ghost node %s", node)
3384 _ErrorIf(not node_image[node].vm_capable, constants.CV_EINSTANCEBADNODE,
3385 instance, "instance lives on non-vm_capable node %s", node)
3387 feedback_fn("* Verifying orphan volumes")
3388 reserved = utils.FieldSet(*cluster.reserved_lvs)
3390 # We will get spurious "unknown volume" warnings if any node of this group
3391 # is secondary for an instance whose primary is in another group. To avoid
3392 # them, we find these instances and add their volumes to node_vol_should.
3393 for inst in self.all_inst_info.values():
3394 for secondary in inst.secondary_nodes:
3395 if (secondary in self.my_node_info
3396 and inst.name not in self.my_inst_info):
3397 inst.MapLVsByNode(node_vol_should)
3400 self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
3402 if constants.VERIFY_NPLUSONE_MEM not in self.op.skip_checks:
3403 feedback_fn("* Verifying N+1 Memory redundancy")
3404 self._VerifyNPlusOneMemory(node_image, self.my_inst_info)
3406 feedback_fn("* Other Notes")
3408 feedback_fn(" - NOTICE: %d non-redundant instance(s) found."
3409 % len(i_non_redundant))
3411 if i_non_a_balanced:
3412 feedback_fn(" - NOTICE: %d non-auto-balanced instance(s) found."
3413 % len(i_non_a_balanced))
3416 feedback_fn(" - NOTICE: %d offline instance(s) found." % i_offline)
3419 feedback_fn(" - NOTICE: %d offline node(s) found." % n_offline)
3422 feedback_fn(" - NOTICE: %d drained node(s) found." % n_drained)
3426 def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
3427 """Analyze the post-hooks' result
3429 This method analyses the hook result, handles it, and sends some
3430 nicely-formatted feedback back to the user.
3432 @param phase: one of L{constants.HOOKS_PHASE_POST} or
3433 L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
3434 @param hooks_results: the results of the multi-node hooks rpc call
3435 @param feedback_fn: function used send feedback back to the caller
3436 @param lu_result: previous Exec result
3437 @return: the new Exec result, based on the previous result
3441 # We only really run POST phase hooks, only for non-empty groups,
3442 # and are only interested in their results
3443 if not self.my_node_names:
3446 elif phase == constants.HOOKS_PHASE_POST:
3447 # Used to change hooks' output to proper indentation
3448 feedback_fn("* Hooks Results")
3449 assert hooks_results, "invalid result from hooks"
3451 for node_name in hooks_results:
3452 res = hooks_results[node_name]
3454 test = msg and not res.offline
3455 self._ErrorIf(test, constants.CV_ENODEHOOKS, node_name,
3456 "Communication failure in hooks execution: %s", msg)
3457 if res.offline or msg:
3458 # No need to investigate payload if node is offline or gave
3461 for script, hkr, output in res.payload:
3462 test = hkr == constants.HKR_FAIL
3463 self._ErrorIf(test, constants.CV_ENODEHOOKS, node_name,
3464 "Script %s failed, output:", script)
3466 output = self._HOOKS_INDENT_RE.sub(" ", output)
3467 feedback_fn("%s" % output)
3473 class LUClusterVerifyDisks(NoHooksLU):
3474 """Verifies the cluster disks status.
3479 def ExpandNames(self):
3480 self.share_locks = _ShareAll()
3481 self.needed_locks = {
3482 locking.LEVEL_NODEGROUP: locking.ALL_SET,
3485 def Exec(self, feedback_fn):
3486 group_names = self.owned_locks(locking.LEVEL_NODEGROUP)
3488 # Submit one instance of L{opcodes.OpGroupVerifyDisks} per node group
3489 return ResultWithJobs([[opcodes.OpGroupVerifyDisks(group_name=group)]
3490 for group in group_names])
3493 class LUGroupVerifyDisks(NoHooksLU):
3494 """Verifies the status of all disks in a node group.
3499 def ExpandNames(self):
3500 # Raises errors.OpPrereqError on its own if group can't be found
3501 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
3503 self.share_locks = _ShareAll()
3504 self.needed_locks = {
3505 locking.LEVEL_INSTANCE: [],
3506 locking.LEVEL_NODEGROUP: [],
3507 locking.LEVEL_NODE: [],
3510 def DeclareLocks(self, level):
3511 if level == locking.LEVEL_INSTANCE:
3512 assert not self.needed_locks[locking.LEVEL_INSTANCE]
3514 # Lock instances optimistically, needs verification once node and group
3515 # locks have been acquired
3516 self.needed_locks[locking.LEVEL_INSTANCE] = \
3517 self.cfg.GetNodeGroupInstances(self.group_uuid)
3519 elif level == locking.LEVEL_NODEGROUP:
3520 assert not self.needed_locks[locking.LEVEL_NODEGROUP]
3522 self.needed_locks[locking.LEVEL_NODEGROUP] = \
3523 set([self.group_uuid] +
3524 # Lock all groups used by instances optimistically; this requires
3525 # going via the node before it's locked, requiring verification
3528 for instance_name in self.owned_locks(locking.LEVEL_INSTANCE)
3529 for group_uuid in self.cfg.GetInstanceNodeGroups(instance_name)])
3531 elif level == locking.LEVEL_NODE:
3532 # This will only lock the nodes in the group to be verified which contain
3534 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
3535 self._LockInstancesNodes()
3537 # Lock all nodes in group to be verified
3538 assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
3539 member_nodes = self.cfg.GetNodeGroup(self.group_uuid).members
3540 self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
3542 def CheckPrereq(self):
3543 owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
3544 owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
3545 owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
3547 assert self.group_uuid in owned_groups
3549 # Check if locked instances are still correct
3550 _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
3552 # Get instance information
3553 self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
3555 # Check if node groups for locked instances are still correct
3556 _CheckInstancesNodeGroups(self.cfg, self.instances,
3557 owned_groups, owned_nodes, self.group_uuid)
3559 def Exec(self, feedback_fn):
3560 """Verify integrity of cluster disks.
3562 @rtype: tuple of three items
3563 @return: a tuple of (dict of node-to-node_error, list of instances
3564 which need activate-disks, dict of instance: (node, volume) for
3569 res_instances = set()
3572 nv_dict = _MapInstanceDisksToNodes([inst
3573 for inst in self.instances.values()
3574 if inst.admin_state == constants.ADMINST_UP])
3577 nodes = utils.NiceSort(set(self.owned_locks(locking.LEVEL_NODE)) &
3578 set(self.cfg.GetVmCapableNodeList()))
3580 node_lvs = self.rpc.call_lv_list(nodes, [])
3582 for (node, node_res) in node_lvs.items():
3583 if node_res.offline:
3586 msg = node_res.fail_msg
3588 logging.warning("Error enumerating LVs on node %s: %s", node, msg)
3589 res_nodes[node] = msg
3592 for lv_name, (_, _, lv_online) in node_res.payload.items():
3593 inst = nv_dict.pop((node, lv_name), None)
3594 if not (lv_online or inst is None):
3595 res_instances.add(inst)
3597 # any leftover items in nv_dict are missing LVs, let's arrange the data
3599 for key, inst in nv_dict.iteritems():
3600 res_missing.setdefault(inst, []).append(list(key))
3602 return (res_nodes, list(res_instances), res_missing)
3605 class LUClusterRepairDiskSizes(NoHooksLU):
3606 """Verifies the cluster disks sizes.
3611 def ExpandNames(self):
3612 if self.op.instances:
3613 self.wanted_names = _GetWantedInstances(self, self.op.instances)
3614 self.needed_locks = {
3615 locking.LEVEL_NODE_RES: [],
3616 locking.LEVEL_INSTANCE: self.wanted_names,
3618 self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
3620 self.wanted_names = None
3621 self.needed_locks = {
3622 locking.LEVEL_NODE_RES: locking.ALL_SET,
3623 locking.LEVEL_INSTANCE: locking.ALL_SET,
3625 self.share_locks = {
3626 locking.LEVEL_NODE_RES: 1,
3627 locking.LEVEL_INSTANCE: 0,
3630 def DeclareLocks(self, level):
3631 if level == locking.LEVEL_NODE_RES and self.wanted_names is not None:
3632 self._LockInstancesNodes(primary_only=True, level=level)
3634 def CheckPrereq(self):
3635 """Check prerequisites.
3637 This only checks the optional instance list against the existing names.
3640 if self.wanted_names is None:
3641 self.wanted_names = self.owned_locks(locking.LEVEL_INSTANCE)
3643 self.wanted_instances = \
3644 map(compat.snd, self.cfg.GetMultiInstanceInfo(self.wanted_names))
3646 def _EnsureChildSizes(self, disk):
3647 """Ensure children of the disk have the needed disk size.
3649 This is valid mainly for DRBD8 and fixes an issue where the
3650 children have smaller disk size.
3652 @param disk: an L{ganeti.objects.Disk} object
3655 if disk.dev_type == constants.LD_DRBD8:
3656 assert disk.children, "Empty children for DRBD8?"
3657 fchild = disk.children[0]
3658 mismatch = fchild.size < disk.size
3660 self.LogInfo("Child disk has size %d, parent %d, fixing",
3661 fchild.size, disk.size)
3662 fchild.size = disk.size
3664 # and we recurse on this child only, not on the metadev
3665 return self._EnsureChildSizes(fchild) or mismatch
3669 def Exec(self, feedback_fn):
3670 """Verify the size of cluster disks.
3673 # TODO: check child disks too
3674 # TODO: check differences in size between primary/secondary nodes
3676 for instance in self.wanted_instances:
3677 pnode = instance.primary_node
3678 if pnode not in per_node_disks:
3679 per_node_disks[pnode] = []
3680 for idx, disk in enumerate(instance.disks):
3681 per_node_disks[pnode].append((instance, idx, disk))
3683 assert not (frozenset(per_node_disks.keys()) -
3684 self.owned_locks(locking.LEVEL_NODE_RES)), \
3685 "Not owning correct locks"
3686 assert not self.owned_locks(locking.LEVEL_NODE)
3689 for node, dskl in per_node_disks.items():
3690 newl = [v[2].Copy() for v in dskl]
3692 self.cfg.SetDiskID(dsk, node)
3693 result = self.rpc.call_blockdev_getsize(node, newl)
3695 self.LogWarning("Failure in blockdev_getsize call to node"
3696 " %s, ignoring", node)
3698 if len(result.payload) != len(dskl):
3699 logging.warning("Invalid result from node %s: len(dksl)=%d,"
3700 " result.payload=%s", node, len(dskl), result.payload)
3701 self.LogWarning("Invalid result from node %s, ignoring node results",
3704 for ((instance, idx, disk), size) in zip(dskl, result.payload):
3706 self.LogWarning("Disk %d of instance %s did not return size"
3707 " information, ignoring", idx, instance.name)
3709 if not isinstance(size, (int, long)):
3710 self.LogWarning("Disk %d of instance %s did not return valid"
3711 " size information, ignoring", idx, instance.name)
3714 if size != disk.size:
3715 self.LogInfo("Disk %d of instance %s has mismatched size,"
3716 " correcting: recorded %d, actual %d", idx,
3717 instance.name, disk.size, size)
3719 self.cfg.Update(instance, feedback_fn)
3720 changed.append((instance.name, idx, size))
3721 if self._EnsureChildSizes(disk):
3722 self.cfg.Update(instance, feedback_fn)
3723 changed.append((instance.name, idx, disk.size))
3727 class LUClusterRename(LogicalUnit):
3728 """Rename the cluster.
3731 HPATH = "cluster-rename"
3732 HTYPE = constants.HTYPE_CLUSTER
3734 def BuildHooksEnv(self):
3739 "OP_TARGET": self.cfg.GetClusterName(),
3740 "NEW_NAME": self.op.name,
3743 def BuildHooksNodes(self):
3744 """Build hooks nodes.
3747 return ([self.cfg.GetMasterNode()], self.cfg.GetNodeList())
3749 def CheckPrereq(self):
3750 """Verify that the passed name is a valid one.
3753 hostname = netutils.GetHostname(name=self.op.name,
3754 family=self.cfg.GetPrimaryIPFamily())
3756 new_name = hostname.name
3757 self.ip = new_ip = hostname.ip
3758 old_name = self.cfg.GetClusterName()
3759 old_ip = self.cfg.GetMasterIP()
3760 if new_name == old_name and new_ip == old_ip:
3761 raise errors.OpPrereqError("Neither the name nor the IP address of the"
3762 " cluster has changed",
3764 if new_ip != old_ip:
3765 if netutils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
3766 raise errors.OpPrereqError("The given cluster IP address (%s) is"
3767 " reachable on the network" %
3768 new_ip, errors.ECODE_NOTUNIQUE)
3770 self.op.name = new_name
3772 def Exec(self, feedback_fn):
3773 """Rename the cluster.
3776 clustername = self.op.name
3779 # shutdown the master IP
3780 master_params = self.cfg.GetMasterNetworkParameters()
3781 ems = self.cfg.GetUseExternalMipScript()
3782 result = self.rpc.call_node_deactivate_master_ip(master_params.name,
3784 result.Raise("Could not disable the master role")
3787 cluster = self.cfg.GetClusterInfo()
3788 cluster.cluster_name = clustername
3789 cluster.master_ip = new_ip
3790 self.cfg.Update(cluster, feedback_fn)
3792 # update the known hosts file
3793 ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
3794 node_list = self.cfg.GetOnlineNodeList()
3796 node_list.remove(master_params.name)
3799 _UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
3801 master_params.ip = new_ip
3802 result = self.rpc.call_node_activate_master_ip(master_params.name,
3804 msg = result.fail_msg
3806 self.LogWarning("Could not re-enable the master role on"
3807 " the master, please restart manually: %s", msg)
3812 def _ValidateNetmask(cfg, netmask):
3813 """Checks if a netmask is valid.
3815 @type cfg: L{config.ConfigWriter}
3816 @param cfg: The cluster configuration
3818 @param netmask: the netmask to be verified
3819 @raise errors.OpPrereqError: if the validation fails
3822 ip_family = cfg.GetPrimaryIPFamily()
3824 ipcls = netutils.IPAddress.GetClassFromIpFamily(ip_family)
3825 except errors.ProgrammerError:
3826 raise errors.OpPrereqError("Invalid primary ip family: %s." %
3828 if not ipcls.ValidateNetmask(netmask):
3829 raise errors.OpPrereqError("CIDR netmask (%s) not valid" %
3833 class LUClusterSetParams(LogicalUnit):
3834 """Change the parameters of the cluster.
3837 HPATH = "cluster-modify"
3838 HTYPE = constants.HTYPE_CLUSTER
3841 def CheckArguments(self):
3845 if self.op.uid_pool:
3846 uidpool.CheckUidPool(self.op.uid_pool)
3848 if self.op.add_uids:
3849 uidpool.CheckUidPool(self.op.add_uids)
3851 if self.op.remove_uids:
3852 uidpool.CheckUidPool(self.op.remove_uids)
3854 if self.op.master_netmask is not None:
3855 _ValidateNetmask(self.cfg, self.op.master_netmask)
3857 if self.op.diskparams:
3858 for dt_params in self.op.diskparams.values():
3859 utils.ForceDictType(dt_params, constants.DISK_DT_TYPES)
3861 utils.VerifyDictOptions(self.op.diskparams, constants.DISK_DT_DEFAULTS)
3862 except errors.OpPrereqError, err:
3863 raise errors.OpPrereqError("While verify diskparams options: %s" % err,
3866 def ExpandNames(self):
3867 # FIXME: in the future maybe other cluster params won't require checking on
3868 # all nodes to be modified.
3869 self.needed_locks = {
3870 locking.LEVEL_NODE: locking.ALL_SET,
3871 locking.LEVEL_INSTANCE: locking.ALL_SET,
3872 locking.LEVEL_NODEGROUP: locking.ALL_SET,
3874 self.share_locks = {
3875 locking.LEVEL_NODE: 1,
3876 locking.LEVEL_INSTANCE: 1,
3877 locking.LEVEL_NODEGROUP: 1,
3880 def BuildHooksEnv(self):
3885 "OP_TARGET": self.cfg.GetClusterName(),
3886 "NEW_VG_NAME": self.op.vg_name,
3889 def BuildHooksNodes(self):
3890 """Build hooks nodes.
3893 mn = self.cfg.GetMasterNode()
3896 def CheckPrereq(self):
3897 """Check prerequisites.
3899 This checks whether the given params don't conflict and
3900 if the given volume group is valid.
3903 if self.op.vg_name is not None and not self.op.vg_name:
3904 if self.cfg.HasAnyDiskOfType(constants.LD_LV):
3905 raise errors.OpPrereqError("Cannot disable lvm storage while lvm-based"
3906 " instances exist", errors.ECODE_INVAL)
3908 if self.op.drbd_helper is not None and not self.op.drbd_helper:
3909 if self.cfg.HasAnyDiskOfType(constants.LD_DRBD8):
3910 raise errors.OpPrereqError("Cannot disable drbd helper while"
3911 " drbd-based instances exist",
3914 node_list = self.owned_locks(locking.LEVEL_NODE)
3916 # if vg_name not None, checks given volume group on all nodes
3918 vglist = self.rpc.call_vg_list(node_list)
3919 for node in node_list:
3920 msg = vglist[node].fail_msg
3922 # ignoring down node
3923 self.LogWarning("Error while gathering data on node %s"
3924 " (ignoring node): %s", node, msg)
3926 vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
3928 constants.MIN_VG_SIZE)
3930 raise errors.OpPrereqError("Error on node '%s': %s" %
3931 (node, vgstatus), errors.ECODE_ENVIRON)
3933 if self.op.drbd_helper:
3934 # checks given drbd helper on all nodes
3935 helpers = self.rpc.call_drbd_helper(node_list)
3936 for (node, ninfo) in self.cfg.GetMultiNodeInfo(node_list):
3938 self.LogInfo("Not checking drbd helper on offline node %s", node)
3940 msg = helpers[node].fail_msg
3942 raise errors.OpPrereqError("Error checking drbd helper on node"
3943 " '%s': %s" % (node, msg),
3944 errors.ECODE_ENVIRON)
3945 node_helper = helpers[node].payload
3946 if node_helper != self.op.drbd_helper:
3947 raise errors.OpPrereqError("Error on node '%s': drbd helper is %s" %
3948 (node, node_helper), errors.ECODE_ENVIRON)
3950 self.cluster = cluster = self.cfg.GetClusterInfo()
3951 # validate params changes
3952 if self.op.beparams:
3953 objects.UpgradeBeParams(self.op.beparams)
3954 utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
3955 self.new_beparams = cluster.SimpleFillBE(self.op.beparams)
3957 if self.op.ndparams:
3958 utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
3959 self.new_ndparams = cluster.SimpleFillND(self.op.ndparams)
3961 # TODO: we need a more general way to handle resetting
3962 # cluster-level parameters to default values
3963 if self.new_ndparams["oob_program"] == "":
3964 self.new_ndparams["oob_program"] = \
3965 constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
3967 if self.op.hv_state:
3968 new_hv_state = _MergeAndVerifyHvState(self.op.hv_state,
3969 self.cluster.hv_state_static)
3970 self.new_hv_state = dict((hv, cluster.SimpleFillHvState(values))
3971 for hv, values in new_hv_state.items())
3973 if self.op.disk_state:
3974 new_disk_state = _MergeAndVerifyDiskState(self.op.disk_state,
3975 self.cluster.disk_state_static)
3976 self.new_disk_state = \
3977 dict((storage, dict((name, cluster.SimpleFillDiskState(values))
3978 for name, values in svalues.items()))
3979 for storage, svalues in new_disk_state.items())
3982 self.new_ipolicy = _GetUpdatedIPolicy(cluster.ipolicy, self.op.ipolicy,
3985 all_instances = self.cfg.GetAllInstancesInfo().values()
3987 for group in self.cfg.GetAllNodeGroupsInfo().values():
3988 instances = frozenset([inst for inst in all_instances
3989 if compat.any(node in group.members
3990 for node in inst.all_nodes)])
3991 new_ipolicy = objects.FillIPolicy(self.new_ipolicy, group.ipolicy)
3992 new = _ComputeNewInstanceViolations(_CalculateGroupIPolicy(cluster,
3994 new_ipolicy, instances)
3996 violations.update(new)
3999 self.LogWarning("After the ipolicy change the following instances"
4000 " violate them: %s",
4001 utils.CommaJoin(utils.NiceSort(violations)))
4003 if self.op.nicparams:
4004 utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
4005 self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
4006 objects.NIC.CheckParameterSyntax(self.new_nicparams)
4009 # check all instances for consistency
4010 for instance in self.cfg.GetAllInstancesInfo().values():
4011 for nic_idx, nic in enumerate(instance.nics):
4012 params_copy = copy.deepcopy(nic.nicparams)
4013 params_filled = objects.FillDict(self.new_nicparams, params_copy)
4015 # check parameter syntax
4017 objects.NIC.CheckParameterSyntax(params_filled)
4018 except errors.ConfigurationError, err:
4019 nic_errors.append("Instance %s, nic/%d: %s" %
4020 (instance.name, nic_idx, err))
4022 # if we're moving instances to routed, check that they have an ip
4023 target_mode = params_filled[constants.NIC_MODE]
4024 if target_mode == constants.NIC_MODE_ROUTED and not nic.ip:
4025 nic_errors.append("Instance %s, nic/%d: routed NIC with no ip"
4026 " address" % (instance.name, nic_idx))
4028 raise errors.OpPrereqError("Cannot apply the change, errors:\n%s" %
4029 "\n".join(nic_errors))
4031 # hypervisor list/parameters
4032 self.new_hvparams = new_hvp = objects.FillDict(cluster.hvparams, {})
4033 if self.op.hvparams:
4034 for hv_name, hv_dict in self.op.hvparams.items():
4035 if hv_name not in self.new_hvparams:
4036 self.new_hvparams[hv_name] = hv_dict
4038 self.new_hvparams[hv_name].update(hv_dict)
4040 # disk template parameters
4041 self.new_diskparams = objects.FillDict(cluster.diskparams, {})
4042 if self.op.diskparams:
4043 for dt_name, dt_params in self.op.diskparams.items():
4044 if dt_name not in self.op.diskparams:
4045 self.new_diskparams[dt_name] = dt_params
4047 self.new_diskparams[dt_name].update(dt_params)
4049 # os hypervisor parameters
4050 self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
4052 for os_name, hvs in self.op.os_hvp.items():
4053 if os_name not in self.new_os_hvp:
4054 self.new_os_hvp[os_name] = hvs
4056 for hv_name, hv_dict in hvs.items():
4057 if hv_name not in self.new_os_hvp[os_name]:
4058 self.new_os_hvp[os_name][hv_name] = hv_dict
4060 self.new_os_hvp[os_name][hv_name].update(hv_dict)
4063 self.new_osp = objects.FillDict(cluster.osparams, {})
4064 if self.op.osparams:
4065 for os_name, osp in self.op.osparams.items():
4066 if os_name not in self.new_osp:
4067 self.new_osp[os_name] = {}
4069 self.new_osp[os_name] = _GetUpdatedParams(self.new_osp[os_name], osp,
4072 if not self.new_osp[os_name]:
4073 # we removed all parameters
4074 del self.new_osp[os_name]
4076 # check the parameter validity (remote check)
4077 _CheckOSParams(self, False, [self.cfg.GetMasterNode()],
4078 os_name, self.new_osp[os_name])
4080 # changes to the hypervisor list
4081 if self.op.enabled_hypervisors is not None:
4082 self.hv_list = self.op.enabled_hypervisors
4083 for hv in self.hv_list:
4084 # if the hypervisor doesn't already exist in the cluster
4085 # hvparams, we initialize it to empty, and then (in both
4086 # cases) we make sure to fill the defaults, as we might not
4087 # have a complete defaults list if the hypervisor wasn't
4089 if hv not in new_hvp:
4091 new_hvp[hv] = objects.FillDict(constants.HVC_DEFAULTS[hv], new_hvp[hv])
4092 utils.ForceDictType(new_hvp[hv], constants.HVS_PARAMETER_TYPES)
4094 self.hv_list = cluster.enabled_hypervisors
4096 if self.op.hvparams or self.op.enabled_hypervisors is not None:
4097 # either the enabled list has changed, or the parameters have, validate
4098 for hv_name, hv_params in self.new_hvparams.items():
4099 if ((self.op.hvparams and hv_name in self.op.hvparams) or
4100 (self.op.enabled_hypervisors and
4101 hv_name in self.op.enabled_hypervisors)):
4102 # either this is a new hypervisor, or its parameters have changed
4103 hv_class = hypervisor.GetHypervisor(hv_name)
4104 utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
4105 hv_class.CheckParameterSyntax(hv_params)
4106 _CheckHVParams(self, node_list, hv_name, hv_params)
4109 # no need to check any newly-enabled hypervisors, since the
4110 # defaults have already been checked in the above code-block
4111 for os_name, os_hvp in self.new_os_hvp.items():
4112 for hv_name, hv_params in os_hvp.items():
4113 utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
4114 # we need to fill in the new os_hvp on top of the actual hv_p
4115 cluster_defaults = self.new_hvparams.get(hv_name, {})
4116 new_osp = objects.FillDict(cluster_defaults, hv_params)
4117 hv_class = hypervisor.GetHypervisor(hv_name)
4118 hv_class.CheckParameterSyntax(new_osp)
4119 _CheckHVParams(self, node_list, hv_name, new_osp)
4121 if self.op.default_iallocator:
4122 alloc_script = utils.FindFile(self.op.default_iallocator,
4123 constants.IALLOCATOR_SEARCH_PATH,
4125 if alloc_script is None:
4126 raise errors.OpPrereqError("Invalid default iallocator script '%s'"
4127 " specified" % self.op.default_iallocator,
4130 def Exec(self, feedback_fn):
4131 """Change the parameters of the cluster.
4134 if self.op.vg_name is not None:
4135 new_volume = self.op.vg_name
4138 if new_volume != self.cfg.GetVGName():
4139 self.cfg.SetVGName(new_volume)
4141 feedback_fn("Cluster LVM configuration already in desired"
4142 " state, not changing")
4143 if self.op.drbd_helper is not None:
4144 new_helper = self.op.drbd_helper
4147 if new_helper != self.cfg.GetDRBDHelper():
4148 self.cfg.SetDRBDHelper(new_helper)
4150 feedback_fn("Cluster DRBD helper already in desired state,"
4152 if self.op.hvparams:
4153 self.cluster.hvparams = self.new_hvparams
4155 self.cluster.os_hvp = self.new_os_hvp
4156 if self.op.enabled_hypervisors is not None:
4157 self.cluster.hvparams = self.new_hvparams
4158 self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
4159 if self.op.beparams:
4160 self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
4161 if self.op.nicparams:
4162 self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
4164 self.cluster.ipolicy = self.new_ipolicy
4165 if self.op.osparams:
4166 self.cluster.osparams = self.new_osp
4167 if self.op.ndparams:
4168 self.cluster.ndparams = self.new_ndparams
4169 if self.op.diskparams:
4170 self.cluster.diskparams = self.new_diskparams
4171 if self.op.hv_state:
4172 self.cluster.hv_state_static = self.new_hv_state
4173 if self.op.disk_state:
4174 self.cluster.disk_state_static = self.new_disk_state
4176 if self.op.candidate_pool_size is not None:
4177 self.cluster.candidate_pool_size = self.op.candidate_pool_size
4178 # we need to update the pool size here, otherwise the save will fail
4179 _AdjustCandidatePool(self, [])
4181 if self.op.maintain_node_health is not None:
4182 if self.op.maintain_node_health and not constants.ENABLE_CONFD:
4183 feedback_fn("Note: CONFD was disabled at build time, node health"
4184 " maintenance is not useful (still enabling it)")
4185 self.cluster.maintain_node_health = self.op.maintain_node_health
4187 if self.op.prealloc_wipe_disks is not None:
4188 self.cluster.prealloc_wipe_disks = self.op.prealloc_wipe_disks
4190 if self.op.add_uids is not None:
4191 uidpool.AddToUidPool(self.cluster.uid_pool, self.op.add_uids)
4193 if self.op.remove_uids is not None:
4194 uidpool.RemoveFromUidPool(self.cluster.uid_pool, self.op.remove_uids)
4196 if self.op.uid_pool is not None:
4197 self.cluster.uid_pool = self.op.uid_pool
4199 if self.op.default_iallocator is not None:
4200 self.cluster.default_iallocator = self.op.default_iallocator
4202 if self.op.reserved_lvs is not None:
4203 self.cluster.reserved_lvs = self.op.reserved_lvs
4205 if self.op.use_external_mip_script is not None:
4206 self.cluster.use_external_mip_script = self.op.use_external_mip_script
4208 def helper_os(aname, mods, desc):
4210 lst = getattr(self.cluster, aname)
4211 for key, val in mods:
4212 if key == constants.DDM_ADD:
4214 feedback_fn("OS %s already in %s, ignoring" % (val, desc))
4217 elif key == constants.DDM_REMOVE:
4221 feedback_fn("OS %s not found in %s, ignoring" % (val, desc))
4223 raise errors.ProgrammerError("Invalid modification '%s'" % key)
4225 if self.op.hidden_os:
4226 helper_os("hidden_os", self.op.hidden_os, "hidden")
4228 if self.op.blacklisted_os:
4229 helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
4231 if self.op.master_netdev:
4232 master_params = self.cfg.GetMasterNetworkParameters()
4233 ems = self.cfg.GetUseExternalMipScript()
4234 feedback_fn("Shutting down master ip on the current netdev (%s)" %
4235 self.cluster.master_netdev)
4236 result = self.rpc.call_node_deactivate_master_ip(master_params.name,
4238 result.Raise("Could not disable the master ip")
4239 feedback_fn("Changing master_netdev from %s to %s" %
4240 (master_params.netdev, self.op.master_netdev))
4241 self.cluster.master_netdev = self.op.master_netdev
4243 if self.op.master_netmask:
4244 master_params = self.cfg.GetMasterNetworkParameters()
4245 feedback_fn("Changing master IP netmask to %s" % self.op.master_netmask)
4246 result = self.rpc.call_node_change_master_netmask(master_params.name,
4247 master_params.netmask,
4248 self.op.master_netmask,
4250 master_params.netdev)
4252 msg = "Could not change the master IP netmask: %s" % result.fail_msg
4255 self.cluster.master_netmask = self.op.master_netmask
4257 self.cfg.Update(self.cluster, feedback_fn)
4259 if self.op.master_netdev:
4260 master_params = self.cfg.GetMasterNetworkParameters()
4261 feedback_fn("Starting the master ip on the new master netdev (%s)" %
4262 self.op.master_netdev)
4263 ems = self.cfg.GetUseExternalMipScript()
4264 result = self.rpc.call_node_activate_master_ip(master_params.name,
4267 self.LogWarning("Could not re-enable the master ip on"
4268 " the master, please restart manually: %s",
4272 def _UploadHelper(lu, nodes, fname):
4273 """Helper for uploading a file and showing warnings.
4276 if os.path.exists(fname):
4277 result = lu.rpc.call_upload_file(nodes, fname)
4278 for to_node, to_result in result.items():
4279 msg = to_result.fail_msg
4281 msg = ("Copy of file %s to node %s failed: %s" %
4282 (fname, to_node, msg))
4283 lu.proc.LogWarning(msg)
4286 def _ComputeAncillaryFiles(cluster, redist):
4287 """Compute files external to Ganeti which need to be consistent.
4289 @type redist: boolean
4290 @param redist: Whether to include files which need to be redistributed
4293 # Compute files for all nodes
4295 constants.SSH_KNOWN_HOSTS_FILE,
4296 constants.CONFD_HMAC_KEY,
4297 constants.CLUSTER_DOMAIN_SECRET_FILE,
4298 constants.SPICE_CERT_FILE,
4299 constants.SPICE_CACERT_FILE,
4300 constants.RAPI_USERS_FILE,
4304 files_all.update(constants.ALL_CERT_FILES)
4305 files_all.update(ssconf.SimpleStore().GetFileList())
4307 # we need to ship at least the RAPI certificate
4308 files_all.add(constants.RAPI_CERT_FILE)
4310 if cluster.modify_etc_hosts:
4311 files_all.add(constants.ETC_HOSTS)
4313 if cluster.use_external_mip_script:
4314 files_all.add(constants.EXTERNAL_MASTER_SETUP_SCRIPT)
4316 # Files which are optional, these must:
4317 # - be present in one other category as well
4318 # - either exist or not exist on all nodes of that category (mc, vm all)
4320 constants.RAPI_USERS_FILE,
4323 # Files which should only be on master candidates
4327 files_mc.add(constants.CLUSTER_CONF_FILE)
4329 # Files which should only be on VM-capable nodes
4330 files_vm = set(filename
4331 for hv_name in cluster.enabled_hypervisors
4332 for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[0])
4334 files_opt |= set(filename
4335 for hv_name in cluster.enabled_hypervisors
4336 for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[1])
4338 # Filenames in each category must be unique
4339 all_files_set = files_all | files_mc | files_vm
4340 assert (len(all_files_set) ==
4341 sum(map(len, [files_all, files_mc, files_vm]))), \
4342 "Found file listed in more than one file list"
4344 # Optional files must be present in one other category
4345 assert all_files_set.issuperset(files_opt), \
4346 "Optional file not in a different required list"
4348 return (files_all, files_opt, files_mc, files_vm)
4351 def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
4352 """Distribute additional files which are part of the cluster configuration.
4354 ConfigWriter takes care of distributing the config and ssconf files, but
4355 there are more files which should be distributed to all nodes. This function
4356 makes sure those are copied.
4358 @param lu: calling logical unit
4359 @param additional_nodes: list of nodes not in the config to distribute to
4360 @type additional_vm: boolean
4361 @param additional_vm: whether the additional nodes are vm-capable or not
4364 # Gather target nodes
4365 cluster = lu.cfg.GetClusterInfo()
4366 master_info = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
4368 online_nodes = lu.cfg.GetOnlineNodeList()
4369 vm_nodes = lu.cfg.GetVmCapableNodeList()
4371 if additional_nodes is not None:
4372 online_nodes.extend(additional_nodes)
4374 vm_nodes.extend(additional_nodes)
4376 # Never distribute to master node
4377 for nodelist in [online_nodes, vm_nodes]:
4378 if master_info.name in nodelist:
4379 nodelist.remove(master_info.name)
4382 (files_all, _, files_mc, files_vm) = \
4383 _ComputeAncillaryFiles(cluster, True)
4385 # Never re-distribute configuration file from here
4386 assert not (constants.CLUSTER_CONF_FILE in files_all or
4387 constants.CLUSTER_CONF_FILE in files_vm)
4388 assert not files_mc, "Master candidates not handled in this function"
4391 (online_nodes, files_all),
4392 (vm_nodes, files_vm),
4396 for (node_list, files) in filemap:
4398 _UploadHelper(lu, node_list, fname)
4401 class LUClusterRedistConf(NoHooksLU):
4402 """Force the redistribution of cluster configuration.
4404 This is a very simple LU.
4409 def ExpandNames(self):
4410 self.needed_locks = {
4411 locking.LEVEL_NODE: locking.ALL_SET,
4413 self.share_locks[locking.LEVEL_NODE] = 1
4415 def Exec(self, feedback_fn):
4416 """Redistribute the configuration.
4419 self.cfg.Update(self.cfg.GetClusterInfo(), feedback_fn)
4420 _RedistributeAncillaryFiles(self)
4423 class LUClusterActivateMasterIp(NoHooksLU):
4424 """Activate the master IP on the master node.
4427 def Exec(self, feedback_fn):
4428 """Activate the master IP.
4431 master_params = self.cfg.GetMasterNetworkParameters()
4432 ems = self.cfg.GetUseExternalMipScript()
4433 result = self.rpc.call_node_activate_master_ip(master_params.name,
4435 result.Raise("Could not activate the master IP")
4438 class LUClusterDeactivateMasterIp(NoHooksLU):
4439 """Deactivate the master IP on the master node.
4442 def Exec(self, feedback_fn):
4443 """Deactivate the master IP.
4446 master_params = self.cfg.GetMasterNetworkParameters()
4447 ems = self.cfg.GetUseExternalMipScript()
4448 result = self.rpc.call_node_deactivate_master_ip(master_params.name,
4450 result.Raise("Could not deactivate the master IP")
4453 def _WaitForSync(lu, instance, disks=None, oneshot=False):
4454 """Sleep and poll for an instance's disk to sync.
4457 if not instance.disks or disks is not None and not disks:
4460 disks = _ExpandCheckDisks(instance, disks)
4463 lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
4465 node = instance.primary_node
4468 lu.cfg.SetDiskID(dev, node)
4470 # TODO: Convert to utils.Retry
4473 degr_retries = 10 # in seconds, as we sleep 1 second each time
4477 cumul_degraded = False
4478 rstats = lu.rpc.call_blockdev_getmirrorstatus(node, (disks, instance))
4479 msg = rstats.fail_msg
4481 lu.LogWarning("Can't get any data from node %s: %s", node, msg)
4484 raise errors.RemoteError("Can't contact node %s for mirror data,"
4485 " aborting." % node)
4488 rstats = rstats.payload
4490 for i, mstat in enumerate(rstats):
4492 lu.LogWarning("Can't compute data for node %s/%s",
4493 node, disks[i].iv_name)
4496 cumul_degraded = (cumul_degraded or
4497 (mstat.is_degraded and mstat.sync_percent is None))
4498 if mstat.sync_percent is not None:
4500 if mstat.estimated_time is not None:
4501 rem_time = ("%s remaining (estimated)" %
4502 utils.FormatSeconds(mstat.estimated_time))
4503 max_time = mstat.estimated_time
4505 rem_time = "no time estimate"
4506 lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
4507 (disks[i].iv_name, mstat.sync_percent, rem_time))
4509 # if we're done but degraded, let's do a few small retries, to
4510 # make sure we see a stable and not transient situation; therefore
4511 # we force restart of the loop
4512 if (done or oneshot) and cumul_degraded and degr_retries > 0:
4513 logging.info("Degraded disks found, %d retries left", degr_retries)
4521 time.sleep(min(60, max_time))
4524 lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
4525 return not cumul_degraded
4528 def _BlockdevFind(lu, node, dev, instance):
4529 """Wrapper around call_blockdev_find to annotate diskparams.
4531 @param lu: A reference to the lu object
4532 @param node: The node to call out
4533 @param dev: The device to find
4534 @param instance: The instance object the device belongs to
4535 @returns The result of the rpc call
4538 (disk,) = _AnnotateDiskParams(instance, [dev], lu.cfg)
4539 return lu.rpc.call_blockdev_find(node, disk)
4542 def _CheckDiskConsistency(lu, instance, dev, node, on_primary, ldisk=False):
4543 """Wrapper around L{_CheckDiskConsistencyInner}.
4546 (disk,) = _AnnotateDiskParams(instance, [dev], lu.cfg)
4547 return _CheckDiskConsistencyInner(lu, instance, disk, node, on_primary,
4551 def _CheckDiskConsistencyInner(lu, instance, dev, node, on_primary,
4553 """Check that mirrors are not degraded.
4555 @attention: The device has to be annotated already.
4557 The ldisk parameter, if True, will change the test from the
4558 is_degraded attribute (which represents overall non-ok status for
4559 the device(s)) to the ldisk (representing the local storage status).
4562 lu.cfg.SetDiskID(dev, node)
4566 if on_primary or dev.AssembleOnSecondary():
4567 rstats = lu.rpc.call_blockdev_find(node, dev)
4568 msg = rstats.fail_msg
4570 lu.LogWarning("Can't find disk on node %s: %s", node, msg)
4572 elif not rstats.payload:
4573 lu.LogWarning("Can't find disk on node %s", node)
4577 result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
4579 result = result and not rstats.payload.is_degraded
4582 for child in dev.children:
4583 result = result and _CheckDiskConsistencyInner(lu, instance, child, node,
4589 class LUOobCommand(NoHooksLU):
4590 """Logical unit for OOB handling.
4594 _SKIP_MASTER = (constants.OOB_POWER_OFF, constants.OOB_POWER_CYCLE)
4596 def ExpandNames(self):
4597 """Gather locks we need.
4600 if self.op.node_names:
4601 self.op.node_names = _GetWantedNodes(self, self.op.node_names)
4602 lock_names = self.op.node_names
4604 lock_names = locking.ALL_SET
4606 self.needed_locks = {
4607 locking.LEVEL_NODE: lock_names,
4610 def CheckPrereq(self):
4611 """Check prerequisites.
4614 - the node exists in the configuration
4617 Any errors are signaled by raising errors.OpPrereqError.
4621 self.master_node = self.cfg.GetMasterNode()
4623 assert self.op.power_delay >= 0.0
4625 if self.op.node_names:
4626 if (self.op.command in self._SKIP_MASTER and
4627 self.master_node in self.op.node_names):
4628 master_node_obj = self.cfg.GetNodeInfo(self.master_node)
4629 master_oob_handler = _SupportsOob(self.cfg, master_node_obj)
4631 if master_oob_handler:
4632 additional_text = ("run '%s %s %s' if you want to operate on the"
4633 " master regardless") % (master_oob_handler,
4637 additional_text = "it does not support out-of-band operations"
4639 raise errors.OpPrereqError(("Operating on the master node %s is not"
4640 " allowed for %s; %s") %
4641 (self.master_node, self.op.command,
4642 additional_text), errors.ECODE_INVAL)
4644 self.op.node_names = self.cfg.GetNodeList()
4645 if self.op.command in self._SKIP_MASTER:
4646 self.op.node_names.remove(self.master_node)
4648 if self.op.command in self._SKIP_MASTER:
4649 assert self.master_node not in self.op.node_names
4651 for (node_name, node) in self.cfg.GetMultiNodeInfo(self.op.node_names):
4653 raise errors.OpPrereqError("Node %s not found" % node_name,
4656 self.nodes.append(node)
4658 if (not self.op.ignore_status and
4659 (self.op.command == constants.OOB_POWER_OFF and not node.offline)):
4660 raise errors.OpPrereqError(("Cannot power off node %s because it is"
4661 " not marked offline") % node_name,
4664 def Exec(self, feedback_fn):
4665 """Execute OOB and return result if we expect any.
4668 master_node = self.master_node
4671 for idx, node in enumerate(utils.NiceSort(self.nodes,
4672 key=lambda node: node.name)):
4673 node_entry = [(constants.RS_NORMAL, node.name)]
4674 ret.append(node_entry)
4676 oob_program = _SupportsOob(self.cfg, node)
4679 node_entry.append((constants.RS_UNAVAIL, None))
4682 logging.info("Executing out-of-band command '%s' using '%s' on %s",
4683 self.op.command, oob_program, node.name)
4684 result = self.rpc.call_run_oob(master_node, oob_program,
4685 self.op.command, node.name,
4689 self.LogWarning("Out-of-band RPC failed on node '%s': %s",
4690 node.name, result.fail_msg)
4691 node_entry.append((constants.RS_NODATA, None))
4694 self._CheckPayload(result)
4695 except errors.OpExecError, err:
4696 self.LogWarning("Payload returned by node '%s' is not valid: %s",
4698 node_entry.append((constants.RS_NODATA, None))
4700 if self.op.command == constants.OOB_HEALTH:
4701 # For health we should log important events
4702 for item, status in result.payload:
4703 if status in [constants.OOB_STATUS_WARNING,
4704 constants.OOB_STATUS_CRITICAL]:
4705 self.LogWarning("Item '%s' on node '%s' has status '%s'",
4706 item, node.name, status)
4708 if self.op.command == constants.OOB_POWER_ON:
4710 elif self.op.command == constants.OOB_POWER_OFF:
4711 node.powered = False
4712 elif self.op.command == constants.OOB_POWER_STATUS:
4713 powered = result.payload[constants.OOB_POWER_STATUS_POWERED]
4714 if powered != node.powered:
4715 logging.warning(("Recorded power state (%s) of node '%s' does not"
4716 " match actual power state (%s)"), node.powered,
4719 # For configuration changing commands we should update the node
4720 if self.op.command in (constants.OOB_POWER_ON,
4721 constants.OOB_POWER_OFF):
4722 self.cfg.Update(node, feedback_fn)
4724 node_entry.append((constants.RS_NORMAL, result.payload))
4726 if (self.op.command == constants.OOB_POWER_ON and
4727 idx < len(self.nodes) - 1):
4728 time.sleep(self.op.power_delay)
4732 def _CheckPayload(self, result):
4733 """Checks if the payload is valid.
4735 @param result: RPC result
4736 @raises errors.OpExecError: If payload is not valid
4740 if self.op.command == constants.OOB_HEALTH:
4741 if not isinstance(result.payload, list):
4742 errs.append("command 'health' is expected to return a list but got %s" %
4743 type(result.payload))
4745 for item, status in result.payload:
4746 if status not in constants.OOB_STATUSES:
4747 errs.append("health item '%s' has invalid status '%s'" %
4750 if self.op.command == constants.OOB_POWER_STATUS:
4751 if not isinstance(result.payload, dict):
4752 errs.append("power-status is expected to return a dict but got %s" %
4753 type(result.payload))
4755 if self.op.command in [
4756 constants.OOB_POWER_ON,
4757 constants.OOB_POWER_OFF,
4758 constants.OOB_POWER_CYCLE,
4760 if result.payload is not None:
4761 errs.append("%s is expected to not return payload but got '%s'" %
4762 (self.op.command, result.payload))
4765 raise errors.OpExecError("Check of out-of-band payload failed due to %s" %
4766 utils.CommaJoin(errs))
4769 class _OsQuery(_QueryBase):
4770 FIELDS = query.OS_FIELDS
4772 def ExpandNames(self, lu):
4773 # Lock all nodes in shared mode
4774 # Temporary removal of locks, should be reverted later
4775 # TODO: reintroduce locks when they are lighter-weight
4776 lu.needed_locks = {}
4777 #self.share_locks[locking.LEVEL_NODE] = 1
4778 #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4780 # The following variables interact with _QueryBase._GetNames
4782 self.wanted = self.names
4784 self.wanted = locking.ALL_SET
4786 self.do_locking = self.use_locking
4788 def DeclareLocks(self, lu, level):
4792 def _DiagnoseByOS(rlist):
4793 """Remaps a per-node return list into an a per-os per-node dictionary
4795 @param rlist: a map with node names as keys and OS objects as values
4798 @return: a dictionary with osnames as keys and as value another
4799 map, with nodes as keys and tuples of (path, status, diagnose,
4800 variants, parameters, api_versions) as values, eg::
4802 {"debian-etch": {"node1": [(/usr/lib/..., True, "", [], []),
4803 (/srv/..., False, "invalid api")],
4804 "node2": [(/srv/..., True, "", [], [])]}
4809 # we build here the list of nodes that didn't fail the RPC (at RPC
4810 # level), so that nodes with a non-responding node daemon don't
4811 # make all OSes invalid
4812 good_nodes = [node_name for node_name in rlist
4813 if not rlist[node_name].fail_msg]
4814 for node_name, nr in rlist.items():
4815 if nr.fail_msg or not nr.payload:
4817 for (name, path, status, diagnose, variants,
4818 params, api_versions) in nr.payload:
4819 if name not in all_os:
4820 # build a list of nodes for this os containing empty lists
4821 # for each node in node_list
4823 for nname in good_nodes:
4824 all_os[name][nname] = []
4825 # convert params from [name, help] to (name, help)
4826 params = [tuple(v) for v in params]
4827 all_os[name][node_name].append((path, status, diagnose,
4828 variants, params, api_versions))
4831 def _GetQueryData(self, lu):
4832 """Computes the list of nodes and their attributes.
4835 # Locking is not used
4836 assert not (compat.any(lu.glm.is_owned(level)
4837 for level in locking.LEVELS
4838 if level != locking.LEVEL_CLUSTER) or
4839 self.do_locking or self.use_locking)
4841 valid_nodes = [node.name
4842 for node in lu.cfg.GetAllNodesInfo().values()
4843 if not node.offline and node.vm_capable]
4844 pol = self._DiagnoseByOS(lu.rpc.call_os_diagnose(valid_nodes))
4845 cluster = lu.cfg.GetClusterInfo()
4849 for (os_name, os_data) in pol.items():
4850 info = query.OsInfo(name=os_name, valid=True, node_status=os_data,
4851 hidden=(os_name in cluster.hidden_os),
4852 blacklisted=(os_name in cluster.blacklisted_os))
4856 api_versions = set()
4858 for idx, osl in enumerate(os_data.values()):
4859 info.valid = bool(info.valid and osl and osl[0][1])
4863 (node_variants, node_params, node_api) = osl[0][3:6]
4866 variants.update(node_variants)
4867 parameters.update(node_params)
4868 api_versions.update(node_api)
4870 # Filter out inconsistent values
4871 variants.intersection_update(node_variants)
4872 parameters.intersection_update(node_params)
4873 api_versions.intersection_update(node_api)
4875 info.variants = list(variants)
4876 info.parameters = list(parameters)
4877 info.api_versions = list(api_versions)
4879 data[os_name] = info
4881 # Prepare data in requested order
4882 return [data[name] for name in self._GetNames(lu, pol.keys(), None)
4886 class LUOsDiagnose(NoHooksLU):
4887 """Logical unit for OS diagnose/query.
4893 def _BuildFilter(fields, names):
4894 """Builds a filter for querying OSes.
4897 name_filter = qlang.MakeSimpleFilter("name", names)
4899 # Legacy behaviour: Hide hidden, blacklisted or invalid OSes if the
4900 # respective field is not requested
4901 status_filter = [[qlang.OP_NOT, [qlang.OP_TRUE, fname]]
4902 for fname in ["hidden", "blacklisted"]
4903 if fname not in fields]
4904 if "valid" not in fields:
4905 status_filter.append([qlang.OP_TRUE, "valid"])
4908 status_filter.insert(0, qlang.OP_AND)
4910 status_filter = None
4912 if name_filter and status_filter:
4913 return [qlang.OP_AND, name_filter, status_filter]
4917 return status_filter
4919 def CheckArguments(self):
4920 self.oq = _OsQuery(self._BuildFilter(self.op.output_fields, self.op.names),
4921 self.op.output_fields, False)
4923 def ExpandNames(self):
4924 self.oq.ExpandNames(self)
4926 def Exec(self, feedback_fn):
4927 return self.oq.OldStyleQuery(self)
4930 class LUNodeRemove(LogicalUnit):
4931 """Logical unit for removing a node.
4934 HPATH = "node-remove"
4935 HTYPE = constants.HTYPE_NODE
4937 def BuildHooksEnv(self):
4942 "OP_TARGET": self.op.node_name,
4943 "NODE_NAME": self.op.node_name,
4946 def BuildHooksNodes(self):
4947 """Build hooks nodes.
4949 This doesn't run on the target node in the pre phase as a failed
4950 node would then be impossible to remove.
4953 all_nodes = self.cfg.GetNodeList()
4955 all_nodes.remove(self.op.node_name)
4958 return (all_nodes, all_nodes)
4960 def CheckPrereq(self):
4961 """Check prerequisites.
4964 - the node exists in the configuration
4965 - it does not have primary or secondary instances
4966 - it's not the master
4968 Any errors are signaled by raising errors.OpPrereqError.
4971 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4972 node = self.cfg.GetNodeInfo(self.op.node_name)
4973 assert node is not None
4975 masternode = self.cfg.GetMasterNode()
4976 if node.name == masternode:
4977 raise errors.OpPrereqError("Node is the master node, failover to another"
4978 " node is required", errors.ECODE_INVAL)
4980 for instance_name, instance in self.cfg.GetAllInstancesInfo().items():
4981 if node.name in instance.all_nodes:
4982 raise errors.OpPrereqError("Instance %s is still running on the node,"
4983 " please remove first" % instance_name,
4985 self.op.node_name = node.name
4988 def Exec(self, feedback_fn):
4989 """Removes the node from the cluster.
4993 logging.info("Stopping the node daemon and removing configs from node %s",
4996 modify_ssh_setup = self.cfg.GetClusterInfo().modify_ssh_setup
4998 assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER), \
5001 # Promote nodes to master candidate as needed
5002 _AdjustCandidatePool(self, exceptions=[node.name])
5003 self.context.RemoveNode(node.name)
5005 # Run post hooks on the node before it's removed
5006 _RunPostHook(self, node.name)
5008 result = self.rpc.call_node_leave_cluster(node.name, modify_ssh_setup)
5009 msg = result.fail_msg
5011 self.LogWarning("Errors encountered on the remote node while leaving"
5012 " the cluster: %s", msg)
5014 # Remove node from our /etc/hosts
5015 if self.cfg.GetClusterInfo().modify_etc_hosts:
5016 master_node = self.cfg.GetMasterNode()
5017 result = self.rpc.call_etc_hosts_modify(master_node,
5018 constants.ETC_HOSTS_REMOVE,
5020 result.Raise("Can't update hosts file with new host data")
5021 _RedistributeAncillaryFiles(self)
5024 class _NodeQuery(_QueryBase):
5025 FIELDS = query.NODE_FIELDS
5027 def ExpandNames(self, lu):
5028 lu.needed_locks = {}
5029 lu.share_locks = _ShareAll()
5032 self.wanted = _GetWantedNodes(lu, self.names)
5034 self.wanted = locking.ALL_SET
5036 self.do_locking = (self.use_locking and
5037 query.NQ_LIVE in self.requested_data)
5040 # If any non-static field is requested we need to lock the nodes
5041 lu.needed_locks[locking.LEVEL_NODE] = self.wanted
5043 def DeclareLocks(self, lu, level):
5046 def _GetQueryData(self, lu):
5047 """Computes the list of nodes and their attributes.
5050 all_info = lu.cfg.GetAllNodesInfo()
5052 nodenames = self._GetNames(lu, all_info.keys(), locking.LEVEL_NODE)
5054 # Gather data as requested
5055 if query.NQ_LIVE in self.requested_data:
5056 # filter out non-vm_capable nodes
5057 toquery_nodes = [name for name in nodenames if all_info[name].vm_capable]
5059 node_data = lu.rpc.call_node_info(toquery_nodes, [lu.cfg.GetVGName()],
5060 [lu.cfg.GetHypervisorType()])
5061 live_data = dict((name, _MakeLegacyNodeInfo(nresult.payload))
5062 for (name, nresult) in node_data.items()
5063 if not nresult.fail_msg and nresult.payload)
5067 if query.NQ_INST in self.requested_data:
5068 node_to_primary = dict([(name, set()) for name in nodenames])
5069 node_to_secondary = dict([(name, set()) for name in nodenames])
5071 inst_data = lu.cfg.GetAllInstancesInfo()
5073 for inst in inst_data.values():
5074 if inst.primary_node in node_to_primary:
5075 node_to_primary[inst.primary_node].add(inst.name)
5076 for secnode in inst.secondary_nodes:
5077 if secnode in node_to_secondary:
5078 node_to_secondary[secnode].add(inst.name)
5080 node_to_primary = None
5081 node_to_secondary = None
5083 if query.NQ_OOB in self.requested_data:
5084 oob_support = dict((name, bool(_SupportsOob(lu.cfg, node)))
5085 for name, node in all_info.iteritems())
5089 if query.NQ_GROUP in self.requested_data:
5090 groups = lu.cfg.GetAllNodeGroupsInfo()
5094 return query.NodeQueryData([all_info[name] for name in nodenames],
5095 live_data, lu.cfg.GetMasterNode(),
5096 node_to_primary, node_to_secondary, groups,
5097 oob_support, lu.cfg.GetClusterInfo())
5100 class LUNodeQuery(NoHooksLU):
5101 """Logical unit for querying nodes.
5104 # pylint: disable=W0142
5107 def CheckArguments(self):
5108 self.nq = _NodeQuery(qlang.MakeSimpleFilter("name", self.op.names),
5109 self.op.output_fields, self.op.use_locking)
5111 def ExpandNames(self):
5112 self.nq.ExpandNames(self)
5114 def DeclareLocks(self, level):
5115 self.nq.DeclareLocks(self, level)
5117 def Exec(self, feedback_fn):
5118 return self.nq.OldStyleQuery(self)
5121 class LUNodeQueryvols(NoHooksLU):
5122 """Logical unit for getting volumes on node(s).
5126 _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
5127 _FIELDS_STATIC = utils.FieldSet("node")
5129 def CheckArguments(self):
5130 _CheckOutputFields(static=self._FIELDS_STATIC,
5131 dynamic=self._FIELDS_DYNAMIC,
5132 selected=self.op.output_fields)
5134 def ExpandNames(self):
5135 self.share_locks = _ShareAll()
5136 self.needed_locks = {}
5138 if not self.op.nodes:
5139 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
5141 self.needed_locks[locking.LEVEL_NODE] = \
5142 _GetWantedNodes(self, self.op.nodes)
5144 def Exec(self, feedback_fn):
5145 """Computes the list of nodes and their attributes.
5148 nodenames = self.owned_locks(locking.LEVEL_NODE)
5149 volumes = self.rpc.call_node_volumes(nodenames)
5151 ilist = self.cfg.GetAllInstancesInfo()
5152 vol2inst = _MapInstanceDisksToNodes(ilist.values())
5155 for node in nodenames:
5156 nresult = volumes[node]
5159 msg = nresult.fail_msg
5161 self.LogWarning("Can't compute volume data on node %s: %s", node, msg)
5164 node_vols = sorted(nresult.payload,
5165 key=operator.itemgetter("dev"))
5167 for vol in node_vols:
5169 for field in self.op.output_fields:
5172 elif field == "phys":
5176 elif field == "name":
5178 elif field == "size":
5179 val = int(float(vol["size"]))
5180 elif field == "instance":
5181 val = vol2inst.get((node, vol["vg"] + "/" + vol["name"]), "-")
5183 raise errors.ParameterError(field)
5184 node_output.append(str(val))
5186 output.append(node_output)
5191 class LUNodeQueryStorage(NoHooksLU):
5192 """Logical unit for getting information on storage units on node(s).
5195 _FIELDS_STATIC = utils.FieldSet(constants.SF_NODE)
5198 def CheckArguments(self):
5199 _CheckOutputFields(static=self._FIELDS_STATIC,
5200 dynamic=utils.FieldSet(*constants.VALID_STORAGE_FIELDS),
5201 selected=self.op.output_fields)
5203 def ExpandNames(self):
5204 self.share_locks = _ShareAll()
5205 self.needed_locks = {}
5208 self.needed_locks[locking.LEVEL_NODE] = \
5209 _GetWantedNodes(self, self.op.nodes)
5211 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
5213 def Exec(self, feedback_fn):
5214 """Computes the list of nodes and their attributes.
5217 self.nodes = self.owned_locks(locking.LEVEL_NODE)
5219 # Always get name to sort by
5220 if constants.SF_NAME in self.op.output_fields:
5221 fields = self.op.output_fields[:]
5223 fields = [constants.SF_NAME] + self.op.output_fields
5225 # Never ask for node or type as it's only known to the LU
5226 for extra in [constants.SF_NODE, constants.SF_TYPE]:
5227 while extra in fields:
5228 fields.remove(extra)
5230 field_idx = dict([(name, idx) for (idx, name) in enumerate(fields)])
5231 name_idx = field_idx[constants.SF_NAME]
5233 st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
5234 data = self.rpc.call_storage_list(self.nodes,
5235 self.op.storage_type, st_args,
5236 self.op.name, fields)
5240 for node in utils.NiceSort(self.nodes):
5241 nresult = data[node]
5245 msg = nresult.fail_msg
5247 self.LogWarning("Can't get storage data from node %s: %s", node, msg)
5250 rows = dict([(row[name_idx], row) for row in nresult.payload])
5252 for name in utils.NiceSort(rows.keys()):
5257 for field in self.op.output_fields:
5258 if field == constants.SF_NODE:
5260 elif field == constants.SF_TYPE:
5261 val = self.op.storage_type
5262 elif field in field_idx:
5263 val = row[field_idx[field]]
5265 raise errors.ParameterError(field)
5274 class _InstanceQuery(_QueryBase):
5275 FIELDS = query.INSTANCE_FIELDS
5277 def ExpandNames(self, lu):
5278 lu.needed_locks = {}
5279 lu.share_locks = _ShareAll()
5282 self.wanted = _GetWantedInstances(lu, self.names)
5284 self.wanted = locking.ALL_SET
5286 self.do_locking = (self.use_locking and
5287 query.IQ_LIVE in self.requested_data)
5289 lu.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
5290 lu.needed_locks[locking.LEVEL_NODEGROUP] = []
5291 lu.needed_locks[locking.LEVEL_NODE] = []
5292 lu.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5294 self.do_grouplocks = (self.do_locking and
5295 query.IQ_NODES in self.requested_data)
5297 def DeclareLocks(self, lu, level):
5299 if level == locking.LEVEL_NODEGROUP and self.do_grouplocks:
5300 assert not lu.needed_locks[locking.LEVEL_NODEGROUP]
5302 # Lock all groups used by instances optimistically; this requires going
5303 # via the node before it's locked, requiring verification later on
5304 lu.needed_locks[locking.LEVEL_NODEGROUP] = \
5306 for instance_name in lu.owned_locks(locking.LEVEL_INSTANCE)
5307 for group_uuid in lu.cfg.GetInstanceNodeGroups(instance_name))
5308 elif level == locking.LEVEL_NODE:
5309 lu._LockInstancesNodes() # pylint: disable=W0212
5312 def _CheckGroupLocks(lu):
5313 owned_instances = frozenset(lu.owned_locks(locking.LEVEL_INSTANCE))
5314 owned_groups = frozenset(lu.owned_locks(locking.LEVEL_NODEGROUP))
5316 # Check if node groups for locked instances are still correct
5317 for instance_name in owned_instances:
5318 _CheckInstanceNodeGroups(lu.cfg, instance_name, owned_groups)
5320 def _GetQueryData(self, lu):
5321 """Computes the list of instances and their attributes.
5324 if self.do_grouplocks:
5325 self._CheckGroupLocks(lu)
5327 cluster = lu.cfg.GetClusterInfo()
5328 all_info = lu.cfg.GetAllInstancesInfo()
5330 instance_names = self._GetNames(lu, all_info.keys(), locking.LEVEL_INSTANCE)
5332 instance_list = [all_info[name] for name in instance_names]
5333 nodes = frozenset(itertools.chain(*(inst.all_nodes
5334 for inst in instance_list)))
5335 hv_list = list(set([inst.hypervisor for inst in instance_list]))
5338 wrongnode_inst = set()
5340 # Gather data as requested
5341 if self.requested_data & set([query.IQ_LIVE, query.IQ_CONSOLE]):
5343 node_data = lu.rpc.call_all_instances_info(nodes, hv_list)
5345 result = node_data[name]
5347 # offline nodes will be in both lists
5348 assert result.fail_msg
5349 offline_nodes.append(name)
5351 bad_nodes.append(name)
5352 elif result.payload:
5353 for inst in result.payload:
5354 if inst in all_info:
5355 if all_info[inst].primary_node == name:
5356 live_data.update(result.payload)
5358 wrongnode_inst.add(inst)
5360 # orphan instance; we don't list it here as we don't
5361 # handle this case yet in the output of instance listing
5362 logging.warning("Orphan instance '%s' found on node %s",
5364 # else no instance is alive
5368 if query.IQ_DISKUSAGE in self.requested_data:
5369 disk_usage = dict((inst.name,
5370 _ComputeDiskSize(inst.disk_template,
5371 [{constants.IDISK_SIZE: disk.size}
5372 for disk in inst.disks]))
5373 for inst in instance_list)
5377 if query.IQ_CONSOLE in self.requested_data:
5379 for inst in instance_list:
5380 if inst.name in live_data:
5381 # Instance is running
5382 consinfo[inst.name] = _GetInstanceConsole(cluster, inst)
5384 consinfo[inst.name] = None
5385 assert set(consinfo.keys()) == set(instance_names)
5389 if query.IQ_NODES in self.requested_data:
5390 node_names = set(itertools.chain(*map(operator.attrgetter("all_nodes"),
5392 nodes = dict(lu.cfg.GetMultiNodeInfo(node_names))
5393 groups = dict((uuid, lu.cfg.GetNodeGroup(uuid))
5394 for uuid in set(map(operator.attrgetter("group"),
5400 return query.InstanceQueryData(instance_list, lu.cfg.GetClusterInfo(),
5401 disk_usage, offline_nodes, bad_nodes,
5402 live_data, wrongnode_inst, consinfo,
5406 class LUQuery(NoHooksLU):
5407 """Query for resources/items of a certain kind.
5410 # pylint: disable=W0142
5413 def CheckArguments(self):
5414 qcls = _GetQueryImplementation(self.op.what)
5416 self.impl = qcls(self.op.qfilter, self.op.fields, self.op.use_locking)
5418 def ExpandNames(self):
5419 self.impl.ExpandNames(self)
5421 def DeclareLocks(self, level):
5422 self.impl.DeclareLocks(self, level)
5424 def Exec(self, feedback_fn):
5425 return self.impl.NewStyleQuery(self)
5428 class LUQueryFields(NoHooksLU):
5429 """Query for resources/items of a certain kind.
5432 # pylint: disable=W0142
5435 def CheckArguments(self):
5436 self.qcls = _GetQueryImplementation(self.op.what)
5438 def ExpandNames(self):
5439 self.needed_locks = {}
5441 def Exec(self, feedback_fn):
5442 return query.QueryFields(self.qcls.FIELDS, self.op.fields)
5445 class LUNodeModifyStorage(NoHooksLU):
5446 """Logical unit for modifying a storage volume on a node.
5451 def CheckArguments(self):
5452 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
5454 storage_type = self.op.storage_type
5457 modifiable = constants.MODIFIABLE_STORAGE_FIELDS[storage_type]
5459 raise errors.OpPrereqError("Storage units of type '%s' can not be"
5460 " modified" % storage_type,
5463 diff = set(self.op.changes.keys()) - modifiable
5465 raise errors.OpPrereqError("The following fields can not be modified for"
5466 " storage units of type '%s': %r" %
5467 (storage_type, list(diff)),
5470 def ExpandNames(self):
5471 self.needed_locks = {
5472 locking.LEVEL_NODE: self.op.node_name,
5475 def Exec(self, feedback_fn):
5476 """Computes the list of nodes and their attributes.
5479 st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
5480 result = self.rpc.call_storage_modify(self.op.node_name,
5481 self.op.storage_type, st_args,
5482 self.op.name, self.op.changes)
5483 result.Raise("Failed to modify storage unit '%s' on %s" %
5484 (self.op.name, self.op.node_name))
5487 class LUNodeAdd(LogicalUnit):
5488 """Logical unit for adding node to the cluster.
5492 HTYPE = constants.HTYPE_NODE
5493 _NFLAGS = ["master_capable", "vm_capable"]
5495 def CheckArguments(self):
5496 self.primary_ip_family = self.cfg.GetPrimaryIPFamily()
5497 # validate/normalize the node name
5498 self.hostname = netutils.GetHostname(name=self.op.node_name,
5499 family=self.primary_ip_family)
5500 self.op.node_name = self.hostname.name
5502 if self.op.readd and self.op.node_name == self.cfg.GetMasterNode():
5503 raise errors.OpPrereqError("Cannot readd the master node",
5506 if self.op.readd and self.op.group:
5507 raise errors.OpPrereqError("Cannot pass a node group when a node is"
5508 " being readded", errors.ECODE_INVAL)
5510 def BuildHooksEnv(self):
5513 This will run on all nodes before, and on all nodes + the new node after.
5517 "OP_TARGET": self.op.node_name,
5518 "NODE_NAME": self.op.node_name,
5519 "NODE_PIP": self.op.primary_ip,
5520 "NODE_SIP": self.op.secondary_ip,
5521 "MASTER_CAPABLE": str(self.op.master_capable),
5522 "VM_CAPABLE": str(self.op.vm_capable),
5525 def BuildHooksNodes(self):
5526 """Build hooks nodes.
5529 # Exclude added node
5530 pre_nodes = list(set(self.cfg.GetNodeList()) - set([self.op.node_name]))
5531 post_nodes = pre_nodes + [self.op.node_name, ]
5533 return (pre_nodes, post_nodes)
5535 def CheckPrereq(self):
5536 """Check prerequisites.
5539 - the new node is not already in the config
5541 - its parameters (single/dual homed) matches the cluster
5543 Any errors are signaled by raising errors.OpPrereqError.
5547 hostname = self.hostname
5548 node = hostname.name
5549 primary_ip = self.op.primary_ip = hostname.ip
5550 if self.op.secondary_ip is None:
5551 if self.primary_ip_family == netutils.IP6Address.family:
5552 raise errors.OpPrereqError("When using a IPv6 primary address, a valid"
5553 " IPv4 address must be given as secondary",
5555 self.op.secondary_ip = primary_ip
5557 secondary_ip = self.op.secondary_ip
5558 if not netutils.IP4Address.IsValid(secondary_ip):
5559 raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
5560 " address" % secondary_ip, errors.ECODE_INVAL)
5562 node_list = cfg.GetNodeList()
5563 if not self.op.readd and node in node_list:
5564 raise errors.OpPrereqError("Node %s is already in the configuration" %
5565 node, errors.ECODE_EXISTS)
5566 elif self.op.readd and node not in node_list:
5567 raise errors.OpPrereqError("Node %s is not in the configuration" % node,
5570 self.changed_primary_ip = False
5572 for existing_node_name, existing_node in cfg.GetMultiNodeInfo(node_list):
5573 if self.op.readd and node == existing_node_name:
5574 if existing_node.secondary_ip != secondary_ip:
5575 raise errors.OpPrereqError("Readded node doesn't have the same IP"
5576 " address configuration as before",
5578 if existing_node.primary_ip != primary_ip:
5579 self.changed_primary_ip = True
5583 if (existing_node.primary_ip == primary_ip or
5584 existing_node.secondary_ip == primary_ip or
5585 existing_node.primary_ip == secondary_ip or
5586 existing_node.secondary_ip == secondary_ip):
5587 raise errors.OpPrereqError("New node ip address(es) conflict with"
5588 " existing node %s" % existing_node.name,
5589 errors.ECODE_NOTUNIQUE)
5591 # After this 'if' block, None is no longer a valid value for the
5592 # _capable op attributes
5594 old_node = self.cfg.GetNodeInfo(node)
5595 assert old_node is not None, "Can't retrieve locked node %s" % node
5596 for attr in self._NFLAGS:
5597 if getattr(self.op, attr) is None:
5598 setattr(self.op, attr, getattr(old_node, attr))
5600 for attr in self._NFLAGS:
5601 if getattr(self.op, attr) is None:
5602 setattr(self.op, attr, True)
5604 if self.op.readd and not self.op.vm_capable:
5605 pri, sec = cfg.GetNodeInstances(node)
5607 raise errors.OpPrereqError("Node %s being re-added with vm_capable"
5608 " flag set to false, but it already holds"
5609 " instances" % node,
5612 # check that the type of the node (single versus dual homed) is the
5613 # same as for the master
5614 myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
5615 master_singlehomed = myself.secondary_ip == myself.primary_ip
5616 newbie_singlehomed = secondary_ip == primary_ip
5617 if master_singlehomed != newbie_singlehomed:
5618 if master_singlehomed:
5619 raise errors.OpPrereqError("The master has no secondary ip but the"
5620 " new node has one",
5623 raise errors.OpPrereqError("The master has a secondary ip but the"
5624 " new node doesn't have one",
5627 # checks reachability
5628 if not netutils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
5629 raise errors.OpPrereqError("Node not reachable by ping",
5630 errors.ECODE_ENVIRON)
5632 if not newbie_singlehomed:
5633 # check reachability from my secondary ip to newbie's secondary ip
5634 if not netutils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
5635 source=myself.secondary_ip):
5636 raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
5637 " based ping to node daemon port",
5638 errors.ECODE_ENVIRON)
5645 if self.op.master_capable:
5646 self.master_candidate = _DecideSelfPromotion(self, exceptions=exceptions)
5648 self.master_candidate = False
5651 self.new_node = old_node
5653 node_group = cfg.LookupNodeGroup(self.op.group)
5654 self.new_node = objects.Node(name=node,
5655 primary_ip=primary_ip,
5656 secondary_ip=secondary_ip,
5657 master_candidate=self.master_candidate,
5658 offline=False, drained=False,
5661 if self.op.ndparams:
5662 utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
5664 if self.op.hv_state:
5665 self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state, None)
5667 if self.op.disk_state:
5668 self.new_disk_state = _MergeAndVerifyDiskState(self.op.disk_state, None)
5670 # TODO: If we need to have multiple DnsOnlyRunner we probably should make
5671 # it a property on the base class.
5672 result = rpc.DnsOnlyRunner().call_version([node])[node]
5673 result.Raise("Can't get version information from node %s" % node)
5674 if constants.PROTOCOL_VERSION == result.payload:
5675 logging.info("Communication to node %s fine, sw version %s match",
5676 node, result.payload)
5678 raise errors.OpPrereqError("Version mismatch master version %s,"
5679 " node version %s" %
5680 (constants.PROTOCOL_VERSION, result.payload),
5681 errors.ECODE_ENVIRON)
5683 def Exec(self, feedback_fn):
5684 """Adds the new node to the cluster.
5687 new_node = self.new_node
5688 node = new_node.name
5690 assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER), \
5693 # We adding a new node so we assume it's powered
5694 new_node.powered = True
5696 # for re-adds, reset the offline/drained/master-candidate flags;
5697 # we need to reset here, otherwise offline would prevent RPC calls
5698 # later in the procedure; this also means that if the re-add
5699 # fails, we are left with a non-offlined, broken node
5701 new_node.drained = new_node.offline = False # pylint: disable=W0201
5702 self.LogInfo("Readding a node, the offline/drained flags were reset")
5703 # if we demote the node, we do cleanup later in the procedure
5704 new_node.master_candidate = self.master_candidate
5705 if self.changed_primary_ip:
5706 new_node.primary_ip = self.op.primary_ip
5708 # copy the master/vm_capable flags
5709 for attr in self._NFLAGS:
5710 setattr(new_node, attr, getattr(self.op, attr))
5712 # notify the user about any possible mc promotion
5713 if new_node.master_candidate:
5714 self.LogInfo("Node will be a master candidate")
5716 if self.op.ndparams:
5717 new_node.ndparams = self.op.ndparams
5719 new_node.ndparams = {}
5721 if self.op.hv_state:
5722 new_node.hv_state_static = self.new_hv_state
5724 if self.op.disk_state:
5725 new_node.disk_state_static = self.new_disk_state
5727 # Add node to our /etc/hosts, and add key to known_hosts
5728 if self.cfg.GetClusterInfo().modify_etc_hosts:
5729 master_node = self.cfg.GetMasterNode()
5730 result = self.rpc.call_etc_hosts_modify(master_node,
5731 constants.ETC_HOSTS_ADD,
5734 result.Raise("Can't update hosts file with new host data")
5736 if new_node.secondary_ip != new_node.primary_ip:
5737 _CheckNodeHasSecondaryIP(self, new_node.name, new_node.secondary_ip,
5740 node_verify_list = [self.cfg.GetMasterNode()]
5741 node_verify_param = {
5742 constants.NV_NODELIST: ([node], {}),
5743 # TODO: do a node-net-test as well?
5746 result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
5747 self.cfg.GetClusterName())
5748 for verifier in node_verify_list:
5749 result[verifier].Raise("Cannot communicate with node %s" % verifier)
5750 nl_payload = result[verifier].payload[constants.NV_NODELIST]
5752 for failed in nl_payload:
5753 feedback_fn("ssh/hostname verification failed"
5754 " (checking from %s): %s" %
5755 (verifier, nl_payload[failed]))
5756 raise errors.OpExecError("ssh/hostname verification failed")
5759 _RedistributeAncillaryFiles(self)
5760 self.context.ReaddNode(new_node)
5761 # make sure we redistribute the config
5762 self.cfg.Update(new_node, feedback_fn)
5763 # and make sure the new node will not have old files around
5764 if not new_node.master_candidate:
5765 result = self.rpc.call_node_demote_from_mc(new_node.name)
5766 msg = result.fail_msg
5768 self.LogWarning("Node failed to demote itself from master"
5769 " candidate status: %s" % msg)
5771 _RedistributeAncillaryFiles(self, additional_nodes=[node],
5772 additional_vm=self.op.vm_capable)
5773 self.context.AddNode(new_node, self.proc.GetECId())
5776 class LUNodeSetParams(LogicalUnit):
5777 """Modifies the parameters of a node.
5779 @cvar _F2R: a dictionary from tuples of flags (mc, drained, offline)
5780 to the node role (as _ROLE_*)
5781 @cvar _R2F: a dictionary from node role to tuples of flags
5782 @cvar _FLAGS: a list of attribute names corresponding to the flags
5785 HPATH = "node-modify"
5786 HTYPE = constants.HTYPE_NODE
5788 (_ROLE_CANDIDATE, _ROLE_DRAINED, _ROLE_OFFLINE, _ROLE_REGULAR) = range(4)
5790 (True, False, False): _ROLE_CANDIDATE,
5791 (False, True, False): _ROLE_DRAINED,
5792 (False, False, True): _ROLE_OFFLINE,
5793 (False, False, False): _ROLE_REGULAR,
5795 _R2F = dict((v, k) for k, v in _F2R.items())
5796 _FLAGS = ["master_candidate", "drained", "offline"]
5798 def CheckArguments(self):
5799 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
5800 all_mods = [self.op.offline, self.op.master_candidate, self.op.drained,
5801 self.op.master_capable, self.op.vm_capable,
5802 self.op.secondary_ip, self.op.ndparams, self.op.hv_state,
5804 if all_mods.count(None) == len(all_mods):
5805 raise errors.OpPrereqError("Please pass at least one modification",
5807 if all_mods.count(True) > 1:
5808 raise errors.OpPrereqError("Can't set the node into more than one"
5809 " state at the same time",
5812 # Boolean value that tells us whether we might be demoting from MC
5813 self.might_demote = (self.op.master_candidate == False or
5814 self.op.offline == True or
5815 self.op.drained == True or
5816 self.op.master_capable == False)
5818 if self.op.secondary_ip:
5819 if not netutils.IP4Address.IsValid(self.op.secondary_ip):
5820 raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
5821 " address" % self.op.secondary_ip,
5824 self.lock_all = self.op.auto_promote and self.might_demote
5825 self.lock_instances = self.op.secondary_ip is not None
5827 def _InstanceFilter(self, instance):
5828 """Filter for getting affected instances.
5831 return (instance.disk_template in constants.DTS_INT_MIRROR and
5832 self.op.node_name in instance.all_nodes)
5834 def ExpandNames(self):
5836 self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET}
5838 self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
5840 # Since modifying a node can have severe effects on currently running
5841 # operations the resource lock is at least acquired in shared mode
5842 self.needed_locks[locking.LEVEL_NODE_RES] = \
5843 self.needed_locks[locking.LEVEL_NODE]
5845 # Get node resource and instance locks in shared mode; they are not used
5846 # for anything but read-only access
5847 self.share_locks[locking.LEVEL_NODE_RES] = 1
5848 self.share_locks[locking.LEVEL_INSTANCE] = 1
5850 if self.lock_instances:
5851 self.needed_locks[locking.LEVEL_INSTANCE] = \
5852 frozenset(self.cfg.GetInstancesInfoByFilter(self._InstanceFilter))
5854 def BuildHooksEnv(self):
5857 This runs on the master node.
5861 "OP_TARGET": self.op.node_name,
5862 "MASTER_CANDIDATE": str(self.op.master_candidate),
5863 "OFFLINE": str(self.op.offline),
5864 "DRAINED": str(self.op.drained),
5865 "MASTER_CAPABLE": str(self.op.master_capable),
5866 "VM_CAPABLE": str(self.op.vm_capable),
5869 def BuildHooksNodes(self):
5870 """Build hooks nodes.
5873 nl = [self.cfg.GetMasterNode(), self.op.node_name]
5876 def CheckPrereq(self):
5877 """Check prerequisites.
5879 This only checks the instance list against the existing names.
5882 node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
5884 if self.lock_instances:
5885 affected_instances = \
5886 self.cfg.GetInstancesInfoByFilter(self._InstanceFilter)
5888 # Verify instance locks
5889 owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
5890 wanted_instances = frozenset(affected_instances.keys())
5891 if wanted_instances - owned_instances:
5892 raise errors.OpPrereqError("Instances affected by changing node %s's"
5893 " secondary IP address have changed since"
5894 " locks were acquired, wanted '%s', have"
5895 " '%s'; retry the operation" %
5897 utils.CommaJoin(wanted_instances),
5898 utils.CommaJoin(owned_instances)),
5901 affected_instances = None
5903 if (self.op.master_candidate is not None or
5904 self.op.drained is not None or
5905 self.op.offline is not None):
5906 # we can't change the master's node flags
5907 if self.op.node_name == self.cfg.GetMasterNode():
5908 raise errors.OpPrereqError("The master role can be changed"
5909 " only via master-failover",
5912 if self.op.master_candidate and not node.master_capable:
5913 raise errors.OpPrereqError("Node %s is not master capable, cannot make"
5914 " it a master candidate" % node.name,
5917 if self.op.vm_capable == False:
5918 (ipri, isec) = self.cfg.GetNodeInstances(self.op.node_name)
5920 raise errors.OpPrereqError("Node %s hosts instances, cannot unset"
5921 " the vm_capable flag" % node.name,
5924 if node.master_candidate and self.might_demote and not self.lock_all:
5925 assert not self.op.auto_promote, "auto_promote set but lock_all not"
5926 # check if after removing the current node, we're missing master
5928 (mc_remaining, mc_should, _) = \
5929 self.cfg.GetMasterCandidateStats(exceptions=[node.name])
5930 if mc_remaining < mc_should:
5931 raise errors.OpPrereqError("Not enough master candidates, please"
5932 " pass auto promote option to allow"
5933 " promotion (--auto-promote or RAPI"
5934 " auto_promote=True)", errors.ECODE_STATE)
5936 self.old_flags = old_flags = (node.master_candidate,
5937 node.drained, node.offline)
5938 assert old_flags in self._F2R, "Un-handled old flags %s" % str(old_flags)
5939 self.old_role = old_role = self._F2R[old_flags]
5941 # Check for ineffective changes
5942 for attr in self._FLAGS:
5943 if (getattr(self.op, attr) == False and getattr(node, attr) == False):
5944 self.LogInfo("Ignoring request to unset flag %s, already unset", attr)
5945 setattr(self.op, attr, None)
5947 # Past this point, any flag change to False means a transition
5948 # away from the respective state, as only real changes are kept
5950 # TODO: We might query the real power state if it supports OOB
5951 if _SupportsOob(self.cfg, node):
5952 if self.op.offline is False and not (node.powered or
5953 self.op.powered == True):
5954 raise errors.OpPrereqError(("Node %s needs to be turned on before its"
5955 " offline status can be reset") %
5957 elif self.op.powered is not None:
5958 raise errors.OpPrereqError(("Unable to change powered state for node %s"
5959 " as it does not support out-of-band"
5960 " handling") % self.op.node_name)
5962 # If we're being deofflined/drained, we'll MC ourself if needed
5963 if (self.op.drained == False or self.op.offline == False or
5964 (self.op.master_capable and not node.master_capable)):
5965 if _DecideSelfPromotion(self):
5966 self.op.master_candidate = True
5967 self.LogInfo("Auto-promoting node to master candidate")
5969 # If we're no longer master capable, we'll demote ourselves from MC
5970 if self.op.master_capable == False and node.master_candidate:
5971 self.LogInfo("Demoting from master candidate")
5972 self.op.master_candidate = False
5975 assert [getattr(self.op, attr) for attr in self._FLAGS].count(True) <= 1
5976 if self.op.master_candidate:
5977 new_role = self._ROLE_CANDIDATE
5978 elif self.op.drained:
5979 new_role = self._ROLE_DRAINED
5980 elif self.op.offline:
5981 new_role = self._ROLE_OFFLINE
5982 elif False in [self.op.master_candidate, self.op.drained, self.op.offline]:
5983 # False is still in new flags, which means we're un-setting (the
5985 new_role = self._ROLE_REGULAR
5986 else: # no new flags, nothing, keep old role
5989 self.new_role = new_role
5991 if old_role == self._ROLE_OFFLINE and new_role != old_role:
5992 # Trying to transition out of offline status
5993 result = self.rpc.call_version([node.name])[node.name]
5995 raise errors.OpPrereqError("Node %s is being de-offlined but fails"
5996 " to report its version: %s" %
5997 (node.name, result.fail_msg),
6000 self.LogWarning("Transitioning node from offline to online state"
6001 " without using re-add. Please make sure the node"
6004 if self.op.secondary_ip:
6005 # Ok even without locking, because this can't be changed by any LU
6006 master = self.cfg.GetNodeInfo(self.cfg.GetMasterNode())
6007 master_singlehomed = master.secondary_ip == master.primary_ip
6008 if master_singlehomed and self.op.secondary_ip:
6009 raise errors.OpPrereqError("Cannot change the secondary ip on a single"
6010 " homed cluster", errors.ECODE_INVAL)
6012 assert not (frozenset(affected_instances) -
6013 self.owned_locks(locking.LEVEL_INSTANCE))
6016 if affected_instances:
6017 raise errors.OpPrereqError("Cannot change secondary IP address:"
6018 " offline node has instances (%s)"
6019 " configured to use it" %
6020 utils.CommaJoin(affected_instances.keys()))
6022 # On online nodes, check that no instances are running, and that
6023 # the node has the new ip and we can reach it.
6024 for instance in affected_instances.values():
6025 _CheckInstanceState(self, instance, INSTANCE_DOWN,
6026 msg="cannot change secondary ip")
6028 _CheckNodeHasSecondaryIP(self, node.name, self.op.secondary_ip, True)
6029 if master.name != node.name:
6030 # check reachability from master secondary ip to new secondary ip
6031 if not netutils.TcpPing(self.op.secondary_ip,
6032 constants.DEFAULT_NODED_PORT,
6033 source=master.secondary_ip):
6034 raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
6035 " based ping to node daemon port",
6036 errors.ECODE_ENVIRON)
6038 if self.op.ndparams:
6039 new_ndparams = _GetUpdatedParams(self.node.ndparams, self.op.ndparams)
6040 utils.ForceDictType(new_ndparams, constants.NDS_PARAMETER_TYPES)
6041 self.new_ndparams = new_ndparams
6043 if self.op.hv_state:
6044 self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state,
6045 self.node.hv_state_static)
6047 if self.op.disk_state:
6048 self.new_disk_state = \
6049 _MergeAndVerifyDiskState(self.op.disk_state,
6050 self.node.disk_state_static)
6052 def Exec(self, feedback_fn):
6057 old_role = self.old_role
6058 new_role = self.new_role
6062 if self.op.ndparams:
6063 node.ndparams = self.new_ndparams
6065 if self.op.powered is not None:
6066 node.powered = self.op.powered
6068 if self.op.hv_state:
6069 node.hv_state_static = self.new_hv_state
6071 if self.op.disk_state:
6072 node.disk_state_static = self.new_disk_state
6074 for attr in ["master_capable", "vm_capable"]:
6075 val = getattr(self.op, attr)
6077 setattr(node, attr, val)
6078 result.append((attr, str(val)))
6080 if new_role != old_role:
6081 # Tell the node to demote itself, if no longer MC and not offline
6082 if old_role == self._ROLE_CANDIDATE and new_role != self._ROLE_OFFLINE:
6083 msg = self.rpc.call_node_demote_from_mc(node.name).fail_msg
6085 self.LogWarning("Node failed to demote itself: %s", msg)
6087 new_flags = self._R2F[new_role]
6088 for of, nf, desc in zip(self.old_flags, new_flags, self._FLAGS):
6090 result.append((desc, str(nf)))
6091 (node.master_candidate, node.drained, node.offline) = new_flags
6093 # we locked all nodes, we adjust the CP before updating this node
6095 _AdjustCandidatePool(self, [node.name])
6097 if self.op.secondary_ip:
6098 node.secondary_ip = self.op.secondary_ip
6099 result.append(("secondary_ip", self.op.secondary_ip))
6101 # this will trigger configuration file update, if needed
6102 self.cfg.Update(node, feedback_fn)
6104 # this will trigger job queue propagation or cleanup if the mc
6106 if [old_role, new_role].count(self._ROLE_CANDIDATE) == 1:
6107 self.context.ReaddNode(node)
6112 class LUNodePowercycle(NoHooksLU):
6113 """Powercycles a node.
6118 def CheckArguments(self):
6119 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
6120 if self.op.node_name == self.cfg.GetMasterNode() and not self.op.force:
6121 raise errors.OpPrereqError("The node is the master and the force"
6122 " parameter was not set",
6125 def ExpandNames(self):
6126 """Locking for PowercycleNode.
6128 This is a last-resort option and shouldn't block on other
6129 jobs. Therefore, we grab no locks.
6132 self.needed_locks = {}
6134 def Exec(self, feedback_fn):
6138 result = self.rpc.call_node_powercycle(self.op.node_name,
6139 self.cfg.GetHypervisorType())
6140 result.Raise("Failed to schedule the reboot")
6141 return result.payload
6144 class LUClusterQuery(NoHooksLU):
6145 """Query cluster configuration.
6150 def ExpandNames(self):
6151 self.needed_locks = {}
6153 def Exec(self, feedback_fn):
6154 """Return cluster config.
6157 cluster = self.cfg.GetClusterInfo()
6160 # Filter just for enabled hypervisors
6161 for os_name, hv_dict in cluster.os_hvp.items():
6162 os_hvp[os_name] = {}
6163 for hv_name, hv_params in hv_dict.items():
6164 if hv_name in cluster.enabled_hypervisors:
6165 os_hvp[os_name][hv_name] = hv_params
6167 # Convert ip_family to ip_version
6168 primary_ip_version = constants.IP4_VERSION
6169 if cluster.primary_ip_family == netutils.IP6Address.family:
6170 primary_ip_version = constants.IP6_VERSION
6173 "software_version": constants.RELEASE_VERSION,
6174 "protocol_version": constants.PROTOCOL_VERSION,
6175 "config_version": constants.CONFIG_VERSION,
6176 "os_api_version": max(constants.OS_API_VERSIONS),
6177 "export_version": constants.EXPORT_VERSION,
6178 "architecture": runtime.GetArchInfo(),
6179 "name": cluster.cluster_name,
6180 "master": cluster.master_node,
6181 "default_hypervisor": cluster.primary_hypervisor,
6182 "enabled_hypervisors": cluster.enabled_hypervisors,
6183 "hvparams": dict([(hypervisor_name, cluster.hvparams[hypervisor_name])
6184 for hypervisor_name in cluster.enabled_hypervisors]),
6186 "beparams": cluster.beparams,
6187 "osparams": cluster.osparams,
6188 "ipolicy": cluster.ipolicy,
6189 "nicparams": cluster.nicparams,
6190 "ndparams": cluster.ndparams,
6191 "diskparams": cluster.diskparams,
6192 "candidate_pool_size": cluster.candidate_pool_size,
6193 "master_netdev": cluster.master_netdev,
6194 "master_netmask": cluster.master_netmask,
6195 "use_external_mip_script": cluster.use_external_mip_script,
6196 "volume_group_name": cluster.volume_group_name,
6197 "drbd_usermode_helper": cluster.drbd_usermode_helper,
6198 "file_storage_dir": cluster.file_storage_dir,
6199 "shared_file_storage_dir": cluster.shared_file_storage_dir,
6200 "maintain_node_health": cluster.maintain_node_health,
6201 "ctime": cluster.ctime,
6202 "mtime": cluster.mtime,
6203 "uuid": cluster.uuid,
6204 "tags": list(cluster.GetTags()),
6205 "uid_pool": cluster.uid_pool,
6206 "default_iallocator": cluster.default_iallocator,
6207 "reserved_lvs": cluster.reserved_lvs,
6208 "primary_ip_version": primary_ip_version,
6209 "prealloc_wipe_disks": cluster.prealloc_wipe_disks,
6210 "hidden_os": cluster.hidden_os,
6211 "blacklisted_os": cluster.blacklisted_os,
6217 class LUClusterConfigQuery(NoHooksLU):
6218 """Return configuration values.
6223 def CheckArguments(self):
6224 self.cq = _ClusterQuery(None, self.op.output_fields, False)
6226 def ExpandNames(self):
6227 self.cq.ExpandNames(self)
6229 def DeclareLocks(self, level):
6230 self.cq.DeclareLocks(self, level)
6232 def Exec(self, feedback_fn):
6233 result = self.cq.OldStyleQuery(self)
6235 assert len(result) == 1
6240 class _ClusterQuery(_QueryBase):
6241 FIELDS = query.CLUSTER_FIELDS
6243 #: Do not sort (there is only one item)
6246 def ExpandNames(self, lu):
6247 lu.needed_locks = {}
6249 # The following variables interact with _QueryBase._GetNames
6250 self.wanted = locking.ALL_SET
6251 self.do_locking = self.use_locking
6254 raise errors.OpPrereqError("Can not use locking for cluster queries",
6257 def DeclareLocks(self, lu, level):
6260 def _GetQueryData(self, lu):
6261 """Computes the list of nodes and their attributes.
6264 # Locking is not used
6265 assert not (compat.any(lu.glm.is_owned(level)
6266 for level in locking.LEVELS
6267 if level != locking.LEVEL_CLUSTER) or
6268 self.do_locking or self.use_locking)
6270 if query.CQ_CONFIG in self.requested_data:
6271 cluster = lu.cfg.GetClusterInfo()
6273 cluster = NotImplemented
6275 if query.CQ_QUEUE_DRAINED in self.requested_data:
6276 drain_flag = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
6278 drain_flag = NotImplemented
6280 if query.CQ_WATCHER_PAUSE in self.requested_data:
6281 watcher_pause = utils.ReadWatcherPauseFile(constants.WATCHER_PAUSEFILE)
6283 watcher_pause = NotImplemented
6285 return query.ClusterQueryData(cluster, drain_flag, watcher_pause)
6288 class LUInstanceActivateDisks(NoHooksLU):
6289 """Bring up an instance's disks.
6294 def ExpandNames(self):
6295 self._ExpandAndLockInstance()
6296 self.needed_locks[locking.LEVEL_NODE] = []
6297 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6299 def DeclareLocks(self, level):
6300 if level == locking.LEVEL_NODE:
6301 self._LockInstancesNodes()
6303 def CheckPrereq(self):
6304 """Check prerequisites.
6306 This checks that the instance is in the cluster.
6309 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6310 assert self.instance is not None, \
6311 "Cannot retrieve locked instance %s" % self.op.instance_name
6312 _CheckNodeOnline(self, self.instance.primary_node)
6314 def Exec(self, feedback_fn):
6315 """Activate the disks.
6318 disks_ok, disks_info = \
6319 _AssembleInstanceDisks(self, self.instance,
6320 ignore_size=self.op.ignore_size)
6322 raise errors.OpExecError("Cannot activate block devices")
6327 def _AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False,
6329 """Prepare the block devices for an instance.
6331 This sets up the block devices on all nodes.
6333 @type lu: L{LogicalUnit}
6334 @param lu: the logical unit on whose behalf we execute
6335 @type instance: L{objects.Instance}
6336 @param instance: the instance for whose disks we assemble
6337 @type disks: list of L{objects.Disk} or None
6338 @param disks: which disks to assemble (or all, if None)
6339 @type ignore_secondaries: boolean
6340 @param ignore_secondaries: if true, errors on secondary nodes
6341 won't result in an error return from the function
6342 @type ignore_size: boolean
6343 @param ignore_size: if true, the current known size of the disk
6344 will not be used during the disk activation, useful for cases
6345 when the size is wrong
6346 @return: False if the operation failed, otherwise a list of
6347 (host, instance_visible_name, node_visible_name)
6348 with the mapping from node devices to instance devices
6353 iname = instance.name
6354 disks = _ExpandCheckDisks(instance, disks)
6356 # With the two passes mechanism we try to reduce the window of
6357 # opportunity for the race condition of switching DRBD to primary
6358 # before handshaking occured, but we do not eliminate it
6360 # The proper fix would be to wait (with some limits) until the
6361 # connection has been made and drbd transitions from WFConnection
6362 # into any other network-connected state (Connected, SyncTarget,
6365 # 1st pass, assemble on all nodes in secondary mode
6366 for idx, inst_disk in enumerate(disks):
6367 for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
6369 node_disk = node_disk.Copy()
6370 node_disk.UnsetSize()
6371 lu.cfg.SetDiskID(node_disk, node)
6372 result = lu.rpc.call_blockdev_assemble(node, (node_disk, instance), iname,
6374 msg = result.fail_msg
6376 lu.proc.LogWarning("Could not prepare block device %s on node %s"
6377 " (is_primary=False, pass=1): %s",
6378 inst_disk.iv_name, node, msg)
6379 if not ignore_secondaries:
6382 # FIXME: race condition on drbd migration to primary
6384 # 2nd pass, do only the primary node
6385 for idx, inst_disk in enumerate(disks):
6388 for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
6389 if node != instance.primary_node:
6392 node_disk = node_disk.Copy()
6393 node_disk.UnsetSize()
6394 lu.cfg.SetDiskID(node_disk, node)
6395 result = lu.rpc.call_blockdev_assemble(node, (node_disk, instance), iname,
6397 msg = result.fail_msg
6399 lu.proc.LogWarning("Could not prepare block device %s on node %s"
6400 " (is_primary=True, pass=2): %s",
6401 inst_disk.iv_name, node, msg)
6404 dev_path = result.payload
6406 device_info.append((instance.primary_node, inst_disk.iv_name, dev_path))
6408 # leave the disks configured for the primary node
6409 # this is a workaround that would be fixed better by
6410 # improving the logical/physical id handling
6412 lu.cfg.SetDiskID(disk, instance.primary_node)
6414 return disks_ok, device_info
6417 def _StartInstanceDisks(lu, instance, force):
6418 """Start the disks of an instance.
6421 disks_ok, _ = _AssembleInstanceDisks(lu, instance,
6422 ignore_secondaries=force)
6424 _ShutdownInstanceDisks(lu, instance)
6425 if force is not None and not force:
6426 lu.proc.LogWarning("", hint="If the message above refers to a"
6428 " you can retry the operation using '--force'.")
6429 raise errors.OpExecError("Disk consistency error")
6432 class LUInstanceDeactivateDisks(NoHooksLU):
6433 """Shutdown an instance's disks.
6438 def ExpandNames(self):
6439 self._ExpandAndLockInstance()
6440 self.needed_locks[locking.LEVEL_NODE] = []
6441 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6443 def DeclareLocks(self, level):
6444 if level == locking.LEVEL_NODE:
6445 self._LockInstancesNodes()
6447 def CheckPrereq(self):
6448 """Check prerequisites.
6450 This checks that the instance is in the cluster.
6453 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6454 assert self.instance is not None, \
6455 "Cannot retrieve locked instance %s" % self.op.instance_name
6457 def Exec(self, feedback_fn):
6458 """Deactivate the disks
6461 instance = self.instance
6463 _ShutdownInstanceDisks(self, instance)
6465 _SafeShutdownInstanceDisks(self, instance)
6468 def _SafeShutdownInstanceDisks(lu, instance, disks=None):
6469 """Shutdown block devices of an instance.
6471 This function checks if an instance is running, before calling
6472 _ShutdownInstanceDisks.
6475 _CheckInstanceState(lu, instance, INSTANCE_DOWN, msg="cannot shutdown disks")
6476 _ShutdownInstanceDisks(lu, instance, disks=disks)
6479 def _ExpandCheckDisks(instance, disks):
6480 """Return the instance disks selected by the disks list
6482 @type disks: list of L{objects.Disk} or None
6483 @param disks: selected disks
6484 @rtype: list of L{objects.Disk}
6485 @return: selected instance disks to act on
6489 return instance.disks
6491 if not set(disks).issubset(instance.disks):
6492 raise errors.ProgrammerError("Can only act on disks belonging to the"
6497 def _ShutdownInstanceDisks(lu, instance, disks=None, ignore_primary=False):
6498 """Shutdown block devices of an instance.
6500 This does the shutdown on all nodes of the instance.
6502 If the ignore_primary is false, errors on the primary node are
6507 disks = _ExpandCheckDisks(instance, disks)
6510 for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
6511 lu.cfg.SetDiskID(top_disk, node)
6512 result = lu.rpc.call_blockdev_shutdown(node, (top_disk, instance))
6513 msg = result.fail_msg
6515 lu.LogWarning("Could not shutdown block device %s on node %s: %s",
6516 disk.iv_name, node, msg)
6517 if ((node == instance.primary_node and not ignore_primary) or
6518 (node != instance.primary_node and not result.offline)):
6523 def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
6524 """Checks if a node has enough free memory.
6526 This function check if a given node has the needed amount of free
6527 memory. In case the node has less memory or we cannot get the
6528 information from the node, this function raise an OpPrereqError
6531 @type lu: C{LogicalUnit}
6532 @param lu: a logical unit from which we get configuration data
6534 @param node: the node to check
6535 @type reason: C{str}
6536 @param reason: string to use in the error message
6537 @type requested: C{int}
6538 @param requested: the amount of memory in MiB to check for
6539 @type hypervisor_name: C{str}
6540 @param hypervisor_name: the hypervisor to ask for memory stats
6542 @return: node current free memory
6543 @raise errors.OpPrereqError: if the node doesn't have enough memory, or
6544 we cannot check the node
6547 nodeinfo = lu.rpc.call_node_info([node], None, [hypervisor_name])
6548 nodeinfo[node].Raise("Can't get data from node %s" % node,
6549 prereq=True, ecode=errors.ECODE_ENVIRON)
6550 (_, _, (hv_info, )) = nodeinfo[node].payload
6552 free_mem = hv_info.get("memory_free", None)
6553 if not isinstance(free_mem, int):
6554 raise errors.OpPrereqError("Can't compute free memory on node %s, result"
6555 " was '%s'" % (node, free_mem),
6556 errors.ECODE_ENVIRON)
6557 if requested > free_mem:
6558 raise errors.OpPrereqError("Not enough memory on node %s for %s:"
6559 " needed %s MiB, available %s MiB" %
6560 (node, reason, requested, free_mem),
6565 def _CheckNodesFreeDiskPerVG(lu, nodenames, req_sizes):
6566 """Checks if nodes have enough free disk space in the all VGs.
6568 This function check if all given nodes have the needed amount of
6569 free disk. In case any node has less disk or we cannot get the
6570 information from the node, this function raise an OpPrereqError
6573 @type lu: C{LogicalUnit}
6574 @param lu: a logical unit from which we get configuration data
6575 @type nodenames: C{list}
6576 @param nodenames: the list of node names to check
6577 @type req_sizes: C{dict}
6578 @param req_sizes: the hash of vg and corresponding amount of disk in
6580 @raise errors.OpPrereqError: if the node doesn't have enough disk,
6581 or we cannot check the node
6584 for vg, req_size in req_sizes.items():
6585 _CheckNodesFreeDiskOnVG(lu, nodenames, vg, req_size)
6588 def _CheckNodesFreeDiskOnVG(lu, nodenames, vg, requested):
6589 """Checks if nodes have enough free disk space in the specified VG.
6591 This function check if all given nodes have the needed amount of
6592 free disk. In case any node has less disk or we cannot get the
6593 information from the node, this function raise an OpPrereqError
6596 @type lu: C{LogicalUnit}
6597 @param lu: a logical unit from which we get configuration data
6598 @type nodenames: C{list}
6599 @param nodenames: the list of node names to check
6601 @param vg: the volume group to check
6602 @type requested: C{int}
6603 @param requested: the amount of disk in MiB to check for
6604 @raise errors.OpPrereqError: if the node doesn't have enough disk,
6605 or we cannot check the node
6608 nodeinfo = lu.rpc.call_node_info(nodenames, [vg], None)
6609 for node in nodenames:
6610 info = nodeinfo[node]
6611 info.Raise("Cannot get current information from node %s" % node,
6612 prereq=True, ecode=errors.ECODE_ENVIRON)
6613 (_, (vg_info, ), _) = info.payload
6614 vg_free = vg_info.get("vg_free", None)
6615 if not isinstance(vg_free, int):
6616 raise errors.OpPrereqError("Can't compute free disk space on node"
6617 " %s for vg %s, result was '%s'" %
6618 (node, vg, vg_free), errors.ECODE_ENVIRON)
6619 if requested > vg_free:
6620 raise errors.OpPrereqError("Not enough disk space on target node %s"
6621 " vg %s: required %d MiB, available %d MiB" %
6622 (node, vg, requested, vg_free),
6626 def _CheckNodesPhysicalCPUs(lu, nodenames, requested, hypervisor_name):
6627 """Checks if nodes have enough physical CPUs
6629 This function checks if all given nodes have the needed number of
6630 physical CPUs. In case any node has less CPUs or we cannot get the
6631 information from the node, this function raises an OpPrereqError
6634 @type lu: C{LogicalUnit}
6635 @param lu: a logical unit from which we get configuration data
6636 @type nodenames: C{list}
6637 @param nodenames: the list of node names to check
6638 @type requested: C{int}
6639 @param requested: the minimum acceptable number of physical CPUs
6640 @raise errors.OpPrereqError: if the node doesn't have enough CPUs,
6641 or we cannot check the node
6644 nodeinfo = lu.rpc.call_node_info(nodenames, None, [hypervisor_name])
6645 for node in nodenames:
6646 info = nodeinfo[node]
6647 info.Raise("Cannot get current information from node %s" % node,
6648 prereq=True, ecode=errors.ECODE_ENVIRON)
6649 (_, _, (hv_info, )) = info.payload
6650 num_cpus = hv_info.get("cpu_total", None)
6651 if not isinstance(num_cpus, int):
6652 raise errors.OpPrereqError("Can't compute the number of physical CPUs"
6653 " on node %s, result was '%s'" %
6654 (node, num_cpus), errors.ECODE_ENVIRON)
6655 if requested > num_cpus:
6656 raise errors.OpPrereqError("Node %s has %s physical CPUs, but %s are "
6657 "required" % (node, num_cpus, requested),
6661 class LUInstanceStartup(LogicalUnit):
6662 """Starts an instance.
6665 HPATH = "instance-start"
6666 HTYPE = constants.HTYPE_INSTANCE
6669 def CheckArguments(self):
6671 if self.op.beparams:
6672 # fill the beparams dict
6673 objects.UpgradeBeParams(self.op.beparams)
6674 utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
6676 def ExpandNames(self):
6677 self._ExpandAndLockInstance()
6678 self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
6680 def DeclareLocks(self, level):
6681 if level == locking.LEVEL_NODE_RES:
6682 self._LockInstancesNodes(primary_only=True, level=locking.LEVEL_NODE_RES)
6684 def BuildHooksEnv(self):
6687 This runs on master, primary and secondary nodes of the instance.
6691 "FORCE": self.op.force,
6694 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6698 def BuildHooksNodes(self):
6699 """Build hooks nodes.
6702 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6705 def CheckPrereq(self):
6706 """Check prerequisites.
6708 This checks that the instance is in the cluster.
6711 self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6712 assert self.instance is not None, \
6713 "Cannot retrieve locked instance %s" % self.op.instance_name
6716 if self.op.hvparams:
6717 # check hypervisor parameter syntax (locally)
6718 cluster = self.cfg.GetClusterInfo()
6719 utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
6720 filled_hvp = cluster.FillHV(instance)
6721 filled_hvp.update(self.op.hvparams)
6722 hv_type = hypervisor.GetHypervisor(instance.hypervisor)
6723 hv_type.CheckParameterSyntax(filled_hvp)
6724 _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
6726 _CheckInstanceState(self, instance, INSTANCE_ONLINE)
6728 self.primary_offline = self.cfg.GetNodeInfo(instance.primary_node).offline
6730 if self.primary_offline and self.op.ignore_offline_nodes:
6731 self.proc.LogWarning("Ignoring offline primary node")
6733 if self.op.hvparams or self.op.beparams:
6734 self.proc.LogWarning("Overridden parameters are ignored")
6736 _CheckNodeOnline(self, instance.primary_node)
6738 bep = self.cfg.GetClusterInfo().FillBE(instance)
6739 bep.update(self.op.beparams)
6741 # check bridges existence
6742 _CheckInstanceBridgesExist(self, instance)
6744 remote_info = self.rpc.call_instance_info(instance.primary_node,
6746 instance.hypervisor)
6747 remote_info.Raise("Error checking node %s" % instance.primary_node,
6748 prereq=True, ecode=errors.ECODE_ENVIRON)
6749 if not remote_info.payload: # not running already
6750 _CheckNodeFreeMemory(self, instance.primary_node,
6751 "starting instance %s" % instance.name,
6752 bep[constants.BE_MINMEM], instance.hypervisor)
6754 def Exec(self, feedback_fn):
6755 """Start the instance.
6758 instance = self.instance
6759 force = self.op.force
6761 if not self.op.no_remember:
6762 self.cfg.MarkInstanceUp(instance.name)
6764 if self.primary_offline:
6765 assert self.op.ignore_offline_nodes
6766 self.proc.LogInfo("Primary node offline, marked instance as started")
6768 node_current = instance.primary_node
6770 _StartInstanceDisks(self, instance, force)
6773 self.rpc.call_instance_start(node_current,
6774 (instance, self.op.hvparams,
6776 self.op.startup_paused)
6777 msg = result.fail_msg
6779 _ShutdownInstanceDisks(self, instance)
6780 raise errors.OpExecError("Could not start instance: %s" % msg)
6783 class LUInstanceReboot(LogicalUnit):
6784 """Reboot an instance.
6787 HPATH = "instance-reboot"
6788 HTYPE = constants.HTYPE_INSTANCE
6791 def ExpandNames(self):
6792 self._ExpandAndLockInstance()
6794 def BuildHooksEnv(self):
6797 This runs on master, primary and secondary nodes of the instance.
6801 "IGNORE_SECONDARIES": self.op.ignore_secondaries,
6802 "REBOOT_TYPE": self.op.reboot_type,
6803 "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
6806 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6810 def BuildHooksNodes(self):
6811 """Build hooks nodes.
6814 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6817 def CheckPrereq(self):
6818 """Check prerequisites.
6820 This checks that the instance is in the cluster.
6823 self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6824 assert self.instance is not None, \
6825 "Cannot retrieve locked instance %s" % self.op.instance_name
6826 _CheckInstanceState(self, instance, INSTANCE_ONLINE)
6827 _CheckNodeOnline(self, instance.primary_node)
6829 # check bridges existence
6830 _CheckInstanceBridgesExist(self, instance)
6832 def Exec(self, feedback_fn):
6833 """Reboot the instance.
6836 instance = self.instance
6837 ignore_secondaries = self.op.ignore_secondaries
6838 reboot_type = self.op.reboot_type
6840 remote_info = self.rpc.call_instance_info(instance.primary_node,
6842 instance.hypervisor)
6843 remote_info.Raise("Error checking node %s" % instance.primary_node)
6844 instance_running = bool(remote_info.payload)
6846 node_current = instance.primary_node
6848 if instance_running and reboot_type in [constants.INSTANCE_REBOOT_SOFT,
6849 constants.INSTANCE_REBOOT_HARD]:
6850 for disk in instance.disks:
6851 self.cfg.SetDiskID(disk, node_current)
6852 result = self.rpc.call_instance_reboot(node_current, instance,
6854 self.op.shutdown_timeout)
6855 result.Raise("Could not reboot instance")
6857 if instance_running:
6858 result = self.rpc.call_instance_shutdown(node_current, instance,
6859 self.op.shutdown_timeout)
6860 result.Raise("Could not shutdown instance for full reboot")
6861 _ShutdownInstanceDisks(self, instance)
6863 self.LogInfo("Instance %s was already stopped, starting now",
6865 _StartInstanceDisks(self, instance, ignore_secondaries)
6866 result = self.rpc.call_instance_start(node_current,
6867 (instance, None, None), False)
6868 msg = result.fail_msg
6870 _ShutdownInstanceDisks(self, instance)
6871 raise errors.OpExecError("Could not start instance for"
6872 " full reboot: %s" % msg)
6874 self.cfg.MarkInstanceUp(instance.name)
6877 class LUInstanceShutdown(LogicalUnit):
6878 """Shutdown an instance.
6881 HPATH = "instance-stop"
6882 HTYPE = constants.HTYPE_INSTANCE
6885 def ExpandNames(self):
6886 self._ExpandAndLockInstance()
6888 def BuildHooksEnv(self):
6891 This runs on master, primary and secondary nodes of the instance.
6894 env = _BuildInstanceHookEnvByObject(self, self.instance)
6895 env["TIMEOUT"] = self.op.timeout
6898 def BuildHooksNodes(self):
6899 """Build hooks nodes.
6902 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6905 def CheckPrereq(self):
6906 """Check prerequisites.
6908 This checks that the instance is in the cluster.
6911 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6912 assert self.instance is not None, \
6913 "Cannot retrieve locked instance %s" % self.op.instance_name
6915 _CheckInstanceState(self, self.instance, INSTANCE_ONLINE)
6917 self.primary_offline = \
6918 self.cfg.GetNodeInfo(self.instance.primary_node).offline
6920 if self.primary_offline and self.op.ignore_offline_nodes:
6921 self.proc.LogWarning("Ignoring offline primary node")
6923 _CheckNodeOnline(self, self.instance.primary_node)
6925 def Exec(self, feedback_fn):
6926 """Shutdown the instance.
6929 instance = self.instance
6930 node_current = instance.primary_node
6931 timeout = self.op.timeout
6933 if not self.op.no_remember:
6934 self.cfg.MarkInstanceDown(instance.name)
6936 if self.primary_offline:
6937 assert self.op.ignore_offline_nodes
6938 self.proc.LogInfo("Primary node offline, marked instance as stopped")
6940 result = self.rpc.call_instance_shutdown(node_current, instance, timeout)
6941 msg = result.fail_msg
6943 self.proc.LogWarning("Could not shutdown instance: %s" % msg)
6945 _ShutdownInstanceDisks(self, instance)
6948 class LUInstanceReinstall(LogicalUnit):
6949 """Reinstall an instance.
6952 HPATH = "instance-reinstall"
6953 HTYPE = constants.HTYPE_INSTANCE
6956 def ExpandNames(self):
6957 self._ExpandAndLockInstance()
6959 def BuildHooksEnv(self):
6962 This runs on master, primary and secondary nodes of the instance.
6965 return _BuildInstanceHookEnvByObject(self, self.instance)
6967 def BuildHooksNodes(self):
6968 """Build hooks nodes.
6971 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6974 def CheckPrereq(self):
6975 """Check prerequisites.
6977 This checks that the instance is in the cluster and is not running.
6980 instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6981 assert instance is not None, \
6982 "Cannot retrieve locked instance %s" % self.op.instance_name
6983 _CheckNodeOnline(self, instance.primary_node, "Instance primary node"
6984 " offline, cannot reinstall")
6985 for node in instance.secondary_nodes:
6986 _CheckNodeOnline(self, node, "Instance secondary node offline,"
6987 " cannot reinstall")
6989 if instance.disk_template == constants.DT_DISKLESS:
6990 raise errors.OpPrereqError("Instance '%s' has no disks" %
6991 self.op.instance_name,
6993 _CheckInstanceState(self, instance, INSTANCE_DOWN, msg="cannot reinstall")
6995 if self.op.os_type is not None:
6997 pnode = _ExpandNodeName(self.cfg, instance.primary_node)
6998 _CheckNodeHasOS(self, pnode, self.op.os_type, self.op.force_variant)
6999 instance_os = self.op.os_type
7001 instance_os = instance.os
7003 nodelist = list(instance.all_nodes)
7005 if self.op.osparams:
7006 i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
7007 _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
7008 self.os_inst = i_osdict # the new dict (without defaults)
7012 self.instance = instance
7014 def Exec(self, feedback_fn):
7015 """Reinstall the instance.
7018 inst = self.instance
7020 if self.op.os_type is not None:
7021 feedback_fn("Changing OS to '%s'..." % self.op.os_type)
7022 inst.os = self.op.os_type
7023 # Write to configuration
7024 self.cfg.Update(inst, feedback_fn)
7026 _StartInstanceDisks(self, inst, None)
7028 feedback_fn("Running the instance OS create scripts...")
7029 # FIXME: pass debug option from opcode to backend
7030 result = self.rpc.call_instance_os_add(inst.primary_node,
7031 (inst, self.os_inst), True,
7032 self.op.debug_level)
7033 result.Raise("Could not install OS for instance %s on node %s" %
7034 (inst.name, inst.primary_node))
7036 _ShutdownInstanceDisks(self, inst)
7039 class LUInstanceRecreateDisks(LogicalUnit):
7040 """Recreate an instance's missing disks.
7043 HPATH = "instance-recreate-disks"
7044 HTYPE = constants.HTYPE_INSTANCE
7047 _MODIFYABLE = frozenset([
7048 constants.IDISK_SIZE,
7049 constants.IDISK_MODE,
7052 # New or changed disk parameters may have different semantics
7053 assert constants.IDISK_PARAMS == (_MODIFYABLE | frozenset([
7054 constants.IDISK_ADOPT,
7056 # TODO: Implement support changing VG while recreating
7058 constants.IDISK_METAVG,
7061 def CheckArguments(self):
7062 if self.op.disks and ht.TPositiveInt(self.op.disks[0]):
7063 # Normalize and convert deprecated list of disk indices
7064 self.op.disks = [(idx, {}) for idx in sorted(frozenset(self.op.disks))]
7066 duplicates = utils.FindDuplicates(map(compat.fst, self.op.disks))
7068 raise errors.OpPrereqError("Some disks have been specified more than"
7069 " once: %s" % utils.CommaJoin(duplicates),
7072 for (idx, params) in self.op.disks:
7073 utils.ForceDictType(params, constants.IDISK_PARAMS_TYPES)
7074 unsupported = frozenset(params.keys()) - self._MODIFYABLE
7076 raise errors.OpPrereqError("Parameters for disk %s try to change"
7077 " unmodifyable parameter(s): %s" %
7078 (idx, utils.CommaJoin(unsupported)),
7081 def ExpandNames(self):
7082 self._ExpandAndLockInstance()
7083 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
7085 self.op.nodes = [_ExpandNodeName(self.cfg, n) for n in self.op.nodes]
7086 self.needed_locks[locking.LEVEL_NODE] = list(self.op.nodes)
7088 self.needed_locks[locking.LEVEL_NODE] = []
7089 self.needed_locks[locking.LEVEL_NODE_RES] = []
7091 def DeclareLocks(self, level):
7092 if level == locking.LEVEL_NODE:
7093 # if we replace the nodes, we only need to lock the old primary,
7094 # otherwise we need to lock all nodes for disk re-creation
7095 primary_only = bool(self.op.nodes)
7096 self._LockInstancesNodes(primary_only=primary_only)
7097 elif level == locking.LEVEL_NODE_RES:
7099 self.needed_locks[locking.LEVEL_NODE_RES] = \
7100 self.needed_locks[locking.LEVEL_NODE][:]
7102 def BuildHooksEnv(self):
7105 This runs on master, primary and secondary nodes of the instance.
7108 return _BuildInstanceHookEnvByObject(self, self.instance)
7110 def BuildHooksNodes(self):
7111 """Build hooks nodes.
7114 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
7117 def CheckPrereq(self):
7118 """Check prerequisites.
7120 This checks that the instance is in the cluster and is not running.
7123 instance = self.cfg.GetInstanceInfo(self.op.instance_name)
7124 assert instance is not None, \
7125 "Cannot retrieve locked instance %s" % self.op.instance_name
7127 if len(self.op.nodes) != len(instance.all_nodes):
7128 raise errors.OpPrereqError("Instance %s currently has %d nodes, but"
7129 " %d replacement nodes were specified" %
7130 (instance.name, len(instance.all_nodes),
7131 len(self.op.nodes)),
7133 assert instance.disk_template != constants.DT_DRBD8 or \
7134 len(self.op.nodes) == 2
7135 assert instance.disk_template != constants.DT_PLAIN or \
7136 len(self.op.nodes) == 1
7137 primary_node = self.op.nodes[0]
7139 primary_node = instance.primary_node
7140 _CheckNodeOnline(self, primary_node)
7142 if instance.disk_template == constants.DT_DISKLESS:
7143 raise errors.OpPrereqError("Instance '%s' has no disks" %
7144 self.op.instance_name, errors.ECODE_INVAL)
7146 # if we replace nodes *and* the old primary is offline, we don't
7148 assert instance.primary_node in self.owned_locks(locking.LEVEL_NODE)
7149 assert instance.primary_node in self.owned_locks(locking.LEVEL_NODE_RES)
7150 old_pnode = self.cfg.GetNodeInfo(instance.primary_node)
7151 if not (self.op.nodes and old_pnode.offline):
7152 _CheckInstanceState(self, instance, INSTANCE_NOT_RUNNING,
7153 msg="cannot recreate disks")
7156 self.disks = dict(self.op.disks)
7158 self.disks = dict((idx, {}) for idx in range(len(instance.disks)))
7160 maxidx = max(self.disks.keys())
7161 if maxidx >= len(instance.disks):
7162 raise errors.OpPrereqError("Invalid disk index '%s'" % maxidx,
7165 if (self.op.nodes and
7166 sorted(self.disks.keys()) != range(len(instance.disks))):
7167 raise errors.OpPrereqError("Can't recreate disks partially and"
7168 " change the nodes at the same time",
7171 self.instance = instance
7173 def Exec(self, feedback_fn):
7174 """Recreate the disks.
7177 instance = self.instance
7179 assert (self.owned_locks(locking.LEVEL_NODE) ==
7180 self.owned_locks(locking.LEVEL_NODE_RES))
7183 mods = [] # keeps track of needed changes
7185 for idx, disk in enumerate(instance.disks):
7187 changes = self.disks[idx]
7189 # Disk should not be recreated
7193 # update secondaries for disks, if needed
7194 if self.op.nodes and disk.dev_type == constants.LD_DRBD8:
7195 # need to update the nodes and minors
7196 assert len(self.op.nodes) == 2
7197 assert len(disk.logical_id) == 6 # otherwise disk internals
7199 (_, _, old_port, _, _, old_secret) = disk.logical_id
7200 new_minors = self.cfg.AllocateDRBDMinor(self.op.nodes, instance.name)
7201 new_id = (self.op.nodes[0], self.op.nodes[1], old_port,
7202 new_minors[0], new_minors[1], old_secret)
7203 assert len(disk.logical_id) == len(new_id)
7207 mods.append((idx, new_id, changes))
7209 # now that we have passed all asserts above, we can apply the mods
7210 # in a single run (to avoid partial changes)
7211 for idx, new_id, changes in mods:
7212 disk = instance.disks[idx]
7213 if new_id is not None:
7214 assert disk.dev_type == constants.LD_DRBD8
7215 disk.logical_id = new_id
7217 disk.Update(size=changes.get(constants.IDISK_SIZE, None),
7218 mode=changes.get(constants.IDISK_MODE, None))
7220 # change primary node, if needed
7222 instance.primary_node = self.op.nodes[0]
7223 self.LogWarning("Changing the instance's nodes, you will have to"
7224 " remove any disks left on the older nodes manually")
7227 self.cfg.Update(instance, feedback_fn)
7229 _CreateDisks(self, instance, to_skip=to_skip)
7232 class LUInstanceRename(LogicalUnit):
7233 """Rename an instance.
7236 HPATH = "instance-rename"
7237 HTYPE = constants.HTYPE_INSTANCE
7239 def CheckArguments(self):
7243 if self.op.ip_check and not self.op.name_check:
7244 # TODO: make the ip check more flexible and not depend on the name check
7245 raise errors.OpPrereqError("IP address check requires a name check",
7248 def BuildHooksEnv(self):
7251 This runs on master, primary and secondary nodes of the instance.
7254 env = _BuildInstanceHookEnvByObject(self, self.instance)
7255 env["INSTANCE_NEW_NAME"] = self.op.new_name
7258 def BuildHooksNodes(self):
7259 """Build hooks nodes.
7262 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
7265 def CheckPrereq(self):
7266 """Check prerequisites.
7268 This checks that the instance is in the cluster and is not running.
7271 self.op.instance_name = _ExpandInstanceName(self.cfg,
7272 self.op.instance_name)
7273 instance = self.cfg.GetInstanceInfo(self.op.instance_name)
7274 assert instance is not None
7275 _CheckNodeOnline(self, instance.primary_node)
7276 _CheckInstanceState(self, instance, INSTANCE_NOT_RUNNING,
7277 msg="cannot rename")
7278 self.instance = instance
7280 new_name = self.op.new_name
7281 if self.op.name_check:
7282 hostname = netutils.GetHostname(name=new_name)
7283 if hostname.name != new_name:
7284 self.LogInfo("Resolved given name '%s' to '%s'", new_name,
7286 if not utils.MatchNameComponent(self.op.new_name, [hostname.name]):
7287 raise errors.OpPrereqError(("Resolved hostname '%s' does not look the"
7288 " same as given hostname '%s'") %
7289 (hostname.name, self.op.new_name),
7291 new_name = self.op.new_name = hostname.name
7292 if (self.op.ip_check and
7293 netutils.TcpPing(hostname.ip, constants.DEFAULT_NODED_PORT)):
7294 raise errors.OpPrereqError("IP %s of instance %s already in use" %
7295 (hostname.ip, new_name),
7296 errors.ECODE_NOTUNIQUE)
7298 instance_list = self.cfg.GetInstanceList()
7299 if new_name in instance_list and new_name != instance.name:
7300 raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
7301 new_name, errors.ECODE_EXISTS)
7303 def Exec(self, feedback_fn):
7304 """Rename the instance.
7307 inst = self.instance
7308 old_name = inst.name
7310 rename_file_storage = False
7311 if (inst.disk_template in constants.DTS_FILEBASED and
7312 self.op.new_name != inst.name):
7313 old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
7314 rename_file_storage = True
7316 self.cfg.RenameInstance(inst.name, self.op.new_name)
7317 # Change the instance lock. This is definitely safe while we hold the BGL.
7318 # Otherwise the new lock would have to be added in acquired mode.
7320 self.glm.remove(locking.LEVEL_INSTANCE, old_name)
7321 self.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
7323 # re-read the instance from the configuration after rename
7324 inst = self.cfg.GetInstanceInfo(self.op.new_name)
7326 if rename_file_storage:
7327 new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
7328 result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
7329 old_file_storage_dir,
7330 new_file_storage_dir)
7331 result.Raise("Could not rename on node %s directory '%s' to '%s'"
7332 " (but the instance has been renamed in Ganeti)" %
7333 (inst.primary_node, old_file_storage_dir,
7334 new_file_storage_dir))
7336 _StartInstanceDisks(self, inst, None)
7338 result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
7339 old_name, self.op.debug_level)
7340 msg = result.fail_msg
7342 msg = ("Could not run OS rename script for instance %s on node %s"
7343 " (but the instance has been renamed in Ganeti): %s" %
7344 (inst.name, inst.primary_node, msg))
7345 self.proc.LogWarning(msg)
7347 _ShutdownInstanceDisks(self, inst)
7352 class LUInstanceRemove(LogicalUnit):
7353 """Remove an instance.
7356 HPATH = "instance-remove"
7357 HTYPE = constants.HTYPE_INSTANCE
7360 def ExpandNames(self):
7361 self._ExpandAndLockInstance()
7362 self.needed_locks[locking.LEVEL_NODE] = []
7363 self.needed_locks[locking.LEVEL_NODE_RES] = []
7364 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
7366 def DeclareLocks(self, level):
7367 if level == locking.LEVEL_NODE:
7368 self._LockInstancesNodes()
7369 elif level == locking.LEVEL_NODE_RES:
7371 self.needed_locks[locking.LEVEL_NODE_RES] = \
7372 self.needed_locks[locking.LEVEL_NODE][:]
7374 def BuildHooksEnv(self):
7377 This runs on master, primary and secondary nodes of the instance.
7380 env = _BuildInstanceHookEnvByObject(self, self.instance)
7381 env["SHUTDOWN_TIMEOUT"] = self.op.shutdown_timeout
7384 def BuildHooksNodes(self):
7385 """Build hooks nodes.
7388 nl = [self.cfg.GetMasterNode()]
7389 nl_post = list(self.instance.all_nodes) + nl
7390 return (nl, nl_post)
7392 def CheckPrereq(self):
7393 """Check prerequisites.
7395 This checks that the instance is in the cluster.
7398 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
7399 assert self.instance is not None, \
7400 "Cannot retrieve locked instance %s" % self.op.instance_name
7402 def Exec(self, feedback_fn):
7403 """Remove the instance.
7406 instance = self.instance
7407 logging.info("Shutting down instance %s on node %s",
7408 instance.name, instance.primary_node)
7410 result = self.rpc.call_instance_shutdown(instance.primary_node, instance,
7411 self.op.shutdown_timeout)
7412 msg = result.fail_msg
7414 if self.op.ignore_failures:
7415 feedback_fn("Warning: can't shutdown instance: %s" % msg)
7417 raise errors.OpExecError("Could not shutdown instance %s on"
7419 (instance.name, instance.primary_node, msg))
7421 assert (self.owned_locks(locking.LEVEL_NODE) ==
7422 self.owned_locks(locking.LEVEL_NODE_RES))
7423 assert not (set(instance.all_nodes) -
7424 self.owned_locks(locking.LEVEL_NODE)), \
7425 "Not owning correct locks"
7427 _RemoveInstance(self, feedback_fn, instance, self.op.ignore_failures)
7430 def _RemoveInstance(lu, feedback_fn, instance, ignore_failures):
7431 """Utility function to remove an instance.
7434 logging.info("Removing block devices for instance %s", instance.name)
7436 if not _RemoveDisks(lu, instance, ignore_failures=ignore_failures):
7437 if not ignore_failures:
7438 raise errors.OpExecError("Can't remove instance's disks")
7439 feedback_fn("Warning: can't remove instance's disks")
7441 logging.info("Removing instance %s out of cluster config", instance.name)
7443 lu.cfg.RemoveInstance(instance.name)
7445 assert not lu.remove_locks.get(locking.LEVEL_INSTANCE), \
7446 "Instance lock removal conflict"
7448 # Remove lock for the instance
7449 lu.remove_locks[locking.LEVEL_INSTANCE] = instance.name
7452 class LUInstanceQuery(NoHooksLU):
7453 """Logical unit for querying instances.
7456 # pylint: disable=W0142
7459 def CheckArguments(self):
7460 self.iq = _InstanceQuery(qlang.MakeSimpleFilter("name", self.op.names),
7461 self.op.output_fields, self.op.use_locking)
7463 def ExpandNames(self):
7464 self.iq.ExpandNames(self)
7466 def DeclareLocks(self, level):
7467 self.iq.DeclareLocks(self, level)
7469 def Exec(self, feedback_fn):
7470 return self.iq.OldStyleQuery(self)
7473 class LUInstanceFailover(LogicalUnit):
7474 """Failover an instance.
7477 HPATH = "instance-failover"
7478 HTYPE = constants.HTYPE_INSTANCE
7481 def CheckArguments(self):
7482 """Check the arguments.
7485 self.iallocator = getattr(self.op, "iallocator", None)
7486 self.target_node = getattr(self.op, "target_node", None)
7488 def ExpandNames(self):
7489 self._ExpandAndLockInstance()
7491 if self.op.target_node is not None:
7492 self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
7494 self.needed_locks[locking.LEVEL_NODE] = []
7495 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
7497 self.needed_locks[locking.LEVEL_NODE_RES] = []
7498 self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
7500 ignore_consistency = self.op.ignore_consistency
7501 shutdown_timeout = self.op.shutdown_timeout
7502 self._migrater = TLMigrateInstance(self, self.op.instance_name,
7505 ignore_consistency=ignore_consistency,
7506 shutdown_timeout=shutdown_timeout,
7507 ignore_ipolicy=self.op.ignore_ipolicy)
7508 self.tasklets = [self._migrater]
7510 def DeclareLocks(self, level):
7511 if level == locking.LEVEL_NODE:
7512 instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
7513 if instance.disk_template in constants.DTS_EXT_MIRROR:
7514 if self.op.target_node is None:
7515 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7517 self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
7518 self.op.target_node]
7519 del self.recalculate_locks[locking.LEVEL_NODE]
7521 self._LockInstancesNodes()
7522 elif level == locking.LEVEL_NODE_RES:
7524 self.needed_locks[locking.LEVEL_NODE_RES] = \
7525 self.needed_locks[locking.LEVEL_NODE][:]
7527 def BuildHooksEnv(self):
7530 This runs on master, primary and secondary nodes of the instance.
7533 instance = self._migrater.instance
7534 source_node = instance.primary_node
7535 target_node = self.op.target_node
7537 "IGNORE_CONSISTENCY": self.op.ignore_consistency,
7538 "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
7539 "OLD_PRIMARY": source_node,
7540 "NEW_PRIMARY": target_node,
7543 if instance.disk_template in constants.DTS_INT_MIRROR:
7544 env["OLD_SECONDARY"] = instance.secondary_nodes[0]
7545 env["NEW_SECONDARY"] = source_node
7547 env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = ""
7549 env.update(_BuildInstanceHookEnvByObject(self, instance))
7553 def BuildHooksNodes(self):
7554 """Build hooks nodes.
7557 instance = self._migrater.instance
7558 nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
7559 return (nl, nl + [instance.primary_node])
7562 class LUInstanceMigrate(LogicalUnit):
7563 """Migrate an instance.
7565 This is migration without shutting down, compared to the failover,
7566 which is done with shutdown.
7569 HPATH = "instance-migrate"
7570 HTYPE = constants.HTYPE_INSTANCE
7573 def ExpandNames(self):
7574 self._ExpandAndLockInstance()
7576 if self.op.target_node is not None:
7577 self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
7579 self.needed_locks[locking.LEVEL_NODE] = []
7580 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
7582 self.needed_locks[locking.LEVEL_NODE] = []
7583 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
7586 TLMigrateInstance(self, self.op.instance_name,
7587 cleanup=self.op.cleanup,
7589 fallback=self.op.allow_failover,
7590 allow_runtime_changes=self.op.allow_runtime_changes,
7591 ignore_ipolicy=self.op.ignore_ipolicy)
7592 self.tasklets = [self._migrater]
7594 def DeclareLocks(self, level):
7595 if level == locking.LEVEL_NODE:
7596 instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
7597 if instance.disk_template in constants.DTS_EXT_MIRROR:
7598 if self.op.target_node is None:
7599 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7601 self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
7602 self.op.target_node]
7603 del self.recalculate_locks[locking.LEVEL_NODE]
7605 self._LockInstancesNodes()
7606 elif level == locking.LEVEL_NODE_RES:
7608 self.needed_locks[locking.LEVEL_NODE_RES] = \
7609 self.needed_locks[locking.LEVEL_NODE][:]
7611 def BuildHooksEnv(self):
7614 This runs on master, primary and secondary nodes of the instance.
7617 instance = self._migrater.instance
7618 source_node = instance.primary_node
7619 target_node = self.op.target_node
7620 env = _BuildInstanceHookEnvByObject(self, instance)
7622 "MIGRATE_LIVE": self._migrater.live,
7623 "MIGRATE_CLEANUP": self.op.cleanup,
7624 "OLD_PRIMARY": source_node,
7625 "NEW_PRIMARY": target_node,
7626 "ALLOW_RUNTIME_CHANGES": self.op.allow_runtime_changes,
7629 if instance.disk_template in constants.DTS_INT_MIRROR:
7630 env["OLD_SECONDARY"] = target_node
7631 env["NEW_SECONDARY"] = source_node
7633 env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = None
7637 def BuildHooksNodes(self):
7638 """Build hooks nodes.
7641 instance = self._migrater.instance
7642 nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
7643 return (nl, nl + [instance.primary_node])
7646 class LUInstanceMove(LogicalUnit):
7647 """Move an instance by data-copying.
7650 HPATH = "instance-move"
7651 HTYPE = constants.HTYPE_INSTANCE
7654 def ExpandNames(self):
7655 self._ExpandAndLockInstance()
7656 target_node = _ExpandNodeName(self.cfg, self.op.target_node)
7657 self.op.target_node = target_node
7658 self.needed_locks[locking.LEVEL_NODE] = [target_node]
7659 self.needed_locks[locking.LEVEL_NODE_RES] = []
7660 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
7662 def DeclareLocks(self, level):
7663 if level == locking.LEVEL_NODE:
7664 self._LockInstancesNodes(primary_only=True)
7665 elif level == locking.LEVEL_NODE_RES:
7667 self.needed_locks[locking.LEVEL_NODE_RES] = \
7668 self.needed_locks[locking.LEVEL_NODE][:]
7670 def BuildHooksEnv(self):
7673 This runs on master, primary and secondary nodes of the instance.
7677 "TARGET_NODE": self.op.target_node,
7678 "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
7680 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
7683 def BuildHooksNodes(self):
7684 """Build hooks nodes.
7688 self.cfg.GetMasterNode(),
7689 self.instance.primary_node,
7690 self.op.target_node,
7694 def CheckPrereq(self):
7695 """Check prerequisites.
7697 This checks that the instance is in the cluster.
7700 self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
7701 assert self.instance is not None, \
7702 "Cannot retrieve locked instance %s" % self.op.instance_name
7704 node = self.cfg.GetNodeInfo(self.op.target_node)
7705 assert node is not None, \
7706 "Cannot retrieve locked node %s" % self.op.target_node
7708 self.target_node = target_node = node.name
7710 if target_node == instance.primary_node:
7711 raise errors.OpPrereqError("Instance %s is already on the node %s" %
7712 (instance.name, target_node),
7715 bep = self.cfg.GetClusterInfo().FillBE(instance)
7717 for idx, dsk in enumerate(instance.disks):
7718 if dsk.dev_type not in (constants.LD_LV, constants.LD_FILE):
7719 raise errors.OpPrereqError("Instance disk %d has a complex layout,"
7720 " cannot copy" % idx, errors.ECODE_STATE)
7722 _CheckNodeOnline(self, target_node)
7723 _CheckNodeNotDrained(self, target_node)
7724 _CheckNodeVmCapable(self, target_node)
7725 ipolicy = _CalculateGroupIPolicy(self.cfg.GetClusterInfo(),
7726 self.cfg.GetNodeGroup(node.group))
7727 _CheckTargetNodeIPolicy(self, ipolicy, instance, node,
7728 ignore=self.op.ignore_ipolicy)
7730 if instance.admin_state == constants.ADMINST_UP:
7731 # check memory requirements on the secondary node
7732 _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
7733 instance.name, bep[constants.BE_MAXMEM],
7734 instance.hypervisor)
7736 self.LogInfo("Not checking memory on the secondary node as"
7737 " instance will not be started")
7739 # check bridge existance
7740 _CheckInstanceBridgesExist(self, instance, node=target_node)
7742 def Exec(self, feedback_fn):
7743 """Move an instance.
7745 The move is done by shutting it down on its present node, copying
7746 the data over (slow) and starting it on the new node.
7749 instance = self.instance
7751 source_node = instance.primary_node
7752 target_node = self.target_node
7754 self.LogInfo("Shutting down instance %s on source node %s",
7755 instance.name, source_node)
7757 assert (self.owned_locks(locking.LEVEL_NODE) ==
7758 self.owned_locks(locking.LEVEL_NODE_RES))
7760 result = self.rpc.call_instance_shutdown(source_node, instance,
7761 self.op.shutdown_timeout)
7762 msg = result.fail_msg
7764 if self.op.ignore_consistency:
7765 self.proc.LogWarning("Could not shutdown instance %s on node %s."
7766 " Proceeding anyway. Please make sure node"
7767 " %s is down. Error details: %s",
7768 instance.name, source_node, source_node, msg)
7770 raise errors.OpExecError("Could not shutdown instance %s on"
7772 (instance.name, source_node, msg))
7774 # create the target disks
7776 _CreateDisks(self, instance, target_node=target_node)
7777 except errors.OpExecError:
7778 self.LogWarning("Device creation failed, reverting...")
7780 _RemoveDisks(self, instance, target_node=target_node)
7782 self.cfg.ReleaseDRBDMinors(instance.name)
7785 cluster_name = self.cfg.GetClusterInfo().cluster_name
7788 # activate, get path, copy the data over
7789 for idx, disk in enumerate(instance.disks):
7790 self.LogInfo("Copying data for disk %d", idx)
7791 result = self.rpc.call_blockdev_assemble(target_node, (disk, instance),
7792 instance.name, True, idx)
7794 self.LogWarning("Can't assemble newly created disk %d: %s",
7795 idx, result.fail_msg)
7796 errs.append(result.fail_msg)
7798 dev_path = result.payload
7799 result = self.rpc.call_blockdev_export(source_node, (disk, instance),
7800 target_node, dev_path,
7803 self.LogWarning("Can't copy data over for disk %d: %s",
7804 idx, result.fail_msg)
7805 errs.append(result.fail_msg)
7809 self.LogWarning("Some disks failed to copy, aborting")
7811 _RemoveDisks(self, instance, target_node=target_node)
7813 self.cfg.ReleaseDRBDMinors(instance.name)
7814 raise errors.OpExecError("Errors during disk copy: %s" %
7817 instance.primary_node = target_node
7818 self.cfg.Update(instance, feedback_fn)
7820 self.LogInfo("Removing the disks on the original node")
7821 _RemoveDisks(self, instance, target_node=source_node)
7823 # Only start the instance if it's marked as up
7824 if instance.admin_state == constants.ADMINST_UP:
7825 self.LogInfo("Starting instance %s on node %s",
7826 instance.name, target_node)
7828 disks_ok, _ = _AssembleInstanceDisks(self, instance,
7829 ignore_secondaries=True)
7831 _ShutdownInstanceDisks(self, instance)
7832 raise errors.OpExecError("Can't activate the instance's disks")
7834 result = self.rpc.call_instance_start(target_node,
7835 (instance, None, None), False)
7836 msg = result.fail_msg
7838 _ShutdownInstanceDisks(self, instance)
7839 raise errors.OpExecError("Could not start instance %s on node %s: %s" %
7840 (instance.name, target_node, msg))
7843 class LUNodeMigrate(LogicalUnit):
7844 """Migrate all instances from a node.
7847 HPATH = "node-migrate"
7848 HTYPE = constants.HTYPE_NODE
7851 def CheckArguments(self):
7854 def ExpandNames(self):
7855 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
7857 self.share_locks = _ShareAll()
7858 self.needed_locks = {
7859 locking.LEVEL_NODE: [self.op.node_name],
7862 def BuildHooksEnv(self):
7865 This runs on the master, the primary and all the secondaries.
7869 "NODE_NAME": self.op.node_name,
7870 "ALLOW_RUNTIME_CHANGES": self.op.allow_runtime_changes,
7873 def BuildHooksNodes(self):
7874 """Build hooks nodes.
7877 nl = [self.cfg.GetMasterNode()]
7880 def CheckPrereq(self):
7883 def Exec(self, feedback_fn):
7884 # Prepare jobs for migration instances
7885 allow_runtime_changes = self.op.allow_runtime_changes
7887 [opcodes.OpInstanceMigrate(instance_name=inst.name,
7890 iallocator=self.op.iallocator,
7891 target_node=self.op.target_node,
7892 allow_runtime_changes=allow_runtime_changes,
7893 ignore_ipolicy=self.op.ignore_ipolicy)]
7894 for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name)
7897 # TODO: Run iallocator in this opcode and pass correct placement options to
7898 # OpInstanceMigrate. Since other jobs can modify the cluster between
7899 # running the iallocator and the actual migration, a good consistency model
7900 # will have to be found.
7902 assert (frozenset(self.owned_locks(locking.LEVEL_NODE)) ==
7903 frozenset([self.op.node_name]))
7905 return ResultWithJobs(jobs)
7908 class TLMigrateInstance(Tasklet):
7909 """Tasklet class for instance migration.
7912 @ivar live: whether the migration will be done live or non-live;
7913 this variable is initalized only after CheckPrereq has run
7914 @type cleanup: boolean
7915 @ivar cleanup: Wheater we cleanup from a failed migration
7916 @type iallocator: string
7917 @ivar iallocator: The iallocator used to determine target_node
7918 @type target_node: string
7919 @ivar target_node: If given, the target_node to reallocate the instance to
7920 @type failover: boolean
7921 @ivar failover: Whether operation results in failover or migration
7922 @type fallback: boolean
7923 @ivar fallback: Whether fallback to failover is allowed if migration not
7925 @type ignore_consistency: boolean
7926 @ivar ignore_consistency: Wheter we should ignore consistency between source
7928 @type shutdown_timeout: int
7929 @ivar shutdown_timeout: In case of failover timeout of the shutdown
7930 @type ignore_ipolicy: bool
7931 @ivar ignore_ipolicy: If true, we can ignore instance policy when migrating
7936 _MIGRATION_POLL_INTERVAL = 1 # seconds
7937 _MIGRATION_FEEDBACK_INTERVAL = 10 # seconds
7939 def __init__(self, lu, instance_name, cleanup=False,
7940 failover=False, fallback=False,
7941 ignore_consistency=False,
7942 allow_runtime_changes=True,
7943 shutdown_timeout=constants.DEFAULT_SHUTDOWN_TIMEOUT,
7944 ignore_ipolicy=False):
7945 """Initializes this class.
7948 Tasklet.__init__(self, lu)
7951 self.instance_name = instance_name
7952 self.cleanup = cleanup
7953 self.live = False # will be overridden later
7954 self.failover = failover
7955 self.fallback = fallback
7956 self.ignore_consistency = ignore_consistency
7957 self.shutdown_timeout = shutdown_timeout
7958 self.ignore_ipolicy = ignore_ipolicy
7959 self.allow_runtime_changes = allow_runtime_changes
7961 def CheckPrereq(self):
7962 """Check prerequisites.
7964 This checks that the instance is in the cluster.
7967 instance_name = _ExpandInstanceName(self.lu.cfg, self.instance_name)
7968 instance = self.cfg.GetInstanceInfo(instance_name)
7969 assert instance is not None
7970 self.instance = instance
7971 cluster = self.cfg.GetClusterInfo()
7973 if (not self.cleanup and
7974 not instance.admin_state == constants.ADMINST_UP and
7975 not self.failover and self.fallback):
7976 self.lu.LogInfo("Instance is marked down or offline, fallback allowed,"
7977 " switching to failover")
7978 self.failover = True
7980 if instance.disk_template not in constants.DTS_MIRRORED:
7985 raise errors.OpPrereqError("Instance's disk layout '%s' does not allow"
7986 " %s" % (instance.disk_template, text),
7989 if instance.disk_template in constants.DTS_EXT_MIRROR:
7990 _CheckIAllocatorOrNode(self.lu, "iallocator", "target_node")
7992 if self.lu.op.iallocator:
7993 self._RunAllocator()
7995 # We set set self.target_node as it is required by
7997 self.target_node = self.lu.op.target_node
7999 # Check that the target node is correct in terms of instance policy
8000 nodeinfo = self.cfg.GetNodeInfo(self.target_node)
8001 group_info = self.cfg.GetNodeGroup(nodeinfo.group)
8002 ipolicy = _CalculateGroupIPolicy(cluster, group_info)
8003 _CheckTargetNodeIPolicy(self.lu, ipolicy, instance, nodeinfo,
8004 ignore=self.ignore_ipolicy)
8006 # self.target_node is already populated, either directly or by the
8008 target_node = self.target_node
8009 if self.target_node == instance.primary_node:
8010 raise errors.OpPrereqError("Cannot migrate instance %s"
8011 " to its primary (%s)" %
8012 (instance.name, instance.primary_node))
8014 if len(self.lu.tasklets) == 1:
8015 # It is safe to release locks only when we're the only tasklet
8017 _ReleaseLocks(self.lu, locking.LEVEL_NODE,
8018 keep=[instance.primary_node, self.target_node])
8021 secondary_nodes = instance.secondary_nodes
8022 if not secondary_nodes:
8023 raise errors.ConfigurationError("No secondary node but using"
8024 " %s disk template" %
8025 instance.disk_template)
8026 target_node = secondary_nodes[0]
8027 if self.lu.op.iallocator or (self.lu.op.target_node and
8028 self.lu.op.target_node != target_node):
8030 text = "failed over"
8033 raise errors.OpPrereqError("Instances with disk template %s cannot"
8034 " be %s to arbitrary nodes"
8035 " (neither an iallocator nor a target"
8036 " node can be passed)" %
8037 (instance.disk_template, text),
8039 nodeinfo = self.cfg.GetNodeInfo(target_node)
8040 group_info = self.cfg.GetNodeGroup(nodeinfo.group)
8041 ipolicy = _CalculateGroupIPolicy(cluster, group_info)
8042 _CheckTargetNodeIPolicy(self.lu, ipolicy, instance, nodeinfo,
8043 ignore=self.ignore_ipolicy)
8045 i_be = cluster.FillBE(instance)
8047 # check memory requirements on the secondary node
8048 if (not self.cleanup and
8049 (not self.failover or instance.admin_state == constants.ADMINST_UP)):
8050 self.tgt_free_mem = _CheckNodeFreeMemory(self.lu, target_node,
8051 "migrating instance %s" %
8053 i_be[constants.BE_MINMEM],
8054 instance.hypervisor)
8056 self.lu.LogInfo("Not checking memory on the secondary node as"
8057 " instance will not be started")
8059 # check if failover must be forced instead of migration
8060 if (not self.cleanup and not self.failover and
8061 i_be[constants.BE_ALWAYS_FAILOVER]):
8063 self.lu.LogInfo("Instance configured to always failover; fallback"
8065 self.failover = True
8067 raise errors.OpPrereqError("This instance has been configured to"
8068 " always failover, please allow failover",
8071 # check bridge existance
8072 _CheckInstanceBridgesExist(self.lu, instance, node=target_node)
8074 if not self.cleanup:
8075 _CheckNodeNotDrained(self.lu, target_node)
8076 if not self.failover:
8077 result = self.rpc.call_instance_migratable(instance.primary_node,
8079 if result.fail_msg and self.fallback:
8080 self.lu.LogInfo("Can't migrate, instance offline, fallback to"
8082 self.failover = True
8084 result.Raise("Can't migrate, please use failover",
8085 prereq=True, ecode=errors.ECODE_STATE)
8087 assert not (self.failover and self.cleanup)
8089 if not self.failover:
8090 if self.lu.op.live is not None and self.lu.op.mode is not None:
8091 raise errors.OpPrereqError("Only one of the 'live' and 'mode'"
8092 " parameters are accepted",
8094 if self.lu.op.live is not None:
8096 self.lu.op.mode = constants.HT_MIGRATION_LIVE
8098 self.lu.op.mode = constants.HT_MIGRATION_NONLIVE
8099 # reset the 'live' parameter to None so that repeated
8100 # invocations of CheckPrereq do not raise an exception
8101 self.lu.op.live = None
8102 elif self.lu.op.mode is None:
8103 # read the default value from the hypervisor
8104 i_hv = cluster.FillHV(self.instance, skip_globals=False)
8105 self.lu.op.mode = i_hv[constants.HV_MIGRATION_MODE]
8107 self.live = self.lu.op.mode == constants.HT_MIGRATION_LIVE
8109 # Failover is never live
8112 if not (self.failover or self.cleanup):
8113 remote_info = self.rpc.call_instance_info(instance.primary_node,
8115 instance.hypervisor)
8116 remote_info.Raise("Error checking instance on node %s" %
8117 instance.primary_node)
8118 instance_running = bool(remote_info.payload)
8119 if instance_running:
8120 self.current_mem = int(remote_info.payload["memory"])
8122 def _RunAllocator(self):
8123 """Run the allocator based on input opcode.
8126 # FIXME: add a self.ignore_ipolicy option
8127 ial = IAllocator(self.cfg, self.rpc,
8128 mode=constants.IALLOCATOR_MODE_RELOC,
8129 name=self.instance_name,
8130 relocate_from=[self.instance.primary_node],
8133 ial.Run(self.lu.op.iallocator)
8136 raise errors.OpPrereqError("Can't compute nodes using"
8137 " iallocator '%s': %s" %
8138 (self.lu.op.iallocator, ial.info),
8140 if len(ial.result) != ial.required_nodes:
8141 raise errors.OpPrereqError("iallocator '%s' returned invalid number"
8142 " of nodes (%s), required %s" %
8143 (self.lu.op.iallocator, len(ial.result),
8144 ial.required_nodes), errors.ECODE_FAULT)
8145 self.target_node = ial.result[0]
8146 self.lu.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
8147 self.instance_name, self.lu.op.iallocator,
8148 utils.CommaJoin(ial.result))
8150 def _WaitUntilSync(self):
8151 """Poll with custom rpc for disk sync.
8153 This uses our own step-based rpc call.
8156 self.feedback_fn("* wait until resync is done")
8160 result = self.rpc.call_drbd_wait_sync(self.all_nodes,
8162 (self.instance.disks,
8165 for node, nres in result.items():
8166 nres.Raise("Cannot resync disks on node %s" % node)
8167 node_done, node_percent = nres.payload
8168 all_done = all_done and node_done
8169 if node_percent is not None:
8170 min_percent = min(min_percent, node_percent)
8172 if min_percent < 100:
8173 self.feedback_fn(" - progress: %.1f%%" % min_percent)
8176 def _EnsureSecondary(self, node):
8177 """Demote a node to secondary.
8180 self.feedback_fn("* switching node %s to secondary mode" % node)
8182 for dev in self.instance.disks:
8183 self.cfg.SetDiskID(dev, node)
8185 result = self.rpc.call_blockdev_close(node, self.instance.name,
8186 self.instance.disks)
8187 result.Raise("Cannot change disk to secondary on node %s" % node)
8189 def _GoStandalone(self):
8190 """Disconnect from the network.
8193 self.feedback_fn("* changing into standalone mode")
8194 result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
8195 self.instance.disks)
8196 for node, nres in result.items():
8197 nres.Raise("Cannot disconnect disks node %s" % node)
8199 def _GoReconnect(self, multimaster):
8200 """Reconnect to the network.
8206 msg = "single-master"
8207 self.feedback_fn("* changing disks into %s mode" % msg)
8208 result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
8209 (self.instance.disks, self.instance),
8210 self.instance.name, multimaster)
8211 for node, nres in result.items():
8212 nres.Raise("Cannot change disks config on node %s" % node)
8214 def _ExecCleanup(self):
8215 """Try to cleanup after a failed migration.
8217 The cleanup is done by:
8218 - check that the instance is running only on one node
8219 (and update the config if needed)
8220 - change disks on its secondary node to secondary
8221 - wait until disks are fully synchronized
8222 - disconnect from the network
8223 - change disks into single-master mode
8224 - wait again until disks are fully synchronized
8227 instance = self.instance
8228 target_node = self.target_node
8229 source_node = self.source_node
8231 # check running on only one node
8232 self.feedback_fn("* checking where the instance actually runs"
8233 " (if this hangs, the hypervisor might be in"
8235 ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
8236 for node, result in ins_l.items():
8237 result.Raise("Can't contact node %s" % node)
8239 runningon_source = instance.name in ins_l[source_node].payload
8240 runningon_target = instance.name in ins_l[target_node].payload
8242 if runningon_source and runningon_target:
8243 raise errors.OpExecError("Instance seems to be running on two nodes,"
8244 " or the hypervisor is confused; you will have"
8245 " to ensure manually that it runs only on one"
8246 " and restart this operation")
8248 if not (runningon_source or runningon_target):
8249 raise errors.OpExecError("Instance does not seem to be running at all;"
8250 " in this case it's safer to repair by"
8251 " running 'gnt-instance stop' to ensure disk"
8252 " shutdown, and then restarting it")
8254 if runningon_target:
8255 # the migration has actually succeeded, we need to update the config
8256 self.feedback_fn("* instance running on secondary node (%s),"
8257 " updating config" % target_node)
8258 instance.primary_node = target_node
8259 self.cfg.Update(instance, self.feedback_fn)
8260 demoted_node = source_node
8262 self.feedback_fn("* instance confirmed to be running on its"
8263 " primary node (%s)" % source_node)
8264 demoted_node = target_node
8266 if instance.disk_template in constants.DTS_INT_MIRROR:
8267 self._EnsureSecondary(demoted_node)
8269 self._WaitUntilSync()
8270 except errors.OpExecError:
8271 # we ignore here errors, since if the device is standalone, it
8272 # won't be able to sync
8274 self._GoStandalone()
8275 self._GoReconnect(False)
8276 self._WaitUntilSync()
8278 self.feedback_fn("* done")
8280 def _RevertDiskStatus(self):
8281 """Try to revert the disk status after a failed migration.
8284 target_node = self.target_node
8285 if self.instance.disk_template in constants.DTS_EXT_MIRROR:
8289 self._EnsureSecondary(target_node)
8290 self._GoStandalone()
8291 self._GoReconnect(False)
8292 self._WaitUntilSync()
8293 except errors.OpExecError, err:
8294 self.lu.LogWarning("Migration failed and I can't reconnect the drives,"
8295 " please try to recover the instance manually;"
8296 " error '%s'" % str(err))
8298 def _AbortMigration(self):
8299 """Call the hypervisor code to abort a started migration.
8302 instance = self.instance
8303 target_node = self.target_node
8304 source_node = self.source_node
8305 migration_info = self.migration_info
8307 abort_result = self.rpc.call_instance_finalize_migration_dst(target_node,
8311 abort_msg = abort_result.fail_msg
8313 logging.error("Aborting migration failed on target node %s: %s",
8314 target_node, abort_msg)
8315 # Don't raise an exception here, as we stil have to try to revert the
8316 # disk status, even if this step failed.
8318 abort_result = self.rpc.call_instance_finalize_migration_src(source_node,
8319 instance, False, self.live)
8320 abort_msg = abort_result.fail_msg
8322 logging.error("Aborting migration failed on source node %s: %s",
8323 source_node, abort_msg)
8325 def _ExecMigration(self):
8326 """Migrate an instance.
8328 The migrate is done by:
8329 - change the disks into dual-master mode
8330 - wait until disks are fully synchronized again
8331 - migrate the instance
8332 - change disks on the new secondary node (the old primary) to secondary
8333 - wait until disks are fully synchronized
8334 - change disks into single-master mode
8337 instance = self.instance
8338 target_node = self.target_node
8339 source_node = self.source_node
8341 # Check for hypervisor version mismatch and warn the user.
8342 nodeinfo = self.rpc.call_node_info([source_node, target_node],
8343 None, [self.instance.hypervisor])
8344 for ninfo in nodeinfo.values():
8345 ninfo.Raise("Unable to retrieve node information from node '%s'" %
8347 (_, _, (src_info, )) = nodeinfo[source_node].payload
8348 (_, _, (dst_info, )) = nodeinfo[target_node].payload
8350 if ((constants.HV_NODEINFO_KEY_VERSION in src_info) and
8351 (constants.HV_NODEINFO_KEY_VERSION in dst_info)):
8352 src_version = src_info[constants.HV_NODEINFO_KEY_VERSION]
8353 dst_version = dst_info[constants.HV_NODEINFO_KEY_VERSION]
8354 if src_version != dst_version:
8355 self.feedback_fn("* warning: hypervisor version mismatch between"
8356 " source (%s) and target (%s) node" %
8357 (src_version, dst_version))
8359 self.feedback_fn("* checking disk consistency between source and target")
8360 for (idx, dev) in enumerate(instance.disks):
8361 if not _CheckDiskConsistency(self.lu, instance, dev, target_node, False):
8362 raise errors.OpExecError("Disk %s is degraded or not fully"
8363 " synchronized on target node,"
8364 " aborting migration" % idx)
8366 if self.current_mem > self.tgt_free_mem:
8367 if not self.allow_runtime_changes:
8368 raise errors.OpExecError("Memory ballooning not allowed and not enough"
8369 " free memory to fit instance %s on target"
8370 " node %s (have %dMB, need %dMB)" %
8371 (instance.name, target_node,
8372 self.tgt_free_mem, self.current_mem))
8373 self.feedback_fn("* setting instance memory to %s" % self.tgt_free_mem)
8374 rpcres = self.rpc.call_instance_balloon_memory(instance.primary_node,
8377 rpcres.Raise("Cannot modify instance runtime memory")
8379 # First get the migration information from the remote node
8380 result = self.rpc.call_migration_info(source_node, instance)
8381 msg = result.fail_msg
8383 log_err = ("Failed fetching source migration information from %s: %s" %
8385 logging.error(log_err)
8386 raise errors.OpExecError(log_err)
8388 self.migration_info = migration_info = result.payload
8390 if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
8391 # Then switch the disks to master/master mode
8392 self._EnsureSecondary(target_node)
8393 self._GoStandalone()
8394 self._GoReconnect(True)
8395 self._WaitUntilSync()
8397 self.feedback_fn("* preparing %s to accept the instance" % target_node)
8398 result = self.rpc.call_accept_instance(target_node,
8401 self.nodes_ip[target_node])
8403 msg = result.fail_msg
8405 logging.error("Instance pre-migration failed, trying to revert"
8406 " disk status: %s", msg)
8407 self.feedback_fn("Pre-migration failed, aborting")
8408 self._AbortMigration()
8409 self._RevertDiskStatus()
8410 raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
8411 (instance.name, msg))
8413 self.feedback_fn("* migrating instance to %s" % target_node)
8414 result = self.rpc.call_instance_migrate(source_node, instance,
8415 self.nodes_ip[target_node],
8417 msg = result.fail_msg
8419 logging.error("Instance migration failed, trying to revert"
8420 " disk status: %s", msg)
8421 self.feedback_fn("Migration failed, aborting")
8422 self._AbortMigration()
8423 self._RevertDiskStatus()
8424 raise errors.OpExecError("Could not migrate instance %s: %s" %
8425 (instance.name, msg))
8427 self.feedback_fn("* starting memory transfer")
8428 last_feedback = time.time()
8430 result = self.rpc.call_instance_get_migration_status(source_node,
8432 msg = result.fail_msg
8433 ms = result.payload # MigrationStatus instance
8434 if msg or (ms.status in constants.HV_MIGRATION_FAILED_STATUSES):
8435 logging.error("Instance migration failed, trying to revert"
8436 " disk status: %s", msg)
8437 self.feedback_fn("Migration failed, aborting")
8438 self._AbortMigration()
8439 self._RevertDiskStatus()
8440 raise errors.OpExecError("Could not migrate instance %s: %s" %
8441 (instance.name, msg))
8443 if result.payload.status != constants.HV_MIGRATION_ACTIVE:
8444 self.feedback_fn("* memory transfer complete")
8447 if (utils.TimeoutExpired(last_feedback,
8448 self._MIGRATION_FEEDBACK_INTERVAL) and
8449 ms.transferred_ram is not None):
8450 mem_progress = 100 * float(ms.transferred_ram) / float(ms.total_ram)
8451 self.feedback_fn("* memory transfer progress: %.2f %%" % mem_progress)
8452 last_feedback = time.time()
8454 time.sleep(self._MIGRATION_POLL_INTERVAL)
8456 result = self.rpc.call_instance_finalize_migration_src(source_node,
8460 msg = result.fail_msg
8462 logging.error("Instance migration succeeded, but finalization failed"
8463 " on the source node: %s", msg)
8464 raise errors.OpExecError("Could not finalize instance migration: %s" %
8467 instance.primary_node = target_node
8469 # distribute new instance config to the other nodes
8470 self.cfg.Update(instance, self.feedback_fn)
8472 result = self.rpc.call_instance_finalize_migration_dst(target_node,
8476 msg = result.fail_msg
8478 logging.error("Instance migration succeeded, but finalization failed"
8479 " on the target node: %s", msg)
8480 raise errors.OpExecError("Could not finalize instance migration: %s" %
8483 if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
8484 self._EnsureSecondary(source_node)
8485 self._WaitUntilSync()
8486 self._GoStandalone()
8487 self._GoReconnect(False)
8488 self._WaitUntilSync()
8490 # If the instance's disk template is `rbd' and there was a successful
8491 # migration, unmap the device from the source node.
8492 if self.instance.disk_template == constants.DT_RBD:
8493 disks = _ExpandCheckDisks(instance, instance.disks)
8494 self.feedback_fn("* unmapping instance's disks from %s" % source_node)
8496 result = self.rpc.call_blockdev_shutdown(source_node, (disk, instance))
8497 msg = result.fail_msg
8499 logging.error("Migration was successful, but couldn't unmap the"
8500 " block device %s on source node %s: %s",
8501 disk.iv_name, source_node, msg)
8502 logging.error("You need to unmap the device %s manually on %s",
8503 disk.iv_name, source_node)
8505 self.feedback_fn("* done")
8507 def _ExecFailover(self):
8508 """Failover an instance.
8510 The failover is done by shutting it down on its present node and
8511 starting it on the secondary.
8514 instance = self.instance
8515 primary_node = self.cfg.GetNodeInfo(instance.primary_node)
8517 source_node = instance.primary_node
8518 target_node = self.target_node
8520 if instance.admin_state == constants.ADMINST_UP:
8521 self.feedback_fn("* checking disk consistency between source and target")
8522 for (idx, dev) in enumerate(instance.disks):
8523 # for drbd, these are drbd over lvm
8524 if not _CheckDiskConsistency(self.lu, instance, dev, target_node,
8526 if primary_node.offline:
8527 self.feedback_fn("Node %s is offline, ignoring degraded disk %s on"
8529 (primary_node.name, idx, target_node))
8530 elif not self.ignore_consistency:
8531 raise errors.OpExecError("Disk %s is degraded on target node,"
8532 " aborting failover" % idx)
8534 self.feedback_fn("* not checking disk consistency as instance is not"
8537 self.feedback_fn("* shutting down instance on source node")
8538 logging.info("Shutting down instance %s on node %s",
8539 instance.name, source_node)
8541 result = self.rpc.call_instance_shutdown(source_node, instance,
8542 self.shutdown_timeout)
8543 msg = result.fail_msg
8545 if self.ignore_consistency or primary_node.offline:
8546 self.lu.LogWarning("Could not shutdown instance %s on node %s,"
8547 " proceeding anyway; please make sure node"
8548 " %s is down; error details: %s",
8549 instance.name, source_node, source_node, msg)
8551 raise errors.OpExecError("Could not shutdown instance %s on"
8553 (instance.name, source_node, msg))
8555 self.feedback_fn("* deactivating the instance's disks on source node")
8556 if not _ShutdownInstanceDisks(self.lu, instance, ignore_primary=True):
8557 raise errors.OpExecError("Can't shut down the instance's disks")
8559 instance.primary_node = target_node
8560 # distribute new instance config to the other nodes
8561 self.cfg.Update(instance, self.feedback_fn)
8563 # Only start the instance if it's marked as up
8564 if instance.admin_state == constants.ADMINST_UP:
8565 self.feedback_fn("* activating the instance's disks on target node %s" %
8567 logging.info("Starting instance %s on node %s",
8568 instance.name, target_node)
8570 disks_ok, _ = _AssembleInstanceDisks(self.lu, instance,
8571 ignore_secondaries=True)
8573 _ShutdownInstanceDisks(self.lu, instance)
8574 raise errors.OpExecError("Can't activate the instance's disks")
8576 self.feedback_fn("* starting the instance on the target node %s" %
8578 result = self.rpc.call_instance_start(target_node, (instance, None, None),
8580 msg = result.fail_msg
8582 _ShutdownInstanceDisks(self.lu, instance)
8583 raise errors.OpExecError("Could not start instance %s on node %s: %s" %
8584 (instance.name, target_node, msg))
8586 def Exec(self, feedback_fn):
8587 """Perform the migration.
8590 self.feedback_fn = feedback_fn
8591 self.source_node = self.instance.primary_node
8593 # FIXME: if we implement migrate-to-any in DRBD, this needs fixing
8594 if self.instance.disk_template in constants.DTS_INT_MIRROR:
8595 self.target_node = self.instance.secondary_nodes[0]
8596 # Otherwise self.target_node has been populated either
8597 # directly, or through an iallocator.
8599 self.all_nodes = [self.source_node, self.target_node]
8600 self.nodes_ip = dict((name, node.secondary_ip) for (name, node)
8601 in self.cfg.GetMultiNodeInfo(self.all_nodes))
8604 feedback_fn("Failover instance %s" % self.instance.name)
8605 self._ExecFailover()
8607 feedback_fn("Migrating instance %s" % self.instance.name)
8610 return self._ExecCleanup()
8612 return self._ExecMigration()
8615 def _CreateBlockDev(lu, node, instance, device, force_create, info,
8617 """Wrapper around L{_CreateBlockDevInner}.
8619 This method annotates the root device first.
8622 (disk,) = _AnnotateDiskParams(instance, [device], lu.cfg)
8623 return _CreateBlockDevInner(lu, node, instance, disk, force_create, info,
8627 def _CreateBlockDevInner(lu, node, instance, device, force_create,
8629 """Create a tree of block devices on a given node.
8631 If this device type has to be created on secondaries, create it and
8634 If not, just recurse to children keeping the same 'force' value.
8636 @attention: The device has to be annotated already.
8638 @param lu: the lu on whose behalf we execute
8639 @param node: the node on which to create the device
8640 @type instance: L{objects.Instance}
8641 @param instance: the instance which owns the device
8642 @type device: L{objects.Disk}
8643 @param device: the device to create
8644 @type force_create: boolean
8645 @param force_create: whether to force creation of this device; this
8646 will be change to True whenever we find a device which has
8647 CreateOnSecondary() attribute
8648 @param info: the extra 'metadata' we should attach to the device
8649 (this will be represented as a LVM tag)
8650 @type force_open: boolean
8651 @param force_open: this parameter will be passes to the
8652 L{backend.BlockdevCreate} function where it specifies
8653 whether we run on primary or not, and it affects both
8654 the child assembly and the device own Open() execution
8657 if device.CreateOnSecondary():
8661 for child in device.children:
8662 _CreateBlockDevInner(lu, node, instance, child, force_create,
8665 if not force_create:
8668 _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
8671 def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
8672 """Create a single block device on a given node.
8674 This will not recurse over children of the device, so they must be
8677 @param lu: the lu on whose behalf we execute
8678 @param node: the node on which to create the device
8679 @type instance: L{objects.Instance}
8680 @param instance: the instance which owns the device
8681 @type device: L{objects.Disk}
8682 @param device: the device to create
8683 @param info: the extra 'metadata' we should attach to the device
8684 (this will be represented as a LVM tag)
8685 @type force_open: boolean
8686 @param force_open: this parameter will be passes to the
8687 L{backend.BlockdevCreate} function where it specifies
8688 whether we run on primary or not, and it affects both
8689 the child assembly and the device own Open() execution
8692 lu.cfg.SetDiskID(device, node)
8693 result = lu.rpc.call_blockdev_create(node, device, device.size,
8694 instance.name, force_open, info)
8695 result.Raise("Can't create block device %s on"
8696 " node %s for instance %s" % (device, node, instance.name))
8697 if device.physical_id is None:
8698 device.physical_id = result.payload
8701 def _GenerateUniqueNames(lu, exts):
8702 """Generate a suitable LV name.
8704 This will generate a logical volume name for the given instance.
8709 new_id = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
8710 results.append("%s%s" % (new_id, val))
8714 def _GenerateDRBD8Branch(lu, primary, secondary, size, vgnames, names,
8715 iv_name, p_minor, s_minor):
8716 """Generate a drbd8 device complete with its children.
8719 assert len(vgnames) == len(names) == 2
8720 port = lu.cfg.AllocatePort()
8721 shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
8723 dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
8724 logical_id=(vgnames[0], names[0]),
8726 dev_meta = objects.Disk(dev_type=constants.LD_LV, size=DRBD_META_SIZE,
8727 logical_id=(vgnames[1], names[1]),
8729 drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
8730 logical_id=(primary, secondary, port,
8733 children=[dev_data, dev_meta],
8734 iv_name=iv_name, params={})
8738 _DISK_TEMPLATE_NAME_PREFIX = {
8739 constants.DT_PLAIN: "",
8740 constants.DT_RBD: ".rbd",
8744 _DISK_TEMPLATE_DEVICE_TYPE = {
8745 constants.DT_PLAIN: constants.LD_LV,
8746 constants.DT_FILE: constants.LD_FILE,
8747 constants.DT_SHARED_FILE: constants.LD_FILE,
8748 constants.DT_BLOCK: constants.LD_BLOCKDEV,
8749 constants.DT_RBD: constants.LD_RBD,
8753 def _GenerateDiskTemplate(lu, template_name, instance_name, primary_node,
8754 secondary_nodes, disk_info, file_storage_dir, file_driver, base_index,
8755 feedback_fn, full_disk_params, _req_file_storage=opcodes.RequireFileStorage,
8756 _req_shr_file_storage=opcodes.RequireSharedFileStorage):
8757 """Generate the entire disk layout for a given template type.
8760 #TODO: compute space requirements
8762 vgname = lu.cfg.GetVGName()
8763 disk_count = len(disk_info)
8766 if template_name == constants.DT_DISKLESS:
8768 elif template_name == constants.DT_DRBD8:
8769 if len(secondary_nodes) != 1:
8770 raise errors.ProgrammerError("Wrong template configuration")
8771 remote_node = secondary_nodes[0]
8772 minors = lu.cfg.AllocateDRBDMinor(
8773 [primary_node, remote_node] * len(disk_info), instance_name)
8775 (drbd_params, _, _) = objects.Disk.ComputeLDParams(template_name,
8777 drbd_default_metavg = drbd_params[constants.LDP_DEFAULT_METAVG]
8780 for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
8781 for i in range(disk_count)]):
8782 names.append(lv_prefix + "_data")
8783 names.append(lv_prefix + "_meta")
8784 for idx, disk in enumerate(disk_info):
8785 disk_index = idx + base_index
8786 data_vg = disk.get(constants.IDISK_VG, vgname)
8787 meta_vg = disk.get(constants.IDISK_METAVG, drbd_default_metavg)
8788 disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
8789 disk[constants.IDISK_SIZE],
8791 names[idx * 2:idx * 2 + 2],
8792 "disk/%d" % disk_index,
8793 minors[idx * 2], minors[idx * 2 + 1])
8794 disk_dev.mode = disk[constants.IDISK_MODE]
8795 disks.append(disk_dev)
8798 raise errors.ProgrammerError("Wrong template configuration")
8800 if template_name == constants.DT_FILE:
8802 elif template_name == constants.DT_SHARED_FILE:
8803 _req_shr_file_storage()
8805 name_prefix = _DISK_TEMPLATE_NAME_PREFIX.get(template_name, None)
8806 if name_prefix is None:
8809 names = _GenerateUniqueNames(lu, ["%s.disk%s" %
8810 (name_prefix, base_index + i)
8811 for i in range(disk_count)])
8813 if template_name == constants.DT_PLAIN:
8814 def logical_id_fn(idx, _, disk):
8815 vg = disk.get(constants.IDISK_VG, vgname)
8816 return (vg, names[idx])
8817 elif template_name in (constants.DT_FILE, constants.DT_SHARED_FILE):
8819 lambda _, disk_index, disk: (file_driver,
8820 "%s/disk%d" % (file_storage_dir,
8822 elif template_name == constants.DT_BLOCK:
8824 lambda idx, disk_index, disk: (constants.BLOCKDEV_DRIVER_MANUAL,
8825 disk[constants.IDISK_ADOPT])
8826 elif template_name == constants.DT_RBD:
8827 logical_id_fn = lambda idx, _, disk: ("rbd", names[idx])
8829 raise errors.ProgrammerError("Unknown disk template '%s'" % template_name)
8831 dev_type = _DISK_TEMPLATE_DEVICE_TYPE[template_name]
8833 for idx, disk in enumerate(disk_info):
8834 disk_index = idx + base_index
8835 size = disk[constants.IDISK_SIZE]
8836 feedback_fn("* disk %s, size %s" %
8837 (disk_index, utils.FormatUnit(size, "h")))
8838 disks.append(objects.Disk(dev_type=dev_type, size=size,
8839 logical_id=logical_id_fn(idx, disk_index, disk),
8840 iv_name="disk/%d" % disk_index,
8841 mode=disk[constants.IDISK_MODE],
8847 def _GetInstanceInfoText(instance):
8848 """Compute that text that should be added to the disk's metadata.
8851 return "originstname+%s" % instance.name
8854 def _CalcEta(time_taken, written, total_size):
8855 """Calculates the ETA based on size written and total size.
8857 @param time_taken: The time taken so far
8858 @param written: amount written so far
8859 @param total_size: The total size of data to be written
8860 @return: The remaining time in seconds
8863 avg_time = time_taken / float(written)
8864 return (total_size - written) * avg_time
8867 def _WipeDisks(lu, instance):
8868 """Wipes instance disks.
8870 @type lu: L{LogicalUnit}
8871 @param lu: the logical unit on whose behalf we execute
8872 @type instance: L{objects.Instance}
8873 @param instance: the instance whose disks we should create
8874 @return: the success of the wipe
8877 node = instance.primary_node
8879 for device in instance.disks:
8880 lu.cfg.SetDiskID(device, node)
8882 logging.info("Pause sync of instance %s disks", instance.name)
8883 result = lu.rpc.call_blockdev_pause_resume_sync(node,
8884 (instance.disks, instance),
8887 for idx, success in enumerate(result.payload):
8889 logging.warn("pause-sync of instance %s for disks %d failed",
8893 for idx, device in enumerate(instance.disks):
8894 # The wipe size is MIN_WIPE_CHUNK_PERCENT % of the instance disk but
8895 # MAX_WIPE_CHUNK at max
8896 wipe_chunk_size = min(constants.MAX_WIPE_CHUNK, device.size / 100.0 *
8897 constants.MIN_WIPE_CHUNK_PERCENT)
8898 # we _must_ make this an int, otherwise rounding errors will
8900 wipe_chunk_size = int(wipe_chunk_size)
8902 lu.LogInfo("* Wiping disk %d", idx)
8903 logging.info("Wiping disk %d for instance %s, node %s using"
8904 " chunk size %s", idx, instance.name, node, wipe_chunk_size)
8909 start_time = time.time()
8911 while offset < size:
8912 wipe_size = min(wipe_chunk_size, size - offset)
8913 logging.debug("Wiping disk %d, offset %s, chunk %s",
8914 idx, offset, wipe_size)
8915 result = lu.rpc.call_blockdev_wipe(node, (device, instance), offset,
8917 result.Raise("Could not wipe disk %d at offset %d for size %d" %
8918 (idx, offset, wipe_size))
8921 if now - last_output >= 60:
8922 eta = _CalcEta(now - start_time, offset, size)
8923 lu.LogInfo(" - done: %.1f%% ETA: %s" %
8924 (offset / float(size) * 100, utils.FormatSeconds(eta)))
8927 logging.info("Resume sync of instance %s disks", instance.name)
8929 result = lu.rpc.call_blockdev_pause_resume_sync(node,
8930 (instance.disks, instance),
8933 for idx, success in enumerate(result.payload):
8935 lu.LogWarning("Resume sync of disk %d failed, please have a"
8936 " look at the status and troubleshoot the issue", idx)
8937 logging.warn("resume-sync of instance %s for disks %d failed",
8941 def _CreateDisks(lu, instance, to_skip=None, target_node=None):
8942 """Create all disks for an instance.
8944 This abstracts away some work from AddInstance.
8946 @type lu: L{LogicalUnit}
8947 @param lu: the logical unit on whose behalf we execute
8948 @type instance: L{objects.Instance}
8949 @param instance: the instance whose disks we should create
8951 @param to_skip: list of indices to skip
8952 @type target_node: string
8953 @param target_node: if passed, overrides the target node for creation
8955 @return: the success of the creation
8958 info = _GetInstanceInfoText(instance)
8959 if target_node is None:
8960 pnode = instance.primary_node
8961 all_nodes = instance.all_nodes
8966 if instance.disk_template in constants.DTS_FILEBASED:
8967 file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
8968 result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
8970 result.Raise("Failed to create directory '%s' on"
8971 " node %s" % (file_storage_dir, pnode))
8973 # Note: this needs to be kept in sync with adding of disks in
8974 # LUInstanceSetParams
8975 for idx, device in enumerate(instance.disks):
8976 if to_skip and idx in to_skip:
8978 logging.info("Creating disk %s for instance '%s'", idx, instance.name)
8980 for node in all_nodes:
8981 f_create = node == pnode
8982 _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
8985 def _RemoveDisks(lu, instance, target_node=None, ignore_failures=False):
8986 """Remove all disks for an instance.
8988 This abstracts away some work from `AddInstance()` and
8989 `RemoveInstance()`. Note that in case some of the devices couldn't
8990 be removed, the removal will continue with the other ones (compare
8991 with `_CreateDisks()`).
8993 @type lu: L{LogicalUnit}
8994 @param lu: the logical unit on whose behalf we execute
8995 @type instance: L{objects.Instance}
8996 @param instance: the instance whose disks we should remove
8997 @type target_node: string
8998 @param target_node: used to override the node on which to remove the disks
9000 @return: the success of the removal
9003 logging.info("Removing block devices for instance %s", instance.name)
9006 ports_to_release = set()
9007 anno_disks = _AnnotateDiskParams(instance, instance.disks, lu.cfg)
9008 for (idx, device) in enumerate(anno_disks):
9010 edata = [(target_node, device)]
9012 edata = device.ComputeNodeTree(instance.primary_node)
9013 for node, disk in edata:
9014 lu.cfg.SetDiskID(disk, node)
9015 msg = lu.rpc.call_blockdev_remove(node, disk).fail_msg
9017 lu.LogWarning("Could not remove disk %s on node %s,"
9018 " continuing anyway: %s", idx, node, msg)
9021 # if this is a DRBD disk, return its port to the pool
9022 if device.dev_type in constants.LDS_DRBD:
9023 ports_to_release.add(device.logical_id[2])
9025 if all_result or ignore_failures:
9026 for port in ports_to_release:
9027 lu.cfg.AddTcpUdpPort(port)
9029 if instance.disk_template == constants.DT_FILE:
9030 file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
9034 tgt = instance.primary_node
9035 result = lu.rpc.call_file_storage_dir_remove(tgt, file_storage_dir)
9037 lu.LogWarning("Could not remove directory '%s' on node %s: %s",
9038 file_storage_dir, instance.primary_node, result.fail_msg)
9044 def _ComputeDiskSizePerVG(disk_template, disks):
9045 """Compute disk size requirements in the volume group
9048 def _compute(disks, payload):
9049 """Universal algorithm.
9054 vgs[disk[constants.IDISK_VG]] = \
9055 vgs.get(constants.IDISK_VG, 0) + disk[constants.IDISK_SIZE] + payload
9059 # Required free disk space as a function of disk and swap space
9061 constants.DT_DISKLESS: {},
9062 constants.DT_PLAIN: _compute(disks, 0),
9063 # 128 MB are added for drbd metadata for each disk
9064 constants.DT_DRBD8: _compute(disks, DRBD_META_SIZE),
9065 constants.DT_FILE: {},
9066 constants.DT_SHARED_FILE: {},
9069 if disk_template not in req_size_dict:
9070 raise errors.ProgrammerError("Disk template '%s' size requirement"
9071 " is unknown" % disk_template)
9073 return req_size_dict[disk_template]
9076 def _ComputeDiskSize(disk_template, disks):
9077 """Compute disk size requirements in the volume group
9080 # Required free disk space as a function of disk and swap space
9082 constants.DT_DISKLESS: None,
9083 constants.DT_PLAIN: sum(d[constants.IDISK_SIZE] for d in disks),
9084 # 128 MB are added for drbd metadata for each disk
9086 sum(d[constants.IDISK_SIZE] + DRBD_META_SIZE for d in disks),
9087 constants.DT_FILE: None,
9088 constants.DT_SHARED_FILE: 0,
9089 constants.DT_BLOCK: 0,
9090 constants.DT_RBD: 0,
9093 if disk_template not in req_size_dict:
9094 raise errors.ProgrammerError("Disk template '%s' size requirement"
9095 " is unknown" % disk_template)
9097 return req_size_dict[disk_template]
9100 def _FilterVmNodes(lu, nodenames):
9101 """Filters out non-vm_capable nodes from a list.
9103 @type lu: L{LogicalUnit}
9104 @param lu: the logical unit for which we check
9105 @type nodenames: list
9106 @param nodenames: the list of nodes on which we should check
9108 @return: the list of vm-capable nodes
9111 vm_nodes = frozenset(lu.cfg.GetNonVmCapableNodeList())
9112 return [name for name in nodenames if name not in vm_nodes]
9115 def _CheckHVParams(lu, nodenames, hvname, hvparams):
9116 """Hypervisor parameter validation.
9118 This function abstract the hypervisor parameter validation to be
9119 used in both instance create and instance modify.
9121 @type lu: L{LogicalUnit}
9122 @param lu: the logical unit for which we check
9123 @type nodenames: list
9124 @param nodenames: the list of nodes on which we should check
9125 @type hvname: string
9126 @param hvname: the name of the hypervisor we should use
9127 @type hvparams: dict
9128 @param hvparams: the parameters which we need to check
9129 @raise errors.OpPrereqError: if the parameters are not valid
9132 nodenames = _FilterVmNodes(lu, nodenames)
9134 cluster = lu.cfg.GetClusterInfo()
9135 hvfull = objects.FillDict(cluster.hvparams.get(hvname, {}), hvparams)
9137 hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames, hvname, hvfull)
9138 for node in nodenames:
9142 info.Raise("Hypervisor parameter validation failed on node %s" % node)
9145 def _CheckOSParams(lu, required, nodenames, osname, osparams):
9146 """OS parameters validation.
9148 @type lu: L{LogicalUnit}
9149 @param lu: the logical unit for which we check
9150 @type required: boolean
9151 @param required: whether the validation should fail if the OS is not
9153 @type nodenames: list
9154 @param nodenames: the list of nodes on which we should check
9155 @type osname: string
9156 @param osname: the name of the hypervisor we should use
9157 @type osparams: dict
9158 @param osparams: the parameters which we need to check
9159 @raise errors.OpPrereqError: if the parameters are not valid
9162 nodenames = _FilterVmNodes(lu, nodenames)
9163 result = lu.rpc.call_os_validate(nodenames, required, osname,
9164 [constants.OS_VALIDATE_PARAMETERS],
9166 for node, nres in result.items():
9167 # we don't check for offline cases since this should be run only
9168 # against the master node and/or an instance's nodes
9169 nres.Raise("OS Parameters validation failed on node %s" % node)
9170 if not nres.payload:
9171 lu.LogInfo("OS %s not found on node %s, validation skipped",
9175 class LUInstanceCreate(LogicalUnit):
9176 """Create an instance.
9179 HPATH = "instance-add"
9180 HTYPE = constants.HTYPE_INSTANCE
9183 def CheckArguments(self):
9187 # do not require name_check to ease forward/backward compatibility
9189 if self.op.no_install and self.op.start:
9190 self.LogInfo("No-installation mode selected, disabling startup")
9191 self.op.start = False
9192 # validate/normalize the instance name
9193 self.op.instance_name = \
9194 netutils.Hostname.GetNormalizedName(self.op.instance_name)
9196 if self.op.ip_check and not self.op.name_check:
9197 # TODO: make the ip check more flexible and not depend on the name check
9198 raise errors.OpPrereqError("Cannot do IP address check without a name"
9199 " check", errors.ECODE_INVAL)
9201 # check nics' parameter names
9202 for nic in self.op.nics:
9203 utils.ForceDictType(nic, constants.INIC_PARAMS_TYPES)
9205 # check disks. parameter names and consistent adopt/no-adopt strategy
9206 has_adopt = has_no_adopt = False
9207 for disk in self.op.disks:
9208 utils.ForceDictType(disk, constants.IDISK_PARAMS_TYPES)
9209 if constants.IDISK_ADOPT in disk:
9213 if has_adopt and has_no_adopt:
9214 raise errors.OpPrereqError("Either all disks are adopted or none is",
9217 if self.op.disk_template not in constants.DTS_MAY_ADOPT:
9218 raise errors.OpPrereqError("Disk adoption is not supported for the"
9219 " '%s' disk template" %
9220 self.op.disk_template,
9222 if self.op.iallocator is not None:
9223 raise errors.OpPrereqError("Disk adoption not allowed with an"
9224 " iallocator script", errors.ECODE_INVAL)
9225 if self.op.mode == constants.INSTANCE_IMPORT:
9226 raise errors.OpPrereqError("Disk adoption not allowed for"
9227 " instance import", errors.ECODE_INVAL)
9229 if self.op.disk_template in constants.DTS_MUST_ADOPT:
9230 raise errors.OpPrereqError("Disk template %s requires disk adoption,"
9231 " but no 'adopt' parameter given" %
9232 self.op.disk_template,
9235 self.adopt_disks = has_adopt
9237 # instance name verification
9238 if self.op.name_check:
9239 self.hostname1 = netutils.GetHostname(name=self.op.instance_name)
9240 self.op.instance_name = self.hostname1.name
9241 # used in CheckPrereq for ip ping check
9242 self.check_ip = self.hostname1.ip
9244 self.check_ip = None
9246 # file storage checks
9247 if (self.op.file_driver and
9248 not self.op.file_driver in constants.FILE_DRIVER):
9249 raise errors.OpPrereqError("Invalid file driver name '%s'" %
9250 self.op.file_driver, errors.ECODE_INVAL)
9252 if self.op.disk_template == constants.DT_FILE:
9253 opcodes.RequireFileStorage()
9254 elif self.op.disk_template == constants.DT_SHARED_FILE:
9255 opcodes.RequireSharedFileStorage()
9257 ### Node/iallocator related checks
9258 _CheckIAllocatorOrNode(self, "iallocator", "pnode")
9260 if self.op.pnode is not None:
9261 if self.op.disk_template in constants.DTS_INT_MIRROR:
9262 if self.op.snode is None:
9263 raise errors.OpPrereqError("The networked disk templates need"
9264 " a mirror node", errors.ECODE_INVAL)
9266 self.LogWarning("Secondary node will be ignored on non-mirrored disk"
9268 self.op.snode = None
9270 self._cds = _GetClusterDomainSecret()
9272 if self.op.mode == constants.INSTANCE_IMPORT:
9273 # On import force_variant must be True, because if we forced it at
9274 # initial install, our only chance when importing it back is that it
9276 self.op.force_variant = True
9278 if self.op.no_install:
9279 self.LogInfo("No-installation mode has no effect during import")
9281 elif self.op.mode == constants.INSTANCE_CREATE:
9282 if self.op.os_type is None:
9283 raise errors.OpPrereqError("No guest OS specified",
9285 if self.op.os_type in self.cfg.GetClusterInfo().blacklisted_os:
9286 raise errors.OpPrereqError("Guest OS '%s' is not allowed for"
9287 " installation" % self.op.os_type,
9289 if self.op.disk_template is None:
9290 raise errors.OpPrereqError("No disk template specified",
9293 elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
9294 # Check handshake to ensure both clusters have the same domain secret
9295 src_handshake = self.op.source_handshake
9296 if not src_handshake:
9297 raise errors.OpPrereqError("Missing source handshake",
9300 errmsg = masterd.instance.CheckRemoteExportHandshake(self._cds,
9303 raise errors.OpPrereqError("Invalid handshake: %s" % errmsg,
9306 # Load and check source CA
9307 self.source_x509_ca_pem = self.op.source_x509_ca
9308 if not self.source_x509_ca_pem:
9309 raise errors.OpPrereqError("Missing source X509 CA",
9313 (cert, _) = utils.LoadSignedX509Certificate(self.source_x509_ca_pem,
9315 except OpenSSL.crypto.Error, err:
9316 raise errors.OpPrereqError("Unable to load source X509 CA (%s)" %
9317 (err, ), errors.ECODE_INVAL)
9319 (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
9320 if errcode is not None:
9321 raise errors.OpPrereqError("Invalid source X509 CA (%s)" % (msg, ),
9324 self.source_x509_ca = cert
9326 src_instance_name = self.op.source_instance_name
9327 if not src_instance_name:
9328 raise errors.OpPrereqError("Missing source instance name",
9331 self.source_instance_name = \
9332 netutils.GetHostname(name=src_instance_name).name
9335 raise errors.OpPrereqError("Invalid instance creation mode %r" %
9336 self.op.mode, errors.ECODE_INVAL)
9338 def ExpandNames(self):
9339 """ExpandNames for CreateInstance.
9341 Figure out the right locks for instance creation.
9344 self.needed_locks = {}
9346 instance_name = self.op.instance_name
9347 # this is just a preventive check, but someone might still add this
9348 # instance in the meantime, and creation will fail at lock-add time
9349 if instance_name in self.cfg.GetInstanceList():
9350 raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
9351 instance_name, errors.ECODE_EXISTS)
9353 self.add_locks[locking.LEVEL_INSTANCE] = instance_name
9355 if self.op.iallocator:
9356 # TODO: Find a solution to not lock all nodes in the cluster, e.g. by
9357 # specifying a group on instance creation and then selecting nodes from
9359 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
9360 self.needed_locks[locking.LEVEL_NODE_RES] = locking.ALL_SET
9362 self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
9363 nodelist = [self.op.pnode]
9364 if self.op.snode is not None:
9365 self.op.snode = _ExpandNodeName(self.cfg, self.op.snode)
9366 nodelist.append(self.op.snode)
9367 self.needed_locks[locking.LEVEL_NODE] = nodelist
9368 # Lock resources of instance's primary and secondary nodes (copy to
9369 # prevent accidential modification)
9370 self.needed_locks[locking.LEVEL_NODE_RES] = list(nodelist)
9372 # in case of import lock the source node too
9373 if self.op.mode == constants.INSTANCE_IMPORT:
9374 src_node = self.op.src_node
9375 src_path = self.op.src_path
9377 if src_path is None:
9378 self.op.src_path = src_path = self.op.instance_name
9380 if src_node is None:
9381 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
9382 self.op.src_node = None
9383 if os.path.isabs(src_path):
9384 raise errors.OpPrereqError("Importing an instance from a path"
9385 " requires a source node option",
9388 self.op.src_node = src_node = _ExpandNodeName(self.cfg, src_node)
9389 if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
9390 self.needed_locks[locking.LEVEL_NODE].append(src_node)
9391 if not os.path.isabs(src_path):
9392 self.op.src_path = src_path = \
9393 utils.PathJoin(constants.EXPORT_DIR, src_path)
9395 def _RunAllocator(self):
9396 """Run the allocator based on input opcode.
9399 nics = [n.ToDict() for n in self.nics]
9400 ial = IAllocator(self.cfg, self.rpc,
9401 mode=constants.IALLOCATOR_MODE_ALLOC,
9402 name=self.op.instance_name,
9403 disk_template=self.op.disk_template,
9406 vcpus=self.be_full[constants.BE_VCPUS],
9407 memory=self.be_full[constants.BE_MAXMEM],
9408 spindle_use=self.be_full[constants.BE_SPINDLE_USE],
9411 hypervisor=self.op.hypervisor,
9414 ial.Run(self.op.iallocator)
9417 raise errors.OpPrereqError("Can't compute nodes using"
9418 " iallocator '%s': %s" %
9419 (self.op.iallocator, ial.info),
9421 if len(ial.result) != ial.required_nodes:
9422 raise errors.OpPrereqError("iallocator '%s' returned invalid number"
9423 " of nodes (%s), required %s" %
9424 (self.op.iallocator, len(ial.result),
9425 ial.required_nodes), errors.ECODE_FAULT)
9426 self.op.pnode = ial.result[0]
9427 self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
9428 self.op.instance_name, self.op.iallocator,
9429 utils.CommaJoin(ial.result))
9430 if ial.required_nodes == 2:
9431 self.op.snode = ial.result[1]
9433 def BuildHooksEnv(self):
9436 This runs on master, primary and secondary nodes of the instance.
9440 "ADD_MODE": self.op.mode,
9442 if self.op.mode == constants.INSTANCE_IMPORT:
9443 env["SRC_NODE"] = self.op.src_node
9444 env["SRC_PATH"] = self.op.src_path
9445 env["SRC_IMAGES"] = self.src_images
9447 env.update(_BuildInstanceHookEnv(
9448 name=self.op.instance_name,
9449 primary_node=self.op.pnode,
9450 secondary_nodes=self.secondaries,
9451 status=self.op.start,
9452 os_type=self.op.os_type,
9453 minmem=self.be_full[constants.BE_MINMEM],
9454 maxmem=self.be_full[constants.BE_MAXMEM],
9455 vcpus=self.be_full[constants.BE_VCPUS],
9456 nics=_NICListToTuple(self, self.nics),
9457 disk_template=self.op.disk_template,
9458 disks=[(d[constants.IDISK_SIZE], d[constants.IDISK_MODE])
9459 for d in self.disks],
9462 hypervisor_name=self.op.hypervisor,
9468 def BuildHooksNodes(self):
9469 """Build hooks nodes.
9472 nl = [self.cfg.GetMasterNode(), self.op.pnode] + self.secondaries
9475 def _ReadExportInfo(self):
9476 """Reads the export information from disk.
9478 It will override the opcode source node and path with the actual
9479 information, if these two were not specified before.
9481 @return: the export information
9484 assert self.op.mode == constants.INSTANCE_IMPORT
9486 src_node = self.op.src_node
9487 src_path = self.op.src_path
9489 if src_node is None:
9490 locked_nodes = self.owned_locks(locking.LEVEL_NODE)
9491 exp_list = self.rpc.call_export_list(locked_nodes)
9493 for node in exp_list:
9494 if exp_list[node].fail_msg:
9496 if src_path in exp_list[node].payload:
9498 self.op.src_node = src_node = node
9499 self.op.src_path = src_path = utils.PathJoin(constants.EXPORT_DIR,
9503 raise errors.OpPrereqError("No export found for relative path %s" %
9504 src_path, errors.ECODE_INVAL)
9506 _CheckNodeOnline(self, src_node)
9507 result = self.rpc.call_export_info(src_node, src_path)
9508 result.Raise("No export or invalid export found in dir %s" % src_path)
9510 export_info = objects.SerializableConfigParser.Loads(str(result.payload))
9511 if not export_info.has_section(constants.INISECT_EXP):
9512 raise errors.ProgrammerError("Corrupted export config",
9513 errors.ECODE_ENVIRON)
9515 ei_version = export_info.get(constants.INISECT_EXP, "version")
9516 if (int(ei_version) != constants.EXPORT_VERSION):
9517 raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
9518 (ei_version, constants.EXPORT_VERSION),
9519 errors.ECODE_ENVIRON)
9522 def _ReadExportParams(self, einfo):
9523 """Use export parameters as defaults.
9525 In case the opcode doesn't specify (as in override) some instance
9526 parameters, then try to use them from the export information, if
9530 self.op.os_type = einfo.get(constants.INISECT_EXP, "os")
9532 if self.op.disk_template is None:
9533 if einfo.has_option(constants.INISECT_INS, "disk_template"):
9534 self.op.disk_template = einfo.get(constants.INISECT_INS,
9536 if self.op.disk_template not in constants.DISK_TEMPLATES:
9537 raise errors.OpPrereqError("Disk template specified in configuration"
9538 " file is not one of the allowed values:"
9539 " %s" % " ".join(constants.DISK_TEMPLATES))
9541 raise errors.OpPrereqError("No disk template specified and the export"
9542 " is missing the disk_template information",
9545 if not self.op.disks:
9547 # TODO: import the disk iv_name too
9548 for idx in range(constants.MAX_DISKS):
9549 if einfo.has_option(constants.INISECT_INS, "disk%d_size" % idx):
9550 disk_sz = einfo.getint(constants.INISECT_INS, "disk%d_size" % idx)
9551 disks.append({constants.IDISK_SIZE: disk_sz})
9552 self.op.disks = disks
9553 if not disks and self.op.disk_template != constants.DT_DISKLESS:
9554 raise errors.OpPrereqError("No disk info specified and the export"
9555 " is missing the disk information",
9558 if not self.op.nics:
9560 for idx in range(constants.MAX_NICS):
9561 if einfo.has_option(constants.INISECT_INS, "nic%d_mac" % idx):
9563 for name in list(constants.NICS_PARAMETERS) + ["ip", "mac"]:
9564 v = einfo.get(constants.INISECT_INS, "nic%d_%s" % (idx, name))
9571 if not self.op.tags and einfo.has_option(constants.INISECT_INS, "tags"):
9572 self.op.tags = einfo.get(constants.INISECT_INS, "tags").split()
9574 if (self.op.hypervisor is None and
9575 einfo.has_option(constants.INISECT_INS, "hypervisor")):
9576 self.op.hypervisor = einfo.get(constants.INISECT_INS, "hypervisor")
9578 if einfo.has_section(constants.INISECT_HYP):
9579 # use the export parameters but do not override the ones
9580 # specified by the user
9581 for name, value in einfo.items(constants.INISECT_HYP):
9582 if name not in self.op.hvparams:
9583 self.op.hvparams[name] = value
9585 if einfo.has_section(constants.INISECT_BEP):
9586 # use the parameters, without overriding
9587 for name, value in einfo.items(constants.INISECT_BEP):
9588 if name not in self.op.beparams:
9589 self.op.beparams[name] = value
9590 # Compatibility for the old "memory" be param
9591 if name == constants.BE_MEMORY:
9592 if constants.BE_MAXMEM not in self.op.beparams:
9593 self.op.beparams[constants.BE_MAXMEM] = value
9594 if constants.BE_MINMEM not in self.op.beparams:
9595 self.op.beparams[constants.BE_MINMEM] = value
9597 # try to read the parameters old style, from the main section
9598 for name in constants.BES_PARAMETERS:
9599 if (name not in self.op.beparams and
9600 einfo.has_option(constants.INISECT_INS, name)):
9601 self.op.beparams[name] = einfo.get(constants.INISECT_INS, name)
9603 if einfo.has_section(constants.INISECT_OSP):
9604 # use the parameters, without overriding
9605 for name, value in einfo.items(constants.INISECT_OSP):
9606 if name not in self.op.osparams:
9607 self.op.osparams[name] = value
9609 def _RevertToDefaults(self, cluster):
9610 """Revert the instance parameters to the default values.
9614 hv_defs = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type, {})
9615 for name in self.op.hvparams.keys():
9616 if name in hv_defs and hv_defs[name] == self.op.hvparams[name]:
9617 del self.op.hvparams[name]
9619 be_defs = cluster.SimpleFillBE({})
9620 for name in self.op.beparams.keys():
9621 if name in be_defs and be_defs[name] == self.op.beparams[name]:
9622 del self.op.beparams[name]
9624 nic_defs = cluster.SimpleFillNIC({})
9625 for nic in self.op.nics:
9626 for name in constants.NICS_PARAMETERS:
9627 if name in nic and name in nic_defs and nic[name] == nic_defs[name]:
9630 os_defs = cluster.SimpleFillOS(self.op.os_type, {})
9631 for name in self.op.osparams.keys():
9632 if name in os_defs and os_defs[name] == self.op.osparams[name]:
9633 del self.op.osparams[name]
9635 def _CalculateFileStorageDir(self):
9636 """Calculate final instance file storage dir.
9639 # file storage dir calculation/check
9640 self.instance_file_storage_dir = None
9641 if self.op.disk_template in constants.DTS_FILEBASED:
9642 # build the full file storage dir path
9645 if self.op.disk_template == constants.DT_SHARED_FILE:
9646 get_fsd_fn = self.cfg.GetSharedFileStorageDir
9648 get_fsd_fn = self.cfg.GetFileStorageDir
9650 cfg_storagedir = get_fsd_fn()
9651 if not cfg_storagedir:
9652 raise errors.OpPrereqError("Cluster file storage dir not defined")
9653 joinargs.append(cfg_storagedir)
9655 if self.op.file_storage_dir is not None:
9656 joinargs.append(self.op.file_storage_dir)
9658 joinargs.append(self.op.instance_name)
9660 # pylint: disable=W0142
9661 self.instance_file_storage_dir = utils.PathJoin(*joinargs)
9663 def CheckPrereq(self): # pylint: disable=R0914
9664 """Check prerequisites.
9667 self._CalculateFileStorageDir()
9669 if self.op.mode == constants.INSTANCE_IMPORT:
9670 export_info = self._ReadExportInfo()
9671 self._ReadExportParams(export_info)
9672 self._old_instance_name = export_info.get(constants.INISECT_INS, "name")
9674 self._old_instance_name = None
9676 if (not self.cfg.GetVGName() and
9677 self.op.disk_template not in constants.DTS_NOT_LVM):
9678 raise errors.OpPrereqError("Cluster does not support lvm-based"
9679 " instances", errors.ECODE_STATE)
9681 if (self.op.hypervisor is None or
9682 self.op.hypervisor == constants.VALUE_AUTO):
9683 self.op.hypervisor = self.cfg.GetHypervisorType()
9685 cluster = self.cfg.GetClusterInfo()
9686 enabled_hvs = cluster.enabled_hypervisors
9687 if self.op.hypervisor not in enabled_hvs:
9688 raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
9689 " cluster (%s)" % (self.op.hypervisor,
9690 ",".join(enabled_hvs)),
9693 # Check tag validity
9694 for tag in self.op.tags:
9695 objects.TaggableObject.ValidateTag(tag)
9697 # check hypervisor parameter syntax (locally)
9698 utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
9699 filled_hvp = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type,
9701 hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
9702 hv_type.CheckParameterSyntax(filled_hvp)
9703 self.hv_full = filled_hvp
9704 # check that we don't specify global parameters on an instance
9705 _CheckGlobalHvParams(self.op.hvparams)
9707 # fill and remember the beparams dict
9708 default_beparams = cluster.beparams[constants.PP_DEFAULT]
9709 for param, value in self.op.beparams.iteritems():
9710 if value == constants.VALUE_AUTO:
9711 self.op.beparams[param] = default_beparams[param]
9712 objects.UpgradeBeParams(self.op.beparams)
9713 utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
9714 self.be_full = cluster.SimpleFillBE(self.op.beparams)
9716 # build os parameters
9717 self.os_full = cluster.SimpleFillOS(self.op.os_type, self.op.osparams)
9719 # now that hvp/bep are in final format, let's reset to defaults,
9721 if self.op.identify_defaults:
9722 self._RevertToDefaults(cluster)
9726 for idx, nic in enumerate(self.op.nics):
9727 nic_mode_req = nic.get(constants.INIC_MODE, None)
9728 nic_mode = nic_mode_req
9729 if nic_mode is None or nic_mode == constants.VALUE_AUTO:
9730 nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
9732 # in routed mode, for the first nic, the default ip is 'auto'
9733 if nic_mode == constants.NIC_MODE_ROUTED and idx == 0:
9734 default_ip_mode = constants.VALUE_AUTO
9736 default_ip_mode = constants.VALUE_NONE
9738 # ip validity checks
9739 ip = nic.get(constants.INIC_IP, default_ip_mode)
9740 if ip is None or ip.lower() == constants.VALUE_NONE:
9742 elif ip.lower() == constants.VALUE_AUTO:
9743 if not self.op.name_check:
9744 raise errors.OpPrereqError("IP address set to auto but name checks"
9745 " have been skipped",
9747 nic_ip = self.hostname1.ip
9749 if not netutils.IPAddress.IsValid(ip):
9750 raise errors.OpPrereqError("Invalid IP address '%s'" % ip,
9754 # TODO: check the ip address for uniqueness
9755 if nic_mode == constants.NIC_MODE_ROUTED and not nic_ip:
9756 raise errors.OpPrereqError("Routed nic mode requires an ip address",
9759 # MAC address verification
9760 mac = nic.get(constants.INIC_MAC, constants.VALUE_AUTO)
9761 if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
9762 mac = utils.NormalizeAndValidateMac(mac)
9765 self.cfg.ReserveMAC(mac, self.proc.GetECId())
9766 except errors.ReservationError:
9767 raise errors.OpPrereqError("MAC address %s already in use"
9768 " in cluster" % mac,
9769 errors.ECODE_NOTUNIQUE)
9771 # Build nic parameters
9772 link = nic.get(constants.INIC_LINK, None)
9773 if link == constants.VALUE_AUTO:
9774 link = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_LINK]
9777 nicparams[constants.NIC_MODE] = nic_mode
9779 nicparams[constants.NIC_LINK] = link
9781 check_params = cluster.SimpleFillNIC(nicparams)
9782 objects.NIC.CheckParameterSyntax(check_params)
9783 self.nics.append(objects.NIC(mac=mac, ip=nic_ip, nicparams=nicparams))
9785 # disk checks/pre-build
9786 default_vg = self.cfg.GetVGName()
9788 for disk in self.op.disks:
9789 mode = disk.get(constants.IDISK_MODE, constants.DISK_RDWR)
9790 if mode not in constants.DISK_ACCESS_SET:
9791 raise errors.OpPrereqError("Invalid disk access mode '%s'" %
9792 mode, errors.ECODE_INVAL)
9793 size = disk.get(constants.IDISK_SIZE, None)
9795 raise errors.OpPrereqError("Missing disk size", errors.ECODE_INVAL)
9798 except (TypeError, ValueError):
9799 raise errors.OpPrereqError("Invalid disk size '%s'" % size,
9802 data_vg = disk.get(constants.IDISK_VG, default_vg)
9804 constants.IDISK_SIZE: size,
9805 constants.IDISK_MODE: mode,
9806 constants.IDISK_VG: data_vg,
9808 if constants.IDISK_METAVG in disk:
9809 new_disk[constants.IDISK_METAVG] = disk[constants.IDISK_METAVG]
9810 if constants.IDISK_ADOPT in disk:
9811 new_disk[constants.IDISK_ADOPT] = disk[constants.IDISK_ADOPT]
9812 self.disks.append(new_disk)
9814 if self.op.mode == constants.INSTANCE_IMPORT:
9816 for idx in range(len(self.disks)):
9817 option = "disk%d_dump" % idx
9818 if export_info.has_option(constants.INISECT_INS, option):
9819 # FIXME: are the old os-es, disk sizes, etc. useful?
9820 export_name = export_info.get(constants.INISECT_INS, option)
9821 image = utils.PathJoin(self.op.src_path, export_name)
9822 disk_images.append(image)
9824 disk_images.append(False)
9826 self.src_images = disk_images
9828 if self.op.instance_name == self._old_instance_name:
9829 for idx, nic in enumerate(self.nics):
9830 if nic.mac == constants.VALUE_AUTO:
9831 nic_mac_ini = "nic%d_mac" % idx
9832 nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
9834 # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
9836 # ip ping checks (we use the same ip that was resolved in ExpandNames)
9837 if self.op.ip_check:
9838 if netutils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
9839 raise errors.OpPrereqError("IP %s of instance %s already in use" %
9840 (self.check_ip, self.op.instance_name),
9841 errors.ECODE_NOTUNIQUE)
9843 #### mac address generation
9844 # By generating here the mac address both the allocator and the hooks get
9845 # the real final mac address rather than the 'auto' or 'generate' value.
9846 # There is a race condition between the generation and the instance object
9847 # creation, which means that we know the mac is valid now, but we're not
9848 # sure it will be when we actually add the instance. If things go bad
9849 # adding the instance will abort because of a duplicate mac, and the
9850 # creation job will fail.
9851 for nic in self.nics:
9852 if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
9853 nic.mac = self.cfg.GenerateMAC(self.proc.GetECId())
9857 if self.op.iallocator is not None:
9858 self._RunAllocator()
9860 # Release all unneeded node locks
9861 _ReleaseLocks(self, locking.LEVEL_NODE,
9862 keep=filter(None, [self.op.pnode, self.op.snode,
9864 _ReleaseLocks(self, locking.LEVEL_NODE_RES,
9865 keep=filter(None, [self.op.pnode, self.op.snode,
9868 #### node related checks
9870 # check primary node
9871 self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
9872 assert self.pnode is not None, \
9873 "Cannot retrieve locked node %s" % self.op.pnode
9875 raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
9876 pnode.name, errors.ECODE_STATE)
9878 raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
9879 pnode.name, errors.ECODE_STATE)
9880 if not pnode.vm_capable:
9881 raise errors.OpPrereqError("Cannot use non-vm_capable primary node"
9882 " '%s'" % pnode.name, errors.ECODE_STATE)
9884 self.secondaries = []
9886 # mirror node verification
9887 if self.op.disk_template in constants.DTS_INT_MIRROR:
9888 if self.op.snode == pnode.name:
9889 raise errors.OpPrereqError("The secondary node cannot be the"
9890 " primary node", errors.ECODE_INVAL)
9891 _CheckNodeOnline(self, self.op.snode)
9892 _CheckNodeNotDrained(self, self.op.snode)
9893 _CheckNodeVmCapable(self, self.op.snode)
9894 self.secondaries.append(self.op.snode)
9896 snode = self.cfg.GetNodeInfo(self.op.snode)
9897 if pnode.group != snode.group:
9898 self.LogWarning("The primary and secondary nodes are in two"
9899 " different node groups; the disk parameters"
9900 " from the first disk's node group will be"
9903 nodenames = [pnode.name] + self.secondaries
9905 # Verify instance specs
9906 spindle_use = self.be_full.get(constants.BE_SPINDLE_USE, None)
9908 constants.ISPEC_MEM_SIZE: self.be_full.get(constants.BE_MAXMEM, None),
9909 constants.ISPEC_CPU_COUNT: self.be_full.get(constants.BE_VCPUS, None),
9910 constants.ISPEC_DISK_COUNT: len(self.disks),
9911 constants.ISPEC_DISK_SIZE: [disk["size"] for disk in self.disks],
9912 constants.ISPEC_NIC_COUNT: len(self.nics),
9913 constants.ISPEC_SPINDLE_USE: spindle_use,
9916 group_info = self.cfg.GetNodeGroup(pnode.group)
9917 ipolicy = _CalculateGroupIPolicy(cluster, group_info)
9918 res = _ComputeIPolicyInstanceSpecViolation(ipolicy, ispec)
9919 if not self.op.ignore_ipolicy and res:
9920 raise errors.OpPrereqError(("Instance allocation to group %s violates"
9921 " policy: %s") % (pnode.group,
9922 utils.CommaJoin(res)),
9925 if not self.adopt_disks:
9926 if self.op.disk_template == constants.DT_RBD:
9927 # _CheckRADOSFreeSpace() is just a placeholder.
9928 # Any function that checks prerequisites can be placed here.
9929 # Check if there is enough space on the RADOS cluster.
9930 _CheckRADOSFreeSpace()
9932 # Check lv size requirements, if not adopting
9933 req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks)
9934 _CheckNodesFreeDiskPerVG(self, nodenames, req_sizes)
9936 elif self.op.disk_template == constants.DT_PLAIN: # Check the adoption data
9937 all_lvs = set(["%s/%s" % (disk[constants.IDISK_VG],
9938 disk[constants.IDISK_ADOPT])
9939 for disk in self.disks])
9940 if len(all_lvs) != len(self.disks):
9941 raise errors.OpPrereqError("Duplicate volume names given for adoption",
9943 for lv_name in all_lvs:
9945 # FIXME: lv_name here is "vg/lv" need to ensure that other calls
9946 # to ReserveLV uses the same syntax
9947 self.cfg.ReserveLV(lv_name, self.proc.GetECId())
9948 except errors.ReservationError:
9949 raise errors.OpPrereqError("LV named %s used by another instance" %
9950 lv_name, errors.ECODE_NOTUNIQUE)
9952 vg_names = self.rpc.call_vg_list([pnode.name])[pnode.name]
9953 vg_names.Raise("Cannot get VG information from node %s" % pnode.name)
9955 node_lvs = self.rpc.call_lv_list([pnode.name],
9956 vg_names.payload.keys())[pnode.name]
9957 node_lvs.Raise("Cannot get LV information from node %s" % pnode.name)
9958 node_lvs = node_lvs.payload
9960 delta = all_lvs.difference(node_lvs.keys())
9962 raise errors.OpPrereqError("Missing logical volume(s): %s" %
9963 utils.CommaJoin(delta),
9965 online_lvs = [lv for lv in all_lvs if node_lvs[lv][2]]
9967 raise errors.OpPrereqError("Online logical volumes found, cannot"
9968 " adopt: %s" % utils.CommaJoin(online_lvs),
9970 # update the size of disk based on what is found
9971 for dsk in self.disks:
9972 dsk[constants.IDISK_SIZE] = \
9973 int(float(node_lvs["%s/%s" % (dsk[constants.IDISK_VG],
9974 dsk[constants.IDISK_ADOPT])][0]))
9976 elif self.op.disk_template == constants.DT_BLOCK:
9977 # Normalize and de-duplicate device paths
9978 all_disks = set([os.path.abspath(disk[constants.IDISK_ADOPT])
9979 for disk in self.disks])
9980 if len(all_disks) != len(self.disks):
9981 raise errors.OpPrereqError("Duplicate disk names given for adoption",
9983 baddisks = [d for d in all_disks
9984 if not d.startswith(constants.ADOPTABLE_BLOCKDEV_ROOT)]
9986 raise errors.OpPrereqError("Device node(s) %s lie outside %s and"
9987 " cannot be adopted" %
9988 (", ".join(baddisks),
9989 constants.ADOPTABLE_BLOCKDEV_ROOT),
9992 node_disks = self.rpc.call_bdev_sizes([pnode.name],
9993 list(all_disks))[pnode.name]
9994 node_disks.Raise("Cannot get block device information from node %s" %
9996 node_disks = node_disks.payload
9997 delta = all_disks.difference(node_disks.keys())
9999 raise errors.OpPrereqError("Missing block device(s): %s" %
10000 utils.CommaJoin(delta),
10001 errors.ECODE_INVAL)
10002 for dsk in self.disks:
10003 dsk[constants.IDISK_SIZE] = \
10004 int(float(node_disks[dsk[constants.IDISK_ADOPT]]))
10006 _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
10008 _CheckNodeHasOS(self, pnode.name, self.op.os_type, self.op.force_variant)
10009 # check OS parameters (remotely)
10010 _CheckOSParams(self, True, nodenames, self.op.os_type, self.os_full)
10012 _CheckNicsBridgesExist(self, self.nics, self.pnode.name)
10014 # memory check on primary node
10015 #TODO(dynmem): use MINMEM for checking
10017 _CheckNodeFreeMemory(self, self.pnode.name,
10018 "creating instance %s" % self.op.instance_name,
10019 self.be_full[constants.BE_MAXMEM],
10020 self.op.hypervisor)
10022 self.dry_run_result = list(nodenames)
10024 def Exec(self, feedback_fn):
10025 """Create and add the instance to the cluster.
10028 instance = self.op.instance_name
10029 pnode_name = self.pnode.name
10031 assert not (self.owned_locks(locking.LEVEL_NODE_RES) -
10032 self.owned_locks(locking.LEVEL_NODE)), \
10033 "Node locks differ from node resource locks"
10035 ht_kind = self.op.hypervisor
10036 if ht_kind in constants.HTS_REQ_PORT:
10037 network_port = self.cfg.AllocatePort()
10039 network_port = None
10041 # This is ugly but we got a chicken-egg problem here
10042 # We can only take the group disk parameters, as the instance
10043 # has no disks yet (we are generating them right here).
10044 node = self.cfg.GetNodeInfo(pnode_name)
10045 nodegroup = self.cfg.GetNodeGroup(node.group)
10046 disks = _GenerateDiskTemplate(self,
10047 self.op.disk_template,
10048 instance, pnode_name,
10051 self.instance_file_storage_dir,
10052 self.op.file_driver,
10055 self.cfg.GetGroupDiskParams(nodegroup))
10057 iobj = objects.Instance(name=instance, os=self.op.os_type,
10058 primary_node=pnode_name,
10059 nics=self.nics, disks=disks,
10060 disk_template=self.op.disk_template,
10061 admin_state=constants.ADMINST_DOWN,
10062 network_port=network_port,
10063 beparams=self.op.beparams,
10064 hvparams=self.op.hvparams,
10065 hypervisor=self.op.hypervisor,
10066 osparams=self.op.osparams,
10070 for tag in self.op.tags:
10073 if self.adopt_disks:
10074 if self.op.disk_template == constants.DT_PLAIN:
10075 # rename LVs to the newly-generated names; we need to construct
10076 # 'fake' LV disks with the old data, plus the new unique_id
10077 tmp_disks = [objects.Disk.FromDict(v.ToDict()) for v in disks]
10079 for t_dsk, a_dsk in zip(tmp_disks, self.disks):
10080 rename_to.append(t_dsk.logical_id)
10081 t_dsk.logical_id = (t_dsk.logical_id[0], a_dsk[constants.IDISK_ADOPT])
10082 self.cfg.SetDiskID(t_dsk, pnode_name)
10083 result = self.rpc.call_blockdev_rename(pnode_name,
10084 zip(tmp_disks, rename_to))
10085 result.Raise("Failed to rename adoped LVs")
10087 feedback_fn("* creating instance disks...")
10089 _CreateDisks(self, iobj)
10090 except errors.OpExecError:
10091 self.LogWarning("Device creation failed, reverting...")
10093 _RemoveDisks(self, iobj)
10095 self.cfg.ReleaseDRBDMinors(instance)
10098 feedback_fn("adding instance %s to cluster config" % instance)
10100 self.cfg.AddInstance(iobj, self.proc.GetECId())
10102 # Declare that we don't want to remove the instance lock anymore, as we've
10103 # added the instance to the config
10104 del self.remove_locks[locking.LEVEL_INSTANCE]
10106 if self.op.mode == constants.INSTANCE_IMPORT:
10107 # Release unused nodes
10108 _ReleaseLocks(self, locking.LEVEL_NODE, keep=[self.op.src_node])
10110 # Release all nodes
10111 _ReleaseLocks(self, locking.LEVEL_NODE)
10114 if not self.adopt_disks and self.cfg.GetClusterInfo().prealloc_wipe_disks:
10115 feedback_fn("* wiping instance disks...")
10117 _WipeDisks(self, iobj)
10118 except errors.OpExecError, err:
10119 logging.exception("Wiping disks failed")
10120 self.LogWarning("Wiping instance disks failed (%s)", err)
10124 # Something is already wrong with the disks, don't do anything else
10126 elif self.op.wait_for_sync:
10127 disk_abort = not _WaitForSync(self, iobj)
10128 elif iobj.disk_template in constants.DTS_INT_MIRROR:
10129 # make sure the disks are not degraded (still sync-ing is ok)
10130 feedback_fn("* checking mirrors status")
10131 disk_abort = not _WaitForSync(self, iobj, oneshot=True)
10136 _RemoveDisks(self, iobj)
10137 self.cfg.RemoveInstance(iobj.name)
10138 # Make sure the instance lock gets removed
10139 self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
10140 raise errors.OpExecError("There are some degraded disks for"
10143 # Release all node resource locks
10144 _ReleaseLocks(self, locking.LEVEL_NODE_RES)
10146 if iobj.disk_template != constants.DT_DISKLESS and not self.adopt_disks:
10147 # we need to set the disks ID to the primary node, since the
10148 # preceding code might or might have not done it, depending on
10149 # disk template and other options
10150 for disk in iobj.disks:
10151 self.cfg.SetDiskID(disk, pnode_name)
10152 if self.op.mode == constants.INSTANCE_CREATE:
10153 if not self.op.no_install:
10154 pause_sync = (iobj.disk_template in constants.DTS_INT_MIRROR and
10155 not self.op.wait_for_sync)
10157 feedback_fn("* pausing disk sync to install instance OS")
10158 result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
10161 for idx, success in enumerate(result.payload):
10163 logging.warn("pause-sync of instance %s for disk %d failed",
10166 feedback_fn("* running the instance OS create scripts...")
10167 # FIXME: pass debug option from opcode to backend
10169 self.rpc.call_instance_os_add(pnode_name, (iobj, None), False,
10170 self.op.debug_level)
10172 feedback_fn("* resuming disk sync")
10173 result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
10176 for idx, success in enumerate(result.payload):
10178 logging.warn("resume-sync of instance %s for disk %d failed",
10181 os_add_result.Raise("Could not add os for instance %s"
10182 " on node %s" % (instance, pnode_name))
10185 if self.op.mode == constants.INSTANCE_IMPORT:
10186 feedback_fn("* running the instance OS import scripts...")
10190 for idx, image in enumerate(self.src_images):
10194 # FIXME: pass debug option from opcode to backend
10195 dt = masterd.instance.DiskTransfer("disk/%s" % idx,
10196 constants.IEIO_FILE, (image, ),
10197 constants.IEIO_SCRIPT,
10198 (iobj.disks[idx], idx),
10200 transfers.append(dt)
10203 masterd.instance.TransferInstanceData(self, feedback_fn,
10204 self.op.src_node, pnode_name,
10205 self.pnode.secondary_ip,
10207 if not compat.all(import_result):
10208 self.LogWarning("Some disks for instance %s on node %s were not"
10209 " imported successfully" % (instance, pnode_name))
10211 rename_from = self._old_instance_name
10213 elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
10214 feedback_fn("* preparing remote import...")
10215 # The source cluster will stop the instance before attempting to make
10216 # a connection. In some cases stopping an instance can take a long
10217 # time, hence the shutdown timeout is added to the connection
10219 connect_timeout = (constants.RIE_CONNECT_TIMEOUT +
10220 self.op.source_shutdown_timeout)
10221 timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
10223 assert iobj.primary_node == self.pnode.name
10225 masterd.instance.RemoteImport(self, feedback_fn, iobj, self.pnode,
10226 self.source_x509_ca,
10227 self._cds, timeouts)
10228 if not compat.all(disk_results):
10229 # TODO: Should the instance still be started, even if some disks
10230 # failed to import (valid for local imports, too)?
10231 self.LogWarning("Some disks for instance %s on node %s were not"
10232 " imported successfully" % (instance, pnode_name))
10234 rename_from = self.source_instance_name
10237 # also checked in the prereq part
10238 raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
10241 # Run rename script on newly imported instance
10242 assert iobj.name == instance
10243 feedback_fn("Running rename script for %s" % instance)
10244 result = self.rpc.call_instance_run_rename(pnode_name, iobj,
10246 self.op.debug_level)
10247 if result.fail_msg:
10248 self.LogWarning("Failed to run rename script for %s on node"
10249 " %s: %s" % (instance, pnode_name, result.fail_msg))
10251 assert not self.owned_locks(locking.LEVEL_NODE_RES)
10254 iobj.admin_state = constants.ADMINST_UP
10255 self.cfg.Update(iobj, feedback_fn)
10256 logging.info("Starting instance %s on node %s", instance, pnode_name)
10257 feedback_fn("* starting instance...")
10258 result = self.rpc.call_instance_start(pnode_name, (iobj, None, None),
10260 result.Raise("Could not start instance")
10262 return list(iobj.all_nodes)
10265 def _CheckRADOSFreeSpace():
10266 """Compute disk size requirements inside the RADOS cluster.
10269 # For the RADOS cluster we assume there is always enough space.
10273 class LUInstanceConsole(NoHooksLU):
10274 """Connect to an instance's console.
10276 This is somewhat special in that it returns the command line that
10277 you need to run on the master node in order to connect to the
10283 def ExpandNames(self):
10284 self.share_locks = _ShareAll()
10285 self._ExpandAndLockInstance()
10287 def CheckPrereq(self):
10288 """Check prerequisites.
10290 This checks that the instance is in the cluster.
10293 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
10294 assert self.instance is not None, \
10295 "Cannot retrieve locked instance %s" % self.op.instance_name
10296 _CheckNodeOnline(self, self.instance.primary_node)
10298 def Exec(self, feedback_fn):
10299 """Connect to the console of an instance
10302 instance = self.instance
10303 node = instance.primary_node
10305 node_insts = self.rpc.call_instance_list([node],
10306 [instance.hypervisor])[node]
10307 node_insts.Raise("Can't get node information from %s" % node)
10309 if instance.name not in node_insts.payload:
10310 if instance.admin_state == constants.ADMINST_UP:
10311 state = constants.INSTST_ERRORDOWN
10312 elif instance.admin_state == constants.ADMINST_DOWN:
10313 state = constants.INSTST_ADMINDOWN
10315 state = constants.INSTST_ADMINOFFLINE
10316 raise errors.OpExecError("Instance %s is not running (state %s)" %
10317 (instance.name, state))
10319 logging.debug("Connecting to console of %s on %s", instance.name, node)
10321 return _GetInstanceConsole(self.cfg.GetClusterInfo(), instance)
10324 def _GetInstanceConsole(cluster, instance):
10325 """Returns console information for an instance.
10327 @type cluster: L{objects.Cluster}
10328 @type instance: L{objects.Instance}
10332 hyper = hypervisor.GetHypervisor(instance.hypervisor)
10333 # beparams and hvparams are passed separately, to avoid editing the
10334 # instance and then saving the defaults in the instance itself.
10335 hvparams = cluster.FillHV(instance)
10336 beparams = cluster.FillBE(instance)
10337 console = hyper.GetInstanceConsole(instance, hvparams, beparams)
10339 assert console.instance == instance.name
10340 assert console.Validate()
10342 return console.ToDict()
10345 class LUInstanceReplaceDisks(LogicalUnit):
10346 """Replace the disks of an instance.
10349 HPATH = "mirrors-replace"
10350 HTYPE = constants.HTYPE_INSTANCE
10353 def CheckArguments(self):
10354 TLReplaceDisks.CheckArguments(self.op.mode, self.op.remote_node,
10355 self.op.iallocator)
10357 def ExpandNames(self):
10358 self._ExpandAndLockInstance()
10360 assert locking.LEVEL_NODE not in self.needed_locks
10361 assert locking.LEVEL_NODE_RES not in self.needed_locks
10362 assert locking.LEVEL_NODEGROUP not in self.needed_locks
10364 assert self.op.iallocator is None or self.op.remote_node is None, \
10365 "Conflicting options"
10367 if self.op.remote_node is not None:
10368 self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
10370 # Warning: do not remove the locking of the new secondary here
10371 # unless DRBD8.AddChildren is changed to work in parallel;
10372 # currently it doesn't since parallel invocations of
10373 # FindUnusedMinor will conflict
10374 self.needed_locks[locking.LEVEL_NODE] = [self.op.remote_node]
10375 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
10377 self.needed_locks[locking.LEVEL_NODE] = []
10378 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
10380 if self.op.iallocator is not None:
10381 # iallocator will select a new node in the same group
10382 self.needed_locks[locking.LEVEL_NODEGROUP] = []
10384 self.needed_locks[locking.LEVEL_NODE_RES] = []
10386 self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
10387 self.op.iallocator, self.op.remote_node,
10388 self.op.disks, False, self.op.early_release,
10389 self.op.ignore_ipolicy)
10391 self.tasklets = [self.replacer]
10393 def DeclareLocks(self, level):
10394 if level == locking.LEVEL_NODEGROUP:
10395 assert self.op.remote_node is None
10396 assert self.op.iallocator is not None
10397 assert not self.needed_locks[locking.LEVEL_NODEGROUP]
10399 self.share_locks[locking.LEVEL_NODEGROUP] = 1
10400 # Lock all groups used by instance optimistically; this requires going
10401 # via the node before it's locked, requiring verification later on
10402 self.needed_locks[locking.LEVEL_NODEGROUP] = \
10403 self.cfg.GetInstanceNodeGroups(self.op.instance_name)
10405 elif level == locking.LEVEL_NODE:
10406 if self.op.iallocator is not None:
10407 assert self.op.remote_node is None
10408 assert not self.needed_locks[locking.LEVEL_NODE]
10410 # Lock member nodes of all locked groups
10411 self.needed_locks[locking.LEVEL_NODE] = [node_name
10412 for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
10413 for node_name in self.cfg.GetNodeGroup(group_uuid).members]
10415 self._LockInstancesNodes()
10416 elif level == locking.LEVEL_NODE_RES:
10418 self.needed_locks[locking.LEVEL_NODE_RES] = \
10419 self.needed_locks[locking.LEVEL_NODE]
10421 def BuildHooksEnv(self):
10422 """Build hooks env.
10424 This runs on the master, the primary and all the secondaries.
10427 instance = self.replacer.instance
10429 "MODE": self.op.mode,
10430 "NEW_SECONDARY": self.op.remote_node,
10431 "OLD_SECONDARY": instance.secondary_nodes[0],
10433 env.update(_BuildInstanceHookEnvByObject(self, instance))
10436 def BuildHooksNodes(self):
10437 """Build hooks nodes.
10440 instance = self.replacer.instance
10442 self.cfg.GetMasterNode(),
10443 instance.primary_node,
10445 if self.op.remote_node is not None:
10446 nl.append(self.op.remote_node)
10449 def CheckPrereq(self):
10450 """Check prerequisites.
10453 assert (self.glm.is_owned(locking.LEVEL_NODEGROUP) or
10454 self.op.iallocator is None)
10456 # Verify if node group locks are still correct
10457 owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
10459 _CheckInstanceNodeGroups(self.cfg, self.op.instance_name, owned_groups)
10461 return LogicalUnit.CheckPrereq(self)
10464 class TLReplaceDisks(Tasklet):
10465 """Replaces disks for an instance.
10467 Note: Locking is not within the scope of this class.
10470 def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
10471 disks, delay_iallocator, early_release, ignore_ipolicy):
10472 """Initializes this class.
10475 Tasklet.__init__(self, lu)
10478 self.instance_name = instance_name
10480 self.iallocator_name = iallocator_name
10481 self.remote_node = remote_node
10483 self.delay_iallocator = delay_iallocator
10484 self.early_release = early_release
10485 self.ignore_ipolicy = ignore_ipolicy
10488 self.instance = None
10489 self.new_node = None
10490 self.target_node = None
10491 self.other_node = None
10492 self.remote_node_info = None
10493 self.node_secondary_ip = None
10496 def CheckArguments(mode, remote_node, iallocator):
10497 """Helper function for users of this class.
10500 # check for valid parameter combination
10501 if mode == constants.REPLACE_DISK_CHG:
10502 if remote_node is None and iallocator is None:
10503 raise errors.OpPrereqError("When changing the secondary either an"
10504 " iallocator script must be used or the"
10505 " new node given", errors.ECODE_INVAL)
10507 if remote_node is not None and iallocator is not None:
10508 raise errors.OpPrereqError("Give either the iallocator or the new"
10509 " secondary, not both", errors.ECODE_INVAL)
10511 elif remote_node is not None or iallocator is not None:
10512 # Not replacing the secondary
10513 raise errors.OpPrereqError("The iallocator and new node options can"
10514 " only be used when changing the"
10515 " secondary node", errors.ECODE_INVAL)
10518 def _RunAllocator(lu, iallocator_name, instance_name, relocate_from):
10519 """Compute a new secondary node using an IAllocator.
10522 ial = IAllocator(lu.cfg, lu.rpc,
10523 mode=constants.IALLOCATOR_MODE_RELOC,
10524 name=instance_name,
10525 relocate_from=list(relocate_from))
10527 ial.Run(iallocator_name)
10529 if not ial.success:
10530 raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
10531 " %s" % (iallocator_name, ial.info),
10532 errors.ECODE_NORES)
10534 if len(ial.result) != ial.required_nodes:
10535 raise errors.OpPrereqError("iallocator '%s' returned invalid number"
10536 " of nodes (%s), required %s" %
10538 len(ial.result), ial.required_nodes),
10539 errors.ECODE_FAULT)
10541 remote_node_name = ial.result[0]
10543 lu.LogInfo("Selected new secondary for instance '%s': %s",
10544 instance_name, remote_node_name)
10546 return remote_node_name
10548 def _FindFaultyDisks(self, node_name):
10549 """Wrapper for L{_FindFaultyInstanceDisks}.
10552 return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
10555 def _CheckDisksActivated(self, instance):
10556 """Checks if the instance disks are activated.
10558 @param instance: The instance to check disks
10559 @return: True if they are activated, False otherwise
10562 nodes = instance.all_nodes
10564 for idx, dev in enumerate(instance.disks):
10566 self.lu.LogInfo("Checking disk/%d on %s", idx, node)
10567 self.cfg.SetDiskID(dev, node)
10569 result = _BlockdevFind(self, node, dev, instance)
10573 elif result.fail_msg or not result.payload:
10578 def CheckPrereq(self):
10579 """Check prerequisites.
10581 This checks that the instance is in the cluster.
10584 self.instance = instance = self.cfg.GetInstanceInfo(self.instance_name)
10585 assert instance is not None, \
10586 "Cannot retrieve locked instance %s" % self.instance_name
10588 if instance.disk_template != constants.DT_DRBD8:
10589 raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
10590 " instances", errors.ECODE_INVAL)
10592 if len(instance.secondary_nodes) != 1:
10593 raise errors.OpPrereqError("The instance has a strange layout,"
10594 " expected one secondary but found %d" %
10595 len(instance.secondary_nodes),
10596 errors.ECODE_FAULT)
10598 if not self.delay_iallocator:
10599 self._CheckPrereq2()
10601 def _CheckPrereq2(self):
10602 """Check prerequisites, second part.
10604 This function should always be part of CheckPrereq. It was separated and is
10605 now called from Exec because during node evacuation iallocator was only
10606 called with an unmodified cluster model, not taking planned changes into
10610 instance = self.instance
10611 secondary_node = instance.secondary_nodes[0]
10613 if self.iallocator_name is None:
10614 remote_node = self.remote_node
10616 remote_node = self._RunAllocator(self.lu, self.iallocator_name,
10617 instance.name, instance.secondary_nodes)
10619 if remote_node is None:
10620 self.remote_node_info = None
10622 assert remote_node in self.lu.owned_locks(locking.LEVEL_NODE), \
10623 "Remote node '%s' is not locked" % remote_node
10625 self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
10626 assert self.remote_node_info is not None, \
10627 "Cannot retrieve locked node %s" % remote_node
10629 if remote_node == self.instance.primary_node:
10630 raise errors.OpPrereqError("The specified node is the primary node of"
10631 " the instance", errors.ECODE_INVAL)
10633 if remote_node == secondary_node:
10634 raise errors.OpPrereqError("The specified node is already the"
10635 " secondary node of the instance",
10636 errors.ECODE_INVAL)
10638 if self.disks and self.mode in (constants.REPLACE_DISK_AUTO,
10639 constants.REPLACE_DISK_CHG):
10640 raise errors.OpPrereqError("Cannot specify disks to be replaced",
10641 errors.ECODE_INVAL)
10643 if self.mode == constants.REPLACE_DISK_AUTO:
10644 if not self._CheckDisksActivated(instance):
10645 raise errors.OpPrereqError("Please run activate-disks on instance %s"
10646 " first" % self.instance_name,
10647 errors.ECODE_STATE)
10648 faulty_primary = self._FindFaultyDisks(instance.primary_node)
10649 faulty_secondary = self._FindFaultyDisks(secondary_node)
10651 if faulty_primary and faulty_secondary:
10652 raise errors.OpPrereqError("Instance %s has faulty disks on more than"
10653 " one node and can not be repaired"
10654 " automatically" % self.instance_name,
10655 errors.ECODE_STATE)
10658 self.disks = faulty_primary
10659 self.target_node = instance.primary_node
10660 self.other_node = secondary_node
10661 check_nodes = [self.target_node, self.other_node]
10662 elif faulty_secondary:
10663 self.disks = faulty_secondary
10664 self.target_node = secondary_node
10665 self.other_node = instance.primary_node
10666 check_nodes = [self.target_node, self.other_node]
10672 # Non-automatic modes
10673 if self.mode == constants.REPLACE_DISK_PRI:
10674 self.target_node = instance.primary_node
10675 self.other_node = secondary_node
10676 check_nodes = [self.target_node, self.other_node]
10678 elif self.mode == constants.REPLACE_DISK_SEC:
10679 self.target_node = secondary_node
10680 self.other_node = instance.primary_node
10681 check_nodes = [self.target_node, self.other_node]
10683 elif self.mode == constants.REPLACE_DISK_CHG:
10684 self.new_node = remote_node
10685 self.other_node = instance.primary_node
10686 self.target_node = secondary_node
10687 check_nodes = [self.new_node, self.other_node]
10689 _CheckNodeNotDrained(self.lu, remote_node)
10690 _CheckNodeVmCapable(self.lu, remote_node)
10692 old_node_info = self.cfg.GetNodeInfo(secondary_node)
10693 assert old_node_info is not None
10694 if old_node_info.offline and not self.early_release:
10695 # doesn't make sense to delay the release
10696 self.early_release = True
10697 self.lu.LogInfo("Old secondary %s is offline, automatically enabling"
10698 " early-release mode", secondary_node)
10701 raise errors.ProgrammerError("Unhandled disk replace mode (%s)" %
10704 # If not specified all disks should be replaced
10706 self.disks = range(len(self.instance.disks))
10708 # TODO: This is ugly, but right now we can't distinguish between internal
10709 # submitted opcode and external one. We should fix that.
10710 if self.remote_node_info:
10711 # We change the node, lets verify it still meets instance policy
10712 new_group_info = self.cfg.GetNodeGroup(self.remote_node_info.group)
10713 ipolicy = _CalculateGroupIPolicy(self.cfg.GetClusterInfo(),
10715 _CheckTargetNodeIPolicy(self, ipolicy, instance, self.remote_node_info,
10716 ignore=self.ignore_ipolicy)
10718 for node in check_nodes:
10719 _CheckNodeOnline(self.lu, node)
10721 touched_nodes = frozenset(node_name for node_name in [self.new_node,
10724 if node_name is not None)
10726 # Release unneeded node and node resource locks
10727 _ReleaseLocks(self.lu, locking.LEVEL_NODE, keep=touched_nodes)
10728 _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES, keep=touched_nodes)
10730 # Release any owned node group
10731 if self.lu.glm.is_owned(locking.LEVEL_NODEGROUP):
10732 _ReleaseLocks(self.lu, locking.LEVEL_NODEGROUP)
10734 # Check whether disks are valid
10735 for disk_idx in self.disks:
10736 instance.FindDisk(disk_idx)
10738 # Get secondary node IP addresses
10739 self.node_secondary_ip = dict((name, node.secondary_ip) for (name, node)
10740 in self.cfg.GetMultiNodeInfo(touched_nodes))
10742 def Exec(self, feedback_fn):
10743 """Execute disk replacement.
10745 This dispatches the disk replacement to the appropriate handler.
10748 if self.delay_iallocator:
10749 self._CheckPrereq2()
10752 # Verify owned locks before starting operation
10753 owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE)
10754 assert set(owned_nodes) == set(self.node_secondary_ip), \
10755 ("Incorrect node locks, owning %s, expected %s" %
10756 (owned_nodes, self.node_secondary_ip.keys()))
10757 assert (self.lu.owned_locks(locking.LEVEL_NODE) ==
10758 self.lu.owned_locks(locking.LEVEL_NODE_RES))
10760 owned_instances = self.lu.owned_locks(locking.LEVEL_INSTANCE)
10761 assert list(owned_instances) == [self.instance_name], \
10762 "Instance '%s' not locked" % self.instance_name
10764 assert not self.lu.glm.is_owned(locking.LEVEL_NODEGROUP), \
10765 "Should not own any node group lock at this point"
10768 feedback_fn("No disks need replacement")
10771 feedback_fn("Replacing disk(s) %s for %s" %
10772 (utils.CommaJoin(self.disks), self.instance.name))
10774 activate_disks = (self.instance.admin_state != constants.ADMINST_UP)
10776 # Activate the instance disks if we're replacing them on a down instance
10778 _StartInstanceDisks(self.lu, self.instance, True)
10781 # Should we replace the secondary node?
10782 if self.new_node is not None:
10783 fn = self._ExecDrbd8Secondary
10785 fn = self._ExecDrbd8DiskOnly
10787 result = fn(feedback_fn)
10789 # Deactivate the instance disks if we're replacing them on a
10792 _SafeShutdownInstanceDisks(self.lu, self.instance)
10794 assert not self.lu.owned_locks(locking.LEVEL_NODE)
10797 # Verify owned locks
10798 owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE_RES)
10799 nodes = frozenset(self.node_secondary_ip)
10800 assert ((self.early_release and not owned_nodes) or
10801 (not self.early_release and not (set(owned_nodes) - nodes))), \
10802 ("Not owning the correct locks, early_release=%s, owned=%r,"
10803 " nodes=%r" % (self.early_release, owned_nodes, nodes))
10807 def _CheckVolumeGroup(self, nodes):
10808 self.lu.LogInfo("Checking volume groups")
10810 vgname = self.cfg.GetVGName()
10812 # Make sure volume group exists on all involved nodes
10813 results = self.rpc.call_vg_list(nodes)
10815 raise errors.OpExecError("Can't list volume groups on the nodes")
10818 res = results[node]
10819 res.Raise("Error checking node %s" % node)
10820 if vgname not in res.payload:
10821 raise errors.OpExecError("Volume group '%s' not found on node %s" %
10824 def _CheckDisksExistence(self, nodes):
10825 # Check disk existence
10826 for idx, dev in enumerate(self.instance.disks):
10827 if idx not in self.disks:
10831 self.lu.LogInfo("Checking disk/%d on %s" % (idx, node))
10832 self.cfg.SetDiskID(dev, node)
10834 result = _BlockdevFind(self, node, dev, self.instance)
10836 msg = result.fail_msg
10837 if msg or not result.payload:
10839 msg = "disk not found"
10840 raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
10843 def _CheckDisksConsistency(self, node_name, on_primary, ldisk):
10844 for idx, dev in enumerate(self.instance.disks):
10845 if idx not in self.disks:
10848 self.lu.LogInfo("Checking disk/%d consistency on node %s" %
10851 if not _CheckDiskConsistency(self.lu, self.instance, dev, node_name,
10852 on_primary, ldisk=ldisk):
10853 raise errors.OpExecError("Node %s has degraded storage, unsafe to"
10854 " replace disks for instance %s" %
10855 (node_name, self.instance.name))
10857 def _CreateNewStorage(self, node_name):
10858 """Create new storage on the primary or secondary node.
10860 This is only used for same-node replaces, not for changing the
10861 secondary node, hence we don't want to modify the existing disk.
10866 disks = _AnnotateDiskParams(self.instance, self.instance.disks, self.cfg)
10867 for idx, dev in enumerate(disks):
10868 if idx not in self.disks:
10871 self.lu.LogInfo("Adding storage on %s for disk/%d" % (node_name, idx))
10873 self.cfg.SetDiskID(dev, node_name)
10875 lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
10876 names = _GenerateUniqueNames(self.lu, lv_names)
10878 (data_disk, meta_disk) = dev.children
10879 vg_data = data_disk.logical_id[0]
10880 lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
10881 logical_id=(vg_data, names[0]),
10882 params=data_disk.params)
10883 vg_meta = meta_disk.logical_id[0]
10884 lv_meta = objects.Disk(dev_type=constants.LD_LV, size=DRBD_META_SIZE,
10885 logical_id=(vg_meta, names[1]),
10886 params=meta_disk.params)
10888 new_lvs = [lv_data, lv_meta]
10889 old_lvs = [child.Copy() for child in dev.children]
10890 iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
10892 # we pass force_create=True to force the LVM creation
10893 for new_lv in new_lvs:
10894 _CreateBlockDevInner(self.lu, node_name, self.instance, new_lv, True,
10895 _GetInstanceInfoText(self.instance), False)
10899 def _CheckDevices(self, node_name, iv_names):
10900 for name, (dev, _, _) in iv_names.iteritems():
10901 self.cfg.SetDiskID(dev, node_name)
10903 result = _BlockdevFind(self, node_name, dev, self.instance)
10905 msg = result.fail_msg
10906 if msg or not result.payload:
10908 msg = "disk not found"
10909 raise errors.OpExecError("Can't find DRBD device %s: %s" %
10912 if result.payload.is_degraded:
10913 raise errors.OpExecError("DRBD device %s is degraded!" % name)
10915 def _RemoveOldStorage(self, node_name, iv_names):
10916 for name, (_, old_lvs, _) in iv_names.iteritems():
10917 self.lu.LogInfo("Remove logical volumes for %s" % name)
10920 self.cfg.SetDiskID(lv, node_name)
10922 msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
10924 self.lu.LogWarning("Can't remove old LV: %s" % msg,
10925 hint="remove unused LVs manually")
10927 def _ExecDrbd8DiskOnly(self, feedback_fn): # pylint: disable=W0613
10928 """Replace a disk on the primary or secondary for DRBD 8.
10930 The algorithm for replace is quite complicated:
10932 1. for each disk to be replaced:
10934 1. create new LVs on the target node with unique names
10935 1. detach old LVs from the drbd device
10936 1. rename old LVs to name_replaced.<time_t>
10937 1. rename new LVs to old LVs
10938 1. attach the new LVs (with the old names now) to the drbd device
10940 1. wait for sync across all devices
10942 1. for each modified disk:
10944 1. remove old LVs (which have the name name_replaces.<time_t>)
10946 Failures are not very well handled.
10951 # Step: check device activation
10952 self.lu.LogStep(1, steps_total, "Check device existence")
10953 self._CheckDisksExistence([self.other_node, self.target_node])
10954 self._CheckVolumeGroup([self.target_node, self.other_node])
10956 # Step: check other node consistency
10957 self.lu.LogStep(2, steps_total, "Check peer consistency")
10958 self._CheckDisksConsistency(self.other_node,
10959 self.other_node == self.instance.primary_node,
10962 # Step: create new storage
10963 self.lu.LogStep(3, steps_total, "Allocate new storage")
10964 iv_names = self._CreateNewStorage(self.target_node)
10966 # Step: for each lv, detach+rename*2+attach
10967 self.lu.LogStep(4, steps_total, "Changing drbd configuration")
10968 for dev, old_lvs, new_lvs in iv_names.itervalues():
10969 self.lu.LogInfo("Detaching %s drbd from local storage" % dev.iv_name)
10971 result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
10973 result.Raise("Can't detach drbd from local storage on node"
10974 " %s for device %s" % (self.target_node, dev.iv_name))
10976 #cfg.Update(instance)
10978 # ok, we created the new LVs, so now we know we have the needed
10979 # storage; as such, we proceed on the target node to rename
10980 # old_lv to _old, and new_lv to old_lv; note that we rename LVs
10981 # using the assumption that logical_id == physical_id (which in
10982 # turn is the unique_id on that node)
10984 # FIXME(iustin): use a better name for the replaced LVs
10985 temp_suffix = int(time.time())
10986 ren_fn = lambda d, suff: (d.physical_id[0],
10987 d.physical_id[1] + "_replaced-%s" % suff)
10989 # Build the rename list based on what LVs exist on the node
10990 rename_old_to_new = []
10991 for to_ren in old_lvs:
10992 result = self.rpc.call_blockdev_find(self.target_node, to_ren)
10993 if not result.fail_msg and result.payload:
10995 rename_old_to_new.append((to_ren, ren_fn(to_ren, temp_suffix)))
10997 self.lu.LogInfo("Renaming the old LVs on the target node")
10998 result = self.rpc.call_blockdev_rename(self.target_node,
11000 result.Raise("Can't rename old LVs on node %s" % self.target_node)
11002 # Now we rename the new LVs to the old LVs
11003 self.lu.LogInfo("Renaming the new LVs on the target node")
11004 rename_new_to_old = [(new, old.physical_id)
11005 for old, new in zip(old_lvs, new_lvs)]
11006 result = self.rpc.call_blockdev_rename(self.target_node,
11008 result.Raise("Can't rename new LVs on node %s" % self.target_node)
11010 # Intermediate steps of in memory modifications
11011 for old, new in zip(old_lvs, new_lvs):
11012 new.logical_id = old.logical_id
11013 self.cfg.SetDiskID(new, self.target_node)
11015 # We need to modify old_lvs so that removal later removes the
11016 # right LVs, not the newly added ones; note that old_lvs is a
11018 for disk in old_lvs:
11019 disk.logical_id = ren_fn(disk, temp_suffix)
11020 self.cfg.SetDiskID(disk, self.target_node)
11022 # Now that the new lvs have the old name, we can add them to the device
11023 self.lu.LogInfo("Adding new mirror component on %s" % self.target_node)
11024 result = self.rpc.call_blockdev_addchildren(self.target_node,
11025 (dev, self.instance), new_lvs)
11026 msg = result.fail_msg
11028 for new_lv in new_lvs:
11029 msg2 = self.rpc.call_blockdev_remove(self.target_node,
11032 self.lu.LogWarning("Can't rollback device %s: %s", dev, msg2,
11033 hint=("cleanup manually the unused logical"
11035 raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
11037 cstep = itertools.count(5)
11039 if self.early_release:
11040 self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
11041 self._RemoveOldStorage(self.target_node, iv_names)
11042 # TODO: Check if releasing locks early still makes sense
11043 _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES)
11045 # Release all resource locks except those used by the instance
11046 _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES,
11047 keep=self.node_secondary_ip.keys())
11049 # Release all node locks while waiting for sync
11050 _ReleaseLocks(self.lu, locking.LEVEL_NODE)
11052 # TODO: Can the instance lock be downgraded here? Take the optional disk
11053 # shutdown in the caller into consideration.
11056 # This can fail as the old devices are degraded and _WaitForSync
11057 # does a combined result over all disks, so we don't check its return value
11058 self.lu.LogStep(cstep.next(), steps_total, "Sync devices")
11059 _WaitForSync(self.lu, self.instance)
11061 # Check all devices manually
11062 self._CheckDevices(self.instance.primary_node, iv_names)
11064 # Step: remove old storage
11065 if not self.early_release:
11066 self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
11067 self._RemoveOldStorage(self.target_node, iv_names)
11069 def _ExecDrbd8Secondary(self, feedback_fn):
11070 """Replace the secondary node for DRBD 8.
11072 The algorithm for replace is quite complicated:
11073 - for all disks of the instance:
11074 - create new LVs on the new node with same names
11075 - shutdown the drbd device on the old secondary
11076 - disconnect the drbd network on the primary
11077 - create the drbd device on the new secondary
11078 - network attach the drbd on the primary, using an artifice:
11079 the drbd code for Attach() will connect to the network if it
11080 finds a device which is connected to the good local disks but
11081 not network enabled
11082 - wait for sync across all devices
11083 - remove all disks from the old secondary
11085 Failures are not very well handled.
11090 pnode = self.instance.primary_node
11092 # Step: check device activation
11093 self.lu.LogStep(1, steps_total, "Check device existence")
11094 self._CheckDisksExistence([self.instance.primary_node])
11095 self._CheckVolumeGroup([self.instance.primary_node])
11097 # Step: check other node consistency
11098 self.lu.LogStep(2, steps_total, "Check peer consistency")
11099 self._CheckDisksConsistency(self.instance.primary_node, True, True)
11101 # Step: create new storage
11102 self.lu.LogStep(3, steps_total, "Allocate new storage")
11103 disks = _AnnotateDiskParams(self.instance, self.instance.disks, self.cfg)
11104 for idx, dev in enumerate(disks):
11105 self.lu.LogInfo("Adding new local storage on %s for disk/%d" %
11106 (self.new_node, idx))
11107 # we pass force_create=True to force LVM creation
11108 for new_lv in dev.children:
11109 _CreateBlockDevInner(self.lu, self.new_node, self.instance, new_lv,
11110 True, _GetInstanceInfoText(self.instance), False)
11112 # Step 4: dbrd minors and drbd setups changes
11113 # after this, we must manually remove the drbd minors on both the
11114 # error and the success paths
11115 self.lu.LogStep(4, steps_total, "Changing drbd configuration")
11116 minors = self.cfg.AllocateDRBDMinor([self.new_node
11117 for dev in self.instance.disks],
11118 self.instance.name)
11119 logging.debug("Allocated minors %r", minors)
11122 for idx, (dev, new_minor) in enumerate(zip(self.instance.disks, minors)):
11123 self.lu.LogInfo("activating a new drbd on %s for disk/%d" %
11124 (self.new_node, idx))
11125 # create new devices on new_node; note that we create two IDs:
11126 # one without port, so the drbd will be activated without
11127 # networking information on the new node at this stage, and one
11128 # with network, for the latter activation in step 4
11129 (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
11130 if self.instance.primary_node == o_node1:
11133 assert self.instance.primary_node == o_node2, "Three-node instance?"
11136 new_alone_id = (self.instance.primary_node, self.new_node, None,
11137 p_minor, new_minor, o_secret)
11138 new_net_id = (self.instance.primary_node, self.new_node, o_port,
11139 p_minor, new_minor, o_secret)
11141 iv_names[idx] = (dev, dev.children, new_net_id)
11142 logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
11144 new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
11145 logical_id=new_alone_id,
11146 children=dev.children,
11149 (anno_new_drbd,) = _AnnotateDiskParams(self.instance, [new_drbd],
11152 _CreateSingleBlockDev(self.lu, self.new_node, self.instance,
11154 _GetInstanceInfoText(self.instance), False)
11155 except errors.GenericError:
11156 self.cfg.ReleaseDRBDMinors(self.instance.name)
11159 # We have new devices, shutdown the drbd on the old secondary
11160 for idx, dev in enumerate(self.instance.disks):
11161 self.lu.LogInfo("Shutting down drbd for disk/%d on old node" % idx)
11162 self.cfg.SetDiskID(dev, self.target_node)
11163 msg = self.rpc.call_blockdev_shutdown(self.target_node,
11164 (dev, self.instance)).fail_msg
11166 self.lu.LogWarning("Failed to shutdown drbd for disk/%d on old"
11167 "node: %s" % (idx, msg),
11168 hint=("Please cleanup this device manually as"
11169 " soon as possible"))
11171 self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
11172 result = self.rpc.call_drbd_disconnect_net([pnode], self.node_secondary_ip,
11173 self.instance.disks)[pnode]
11175 msg = result.fail_msg
11177 # detaches didn't succeed (unlikely)
11178 self.cfg.ReleaseDRBDMinors(self.instance.name)
11179 raise errors.OpExecError("Can't detach the disks from the network on"
11180 " old node: %s" % (msg,))
11182 # if we managed to detach at least one, we update all the disks of
11183 # the instance to point to the new secondary
11184 self.lu.LogInfo("Updating instance configuration")
11185 for dev, _, new_logical_id in iv_names.itervalues():
11186 dev.logical_id = new_logical_id
11187 self.cfg.SetDiskID(dev, self.instance.primary_node)
11189 self.cfg.Update(self.instance, feedback_fn)
11191 # Release all node locks (the configuration has been updated)
11192 _ReleaseLocks(self.lu, locking.LEVEL_NODE)
11194 # and now perform the drbd attach
11195 self.lu.LogInfo("Attaching primary drbds to new secondary"
11196 " (standalone => connected)")
11197 result = self.rpc.call_drbd_attach_net([self.instance.primary_node,
11199 self.node_secondary_ip,
11200 (self.instance.disks, self.instance),
11201 self.instance.name,
11203 for to_node, to_result in result.items():
11204 msg = to_result.fail_msg
11206 self.lu.LogWarning("Can't attach drbd disks on node %s: %s",
11208 hint=("please do a gnt-instance info to see the"
11209 " status of disks"))
11211 cstep = itertools.count(5)
11213 if self.early_release:
11214 self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
11215 self._RemoveOldStorage(self.target_node, iv_names)
11216 # TODO: Check if releasing locks early still makes sense
11217 _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES)
11219 # Release all resource locks except those used by the instance
11220 _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES,
11221 keep=self.node_secondary_ip.keys())
11223 # TODO: Can the instance lock be downgraded here? Take the optional disk
11224 # shutdown in the caller into consideration.
11227 # This can fail as the old devices are degraded and _WaitForSync
11228 # does a combined result over all disks, so we don't check its return value
11229 self.lu.LogStep(cstep.next(), steps_total, "Sync devices")
11230 _WaitForSync(self.lu, self.instance)
11232 # Check all devices manually
11233 self._CheckDevices(self.instance.primary_node, iv_names)
11235 # Step: remove old storage
11236 if not self.early_release:
11237 self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
11238 self._RemoveOldStorage(self.target_node, iv_names)
11241 class LURepairNodeStorage(NoHooksLU):
11242 """Repairs the volume group on a node.
11247 def CheckArguments(self):
11248 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
11250 storage_type = self.op.storage_type
11252 if (constants.SO_FIX_CONSISTENCY not in
11253 constants.VALID_STORAGE_OPERATIONS.get(storage_type, [])):
11254 raise errors.OpPrereqError("Storage units of type '%s' can not be"
11255 " repaired" % storage_type,
11256 errors.ECODE_INVAL)
11258 def ExpandNames(self):
11259 self.needed_locks = {
11260 locking.LEVEL_NODE: [self.op.node_name],
11263 def _CheckFaultyDisks(self, instance, node_name):
11264 """Ensure faulty disks abort the opcode or at least warn."""
11266 if _FindFaultyInstanceDisks(self.cfg, self.rpc, instance,
11268 raise errors.OpPrereqError("Instance '%s' has faulty disks on"
11269 " node '%s'" % (instance.name, node_name),
11270 errors.ECODE_STATE)
11271 except errors.OpPrereqError, err:
11272 if self.op.ignore_consistency:
11273 self.proc.LogWarning(str(err.args[0]))
11277 def CheckPrereq(self):
11278 """Check prerequisites.
11281 # Check whether any instance on this node has faulty disks
11282 for inst in _GetNodeInstances(self.cfg, self.op.node_name):
11283 if inst.admin_state != constants.ADMINST_UP:
11285 check_nodes = set(inst.all_nodes)
11286 check_nodes.discard(self.op.node_name)
11287 for inst_node_name in check_nodes:
11288 self._CheckFaultyDisks(inst, inst_node_name)
11290 def Exec(self, feedback_fn):
11291 feedback_fn("Repairing storage unit '%s' on %s ..." %
11292 (self.op.name, self.op.node_name))
11294 st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
11295 result = self.rpc.call_storage_execute(self.op.node_name,
11296 self.op.storage_type, st_args,
11298 constants.SO_FIX_CONSISTENCY)
11299 result.Raise("Failed to repair storage unit '%s' on %s" %
11300 (self.op.name, self.op.node_name))
11303 class LUNodeEvacuate(NoHooksLU):
11304 """Evacuates instances off a list of nodes.
11309 _MODE2IALLOCATOR = {
11310 constants.NODE_EVAC_PRI: constants.IALLOCATOR_NEVAC_PRI,
11311 constants.NODE_EVAC_SEC: constants.IALLOCATOR_NEVAC_SEC,
11312 constants.NODE_EVAC_ALL: constants.IALLOCATOR_NEVAC_ALL,
11314 assert frozenset(_MODE2IALLOCATOR.keys()) == constants.NODE_EVAC_MODES
11315 assert (frozenset(_MODE2IALLOCATOR.values()) ==
11316 constants.IALLOCATOR_NEVAC_MODES)
11318 def CheckArguments(self):
11319 _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
11321 def ExpandNames(self):
11322 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
11324 if self.op.remote_node is not None:
11325 self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
11326 assert self.op.remote_node
11328 if self.op.remote_node == self.op.node_name:
11329 raise errors.OpPrereqError("Can not use evacuated node as a new"
11330 " secondary node", errors.ECODE_INVAL)
11332 if self.op.mode != constants.NODE_EVAC_SEC:
11333 raise errors.OpPrereqError("Without the use of an iallocator only"
11334 " secondary instances can be evacuated",
11335 errors.ECODE_INVAL)
11338 self.share_locks = _ShareAll()
11339 self.needed_locks = {
11340 locking.LEVEL_INSTANCE: [],
11341 locking.LEVEL_NODEGROUP: [],
11342 locking.LEVEL_NODE: [],
11345 # Determine nodes (via group) optimistically, needs verification once locks
11346 # have been acquired
11347 self.lock_nodes = self._DetermineNodes()
11349 def _DetermineNodes(self):
11350 """Gets the list of nodes to operate on.
11353 if self.op.remote_node is None:
11354 # Iallocator will choose any node(s) in the same group
11355 group_nodes = self.cfg.GetNodeGroupMembersByNodes([self.op.node_name])
11357 group_nodes = frozenset([self.op.remote_node])
11359 # Determine nodes to be locked
11360 return set([self.op.node_name]) | group_nodes
11362 def _DetermineInstances(self):
11363 """Builds list of instances to operate on.
11366 assert self.op.mode in constants.NODE_EVAC_MODES
11368 if self.op.mode == constants.NODE_EVAC_PRI:
11369 # Primary instances only
11370 inst_fn = _GetNodePrimaryInstances
11371 assert self.op.remote_node is None, \
11372 "Evacuating primary instances requires iallocator"
11373 elif self.op.mode == constants.NODE_EVAC_SEC:
11374 # Secondary instances only
11375 inst_fn = _GetNodeSecondaryInstances
11378 assert self.op.mode == constants.NODE_EVAC_ALL
11379 inst_fn = _GetNodeInstances
11380 # TODO: In 2.6, change the iallocator interface to take an evacuation mode
11382 raise errors.OpPrereqError("Due to an issue with the iallocator"
11383 " interface it is not possible to evacuate"
11384 " all instances at once; specify explicitly"
11385 " whether to evacuate primary or secondary"
11387 errors.ECODE_INVAL)
11389 return inst_fn(self.cfg, self.op.node_name)
11391 def DeclareLocks(self, level):
11392 if level == locking.LEVEL_INSTANCE:
11393 # Lock instances optimistically, needs verification once node and group
11394 # locks have been acquired
11395 self.needed_locks[locking.LEVEL_INSTANCE] = \
11396 set(i.name for i in self._DetermineInstances())
11398 elif level == locking.LEVEL_NODEGROUP:
11399 # Lock node groups for all potential target nodes optimistically, needs
11400 # verification once nodes have been acquired
11401 self.needed_locks[locking.LEVEL_NODEGROUP] = \
11402 self.cfg.GetNodeGroupsFromNodes(self.lock_nodes)
11404 elif level == locking.LEVEL_NODE:
11405 self.needed_locks[locking.LEVEL_NODE] = self.lock_nodes
11407 def CheckPrereq(self):
11409 owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
11410 owned_nodes = self.owned_locks(locking.LEVEL_NODE)
11411 owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
11413 need_nodes = self._DetermineNodes()
11415 if not owned_nodes.issuperset(need_nodes):
11416 raise errors.OpPrereqError("Nodes in same group as '%s' changed since"
11417 " locks were acquired, current nodes are"
11418 " are '%s', used to be '%s'; retry the"
11420 (self.op.node_name,
11421 utils.CommaJoin(need_nodes),
11422 utils.CommaJoin(owned_nodes)),
11423 errors.ECODE_STATE)
11425 wanted_groups = self.cfg.GetNodeGroupsFromNodes(owned_nodes)
11426 if owned_groups != wanted_groups:
11427 raise errors.OpExecError("Node groups changed since locks were acquired,"
11428 " current groups are '%s', used to be '%s';"
11429 " retry the operation" %
11430 (utils.CommaJoin(wanted_groups),
11431 utils.CommaJoin(owned_groups)))
11433 # Determine affected instances
11434 self.instances = self._DetermineInstances()
11435 self.instance_names = [i.name for i in self.instances]
11437 if set(self.instance_names) != owned_instances:
11438 raise errors.OpExecError("Instances on node '%s' changed since locks"
11439 " were acquired, current instances are '%s',"
11440 " used to be '%s'; retry the operation" %
11441 (self.op.node_name,
11442 utils.CommaJoin(self.instance_names),
11443 utils.CommaJoin(owned_instances)))
11445 if self.instance_names:
11446 self.LogInfo("Evacuating instances from node '%s': %s",
11448 utils.CommaJoin(utils.NiceSort(self.instance_names)))
11450 self.LogInfo("No instances to evacuate from node '%s'",
11453 if self.op.remote_node is not None:
11454 for i in self.instances:
11455 if i.primary_node == self.op.remote_node:
11456 raise errors.OpPrereqError("Node %s is the primary node of"
11457 " instance %s, cannot use it as"
11459 (self.op.remote_node, i.name),
11460 errors.ECODE_INVAL)
11462 def Exec(self, feedback_fn):
11463 assert (self.op.iallocator is not None) ^ (self.op.remote_node is not None)
11465 if not self.instance_names:
11466 # No instances to evacuate
11469 elif self.op.iallocator is not None:
11470 # TODO: Implement relocation to other group
11471 ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_NODE_EVAC,
11472 evac_mode=self._MODE2IALLOCATOR[self.op.mode],
11473 instances=list(self.instance_names))
11475 ial.Run(self.op.iallocator)
11477 if not ial.success:
11478 raise errors.OpPrereqError("Can't compute node evacuation using"
11479 " iallocator '%s': %s" %
11480 (self.op.iallocator, ial.info),
11481 errors.ECODE_NORES)
11483 jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, True)
11485 elif self.op.remote_node is not None:
11486 assert self.op.mode == constants.NODE_EVAC_SEC
11488 [opcodes.OpInstanceReplaceDisks(instance_name=instance_name,
11489 remote_node=self.op.remote_node,
11491 mode=constants.REPLACE_DISK_CHG,
11492 early_release=self.op.early_release)]
11493 for instance_name in self.instance_names
11497 raise errors.ProgrammerError("No iallocator or remote node")
11499 return ResultWithJobs(jobs)
11502 def _SetOpEarlyRelease(early_release, op):
11503 """Sets C{early_release} flag on opcodes if available.
11507 op.early_release = early_release
11508 except AttributeError:
11509 assert not isinstance(op, opcodes.OpInstanceReplaceDisks)
11514 def _NodeEvacDest(use_nodes, group, nodes):
11515 """Returns group or nodes depending on caller's choice.
11519 return utils.CommaJoin(nodes)
11524 def _LoadNodeEvacResult(lu, alloc_result, early_release, use_nodes):
11525 """Unpacks the result of change-group and node-evacuate iallocator requests.
11527 Iallocator modes L{constants.IALLOCATOR_MODE_NODE_EVAC} and
11528 L{constants.IALLOCATOR_MODE_CHG_GROUP}.
11530 @type lu: L{LogicalUnit}
11531 @param lu: Logical unit instance
11532 @type alloc_result: tuple/list
11533 @param alloc_result: Result from iallocator
11534 @type early_release: bool
11535 @param early_release: Whether to release locks early if possible
11536 @type use_nodes: bool
11537 @param use_nodes: Whether to display node names instead of groups
11540 (moved, failed, jobs) = alloc_result
11543 failreason = utils.CommaJoin("%s (%s)" % (name, reason)
11544 for (name, reason) in failed)
11545 lu.LogWarning("Unable to evacuate instances %s", failreason)
11546 raise errors.OpExecError("Unable to evacuate instances %s" % failreason)
11549 lu.LogInfo("Instances to be moved: %s",
11550 utils.CommaJoin("%s (to %s)" %
11551 (name, _NodeEvacDest(use_nodes, group, nodes))
11552 for (name, group, nodes) in moved))
11554 return [map(compat.partial(_SetOpEarlyRelease, early_release),
11555 map(opcodes.OpCode.LoadOpCode, ops))
11559 class LUInstanceGrowDisk(LogicalUnit):
11560 """Grow a disk of an instance.
11563 HPATH = "disk-grow"
11564 HTYPE = constants.HTYPE_INSTANCE
11567 def ExpandNames(self):
11568 self._ExpandAndLockInstance()
11569 self.needed_locks[locking.LEVEL_NODE] = []
11570 self.needed_locks[locking.LEVEL_NODE_RES] = []
11571 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
11572 self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
11574 def DeclareLocks(self, level):
11575 if level == locking.LEVEL_NODE:
11576 self._LockInstancesNodes()
11577 elif level == locking.LEVEL_NODE_RES:
11579 self.needed_locks[locking.LEVEL_NODE_RES] = \
11580 self.needed_locks[locking.LEVEL_NODE][:]
11582 def BuildHooksEnv(self):
11583 """Build hooks env.
11585 This runs on the master, the primary and all the secondaries.
11589 "DISK": self.op.disk,
11590 "AMOUNT": self.op.amount,
11591 "ABSOLUTE": self.op.absolute,
11593 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
11596 def BuildHooksNodes(self):
11597 """Build hooks nodes.
11600 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
11603 def CheckPrereq(self):
11604 """Check prerequisites.
11606 This checks that the instance is in the cluster.
11609 instance = self.cfg.GetInstanceInfo(self.op.instance_name)
11610 assert instance is not None, \
11611 "Cannot retrieve locked instance %s" % self.op.instance_name
11612 nodenames = list(instance.all_nodes)
11613 for node in nodenames:
11614 _CheckNodeOnline(self, node)
11616 self.instance = instance
11618 if instance.disk_template not in constants.DTS_GROWABLE:
11619 raise errors.OpPrereqError("Instance's disk layout does not support"
11620 " growing", errors.ECODE_INVAL)
11622 self.disk = instance.FindDisk(self.op.disk)
11624 if self.op.absolute:
11625 self.target = self.op.amount
11626 self.delta = self.target - self.disk.size
11628 raise errors.OpPrereqError("Requested size (%s) is smaller than "
11629 "current disk size (%s)" %
11630 (utils.FormatUnit(self.target, "h"),
11631 utils.FormatUnit(self.disk.size, "h")),
11632 errors.ECODE_STATE)
11634 self.delta = self.op.amount
11635 self.target = self.disk.size + self.delta
11637 raise errors.OpPrereqError("Requested increment (%s) is negative" %
11638 utils.FormatUnit(self.delta, "h"),
11639 errors.ECODE_INVAL)
11641 if instance.disk_template not in (constants.DT_FILE,
11642 constants.DT_SHARED_FILE,
11644 # TODO: check the free disk space for file, when that feature will be
11646 _CheckNodesFreeDiskPerVG(self, nodenames,
11647 self.disk.ComputeGrowth(self.delta))
11649 def Exec(self, feedback_fn):
11650 """Execute disk grow.
11653 instance = self.instance
11656 assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
11657 assert (self.owned_locks(locking.LEVEL_NODE) ==
11658 self.owned_locks(locking.LEVEL_NODE_RES))
11660 disks_ok, _ = _AssembleInstanceDisks(self, self.instance, disks=[disk])
11662 raise errors.OpExecError("Cannot activate block device to grow")
11664 feedback_fn("Growing disk %s of instance '%s' by %s to %s" %
11665 (self.op.disk, instance.name,
11666 utils.FormatUnit(self.delta, "h"),
11667 utils.FormatUnit(self.target, "h")))
11669 # First run all grow ops in dry-run mode
11670 for node in instance.all_nodes:
11671 self.cfg.SetDiskID(disk, node)
11672 result = self.rpc.call_blockdev_grow(node, (disk, instance), self.delta,
11674 result.Raise("Grow request failed to node %s" % node)
11676 # We know that (as far as we can test) operations across different
11677 # nodes will succeed, time to run it for real
11678 for node in instance.all_nodes:
11679 self.cfg.SetDiskID(disk, node)
11680 result = self.rpc.call_blockdev_grow(node, (disk, instance), self.delta,
11682 result.Raise("Grow request failed to node %s" % node)
11684 # TODO: Rewrite code to work properly
11685 # DRBD goes into sync mode for a short amount of time after executing the
11686 # "resize" command. DRBD 8.x below version 8.0.13 contains a bug whereby
11687 # calling "resize" in sync mode fails. Sleeping for a short amount of
11688 # time is a work-around.
11691 disk.RecordGrow(self.delta)
11692 self.cfg.Update(instance, feedback_fn)
11694 # Changes have been recorded, release node lock
11695 _ReleaseLocks(self, locking.LEVEL_NODE)
11697 # Downgrade lock while waiting for sync
11698 self.glm.downgrade(locking.LEVEL_INSTANCE)
11700 if self.op.wait_for_sync:
11701 disk_abort = not _WaitForSync(self, instance, disks=[disk])
11703 self.proc.LogWarning("Disk sync-ing has not returned a good"
11704 " status; please check the instance")
11705 if instance.admin_state != constants.ADMINST_UP:
11706 _SafeShutdownInstanceDisks(self, instance, disks=[disk])
11707 elif instance.admin_state != constants.ADMINST_UP:
11708 self.proc.LogWarning("Not shutting down the disk even if the instance is"
11709 " not supposed to be running because no wait for"
11710 " sync mode was requested")
11712 assert self.owned_locks(locking.LEVEL_NODE_RES)
11713 assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
11716 class LUInstanceQueryData(NoHooksLU):
11717 """Query runtime instance data.
11722 def ExpandNames(self):
11723 self.needed_locks = {}
11725 # Use locking if requested or when non-static information is wanted
11726 if not (self.op.static or self.op.use_locking):
11727 self.LogWarning("Non-static data requested, locks need to be acquired")
11728 self.op.use_locking = True
11730 if self.op.instances or not self.op.use_locking:
11731 # Expand instance names right here
11732 self.wanted_names = _GetWantedInstances(self, self.op.instances)
11734 # Will use acquired locks
11735 self.wanted_names = None
11737 if self.op.use_locking:
11738 self.share_locks = _ShareAll()
11740 if self.wanted_names is None:
11741 self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
11743 self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
11745 self.needed_locks[locking.LEVEL_NODEGROUP] = []
11746 self.needed_locks[locking.LEVEL_NODE] = []
11747 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
11749 def DeclareLocks(self, level):
11750 if self.op.use_locking:
11751 if level == locking.LEVEL_NODEGROUP:
11752 owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
11754 # Lock all groups used by instances optimistically; this requires going
11755 # via the node before it's locked, requiring verification later on
11756 self.needed_locks[locking.LEVEL_NODEGROUP] = \
11757 frozenset(group_uuid
11758 for instance_name in owned_instances
11760 self.cfg.GetInstanceNodeGroups(instance_name))
11762 elif level == locking.LEVEL_NODE:
11763 self._LockInstancesNodes()
11765 def CheckPrereq(self):
11766 """Check prerequisites.
11768 This only checks the optional instance list against the existing names.
11771 owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
11772 owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
11773 owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
11775 if self.wanted_names is None:
11776 assert self.op.use_locking, "Locking was not used"
11777 self.wanted_names = owned_instances
11779 instances = dict(self.cfg.GetMultiInstanceInfo(self.wanted_names))
11781 if self.op.use_locking:
11782 _CheckInstancesNodeGroups(self.cfg, instances, owned_groups, owned_nodes,
11785 assert not (owned_instances or owned_groups or owned_nodes)
11787 self.wanted_instances = instances.values()
11789 def _ComputeBlockdevStatus(self, node, instance, dev):
11790 """Returns the status of a block device
11793 if self.op.static or not node:
11796 self.cfg.SetDiskID(dev, node)
11798 result = self.rpc.call_blockdev_find(node, dev)
11802 result.Raise("Can't compute disk status for %s" % instance.name)
11804 status = result.payload
11808 return (status.dev_path, status.major, status.minor,
11809 status.sync_percent, status.estimated_time,
11810 status.is_degraded, status.ldisk_status)
11812 def _ComputeDiskStatus(self, instance, snode, dev):
11813 """Compute block device status.
11816 (anno_dev,) = _AnnotateDiskParams(instance, [dev], self.cfg)
11818 return self._ComputeDiskStatusInner(instance, snode, anno_dev)
11820 def _ComputeDiskStatusInner(self, instance, snode, dev):
11821 """Compute block device status.
11823 @attention: The device has to be annotated already.
11826 if dev.dev_type in constants.LDS_DRBD:
11827 # we change the snode then (otherwise we use the one passed in)
11828 if dev.logical_id[0] == instance.primary_node:
11829 snode = dev.logical_id[1]
11831 snode = dev.logical_id[0]
11833 dev_pstatus = self._ComputeBlockdevStatus(instance.primary_node,
11835 dev_sstatus = self._ComputeBlockdevStatus(snode, instance, dev)
11838 dev_children = map(compat.partial(self._ComputeDiskStatusInner,
11845 "iv_name": dev.iv_name,
11846 "dev_type": dev.dev_type,
11847 "logical_id": dev.logical_id,
11848 "physical_id": dev.physical_id,
11849 "pstatus": dev_pstatus,
11850 "sstatus": dev_sstatus,
11851 "children": dev_children,
11856 def Exec(self, feedback_fn):
11857 """Gather and return data"""
11860 cluster = self.cfg.GetClusterInfo()
11862 node_names = itertools.chain(*(i.all_nodes for i in self.wanted_instances))
11863 nodes = dict(self.cfg.GetMultiNodeInfo(node_names))
11865 groups = dict(self.cfg.GetMultiNodeGroupInfo(node.group
11866 for node in nodes.values()))
11868 group2name_fn = lambda uuid: groups[uuid].name
11870 for instance in self.wanted_instances:
11871 pnode = nodes[instance.primary_node]
11873 if self.op.static or pnode.offline:
11874 remote_state = None
11876 self.LogWarning("Primary node %s is marked offline, returning static"
11877 " information only for instance %s" %
11878 (pnode.name, instance.name))
11880 remote_info = self.rpc.call_instance_info(instance.primary_node,
11882 instance.hypervisor)
11883 remote_info.Raise("Error checking node %s" % instance.primary_node)
11884 remote_info = remote_info.payload
11885 if remote_info and "state" in remote_info:
11886 remote_state = "up"
11888 if instance.admin_state == constants.ADMINST_UP:
11889 remote_state = "down"
11891 remote_state = instance.admin_state
11893 disks = map(compat.partial(self._ComputeDiskStatus, instance, None),
11896 snodes_group_uuids = [nodes[snode_name].group
11897 for snode_name in instance.secondary_nodes]
11899 result[instance.name] = {
11900 "name": instance.name,
11901 "config_state": instance.admin_state,
11902 "run_state": remote_state,
11903 "pnode": instance.primary_node,
11904 "pnode_group_uuid": pnode.group,
11905 "pnode_group_name": group2name_fn(pnode.group),
11906 "snodes": instance.secondary_nodes,
11907 "snodes_group_uuids": snodes_group_uuids,
11908 "snodes_group_names": map(group2name_fn, snodes_group_uuids),
11910 # this happens to be the same format used for hooks
11911 "nics": _NICListToTuple(self, instance.nics),
11912 "disk_template": instance.disk_template,
11914 "hypervisor": instance.hypervisor,
11915 "network_port": instance.network_port,
11916 "hv_instance": instance.hvparams,
11917 "hv_actual": cluster.FillHV(instance, skip_globals=True),
11918 "be_instance": instance.beparams,
11919 "be_actual": cluster.FillBE(instance),
11920 "os_instance": instance.osparams,
11921 "os_actual": cluster.SimpleFillOS(instance.os, instance.osparams),
11922 "serial_no": instance.serial_no,
11923 "mtime": instance.mtime,
11924 "ctime": instance.ctime,
11925 "uuid": instance.uuid,
11931 def PrepareContainerMods(mods, private_fn):
11932 """Prepares a list of container modifications by adding a private data field.
11934 @type mods: list of tuples; (operation, index, parameters)
11935 @param mods: List of modifications
11936 @type private_fn: callable or None
11937 @param private_fn: Callable for constructing a private data field for a
11942 if private_fn is None:
11947 return [(op, idx, params, fn()) for (op, idx, params) in mods]
11950 #: Type description for changes as returned by L{ApplyContainerMods}'s
11952 _TApplyContModsCbChanges = \
11953 ht.TMaybeListOf(ht.TAnd(ht.TIsLength(2), ht.TItems([
11954 ht.TNonEmptyString,
11959 def ApplyContainerMods(kind, container, chgdesc, mods,
11960 create_fn, modify_fn, remove_fn):
11961 """Applies descriptions in C{mods} to C{container}.
11964 @param kind: One-word item description
11965 @type container: list
11966 @param container: Container to modify
11967 @type chgdesc: None or list
11968 @param chgdesc: List of applied changes
11970 @param mods: Modifications as returned by L{PrepareContainerMods}
11971 @type create_fn: callable
11972 @param create_fn: Callback for creating a new item (L{constants.DDM_ADD});
11973 receives absolute item index, parameters and private data object as added
11974 by L{PrepareContainerMods}, returns tuple containing new item and changes
11976 @type modify_fn: callable
11977 @param modify_fn: Callback for modifying an existing item
11978 (L{constants.DDM_MODIFY}); receives absolute item index, item, parameters
11979 and private data object as added by L{PrepareContainerMods}, returns
11981 @type remove_fn: callable
11982 @param remove_fn: Callback on removing item; receives absolute item index,
11983 item and private data object as added by L{PrepareContainerMods}
11986 for (op, idx, params, private) in mods:
11989 absidx = len(container) - 1
11991 raise IndexError("Not accepting negative indices other than -1")
11992 elif idx > len(container):
11993 raise IndexError("Got %s index %s, but there are only %s" %
11994 (kind, idx, len(container)))
12000 if op == constants.DDM_ADD:
12001 # Calculate where item will be added
12003 addidx = len(container)
12007 if create_fn is None:
12010 (item, changes) = create_fn(addidx, params, private)
12013 container.append(item)
12016 assert idx <= len(container)
12017 # list.insert does so before the specified index
12018 container.insert(idx, item)
12020 # Retrieve existing item
12022 item = container[absidx]
12024 raise IndexError("Invalid %s index %s" % (kind, idx))
12026 if op == constants.DDM_REMOVE:
12029 if remove_fn is not None:
12030 remove_fn(absidx, item, private)
12032 changes = [("%s/%s" % (kind, absidx), "remove")]
12034 assert container[absidx] == item
12035 del container[absidx]
12036 elif op == constants.DDM_MODIFY:
12037 if modify_fn is not None:
12038 changes = modify_fn(absidx, item, params, private)
12040 raise errors.ProgrammerError("Unhandled operation '%s'" % op)
12042 assert _TApplyContModsCbChanges(changes)
12044 if not (chgdesc is None or changes is None):
12045 chgdesc.extend(changes)
12048 def _UpdateIvNames(base_index, disks):
12049 """Updates the C{iv_name} attribute of disks.
12051 @type disks: list of L{objects.Disk}
12054 for (idx, disk) in enumerate(disks):
12055 disk.iv_name = "disk/%s" % (base_index + idx, )
12058 class _InstNicModPrivate:
12059 """Data structure for network interface modifications.
12061 Used by L{LUInstanceSetParams}.
12064 def __init__(self):
12069 class LUInstanceSetParams(LogicalUnit):
12070 """Modifies an instances's parameters.
12073 HPATH = "instance-modify"
12074 HTYPE = constants.HTYPE_INSTANCE
12078 def _UpgradeDiskNicMods(kind, mods, verify_fn):
12079 assert ht.TList(mods)
12080 assert not mods or len(mods[0]) in (2, 3)
12082 if mods and len(mods[0]) == 2:
12086 for op, params in mods:
12087 if op in (constants.DDM_ADD, constants.DDM_REMOVE):
12088 result.append((op, -1, params))
12092 raise errors.OpPrereqError("Only one %s add or remove operation is"
12093 " supported at a time" % kind,
12094 errors.ECODE_INVAL)
12096 result.append((constants.DDM_MODIFY, op, params))
12098 assert verify_fn(result)
12105 def _CheckMods(kind, mods, key_types, item_fn):
12106 """Ensures requested disk/NIC modifications are valid.
12109 for (op, _, params) in mods:
12110 assert ht.TDict(params)
12112 utils.ForceDictType(params, key_types)
12114 if op == constants.DDM_REMOVE:
12116 raise errors.OpPrereqError("No settings should be passed when"
12117 " removing a %s" % kind,
12118 errors.ECODE_INVAL)
12119 elif op in (constants.DDM_ADD, constants.DDM_MODIFY):
12120 item_fn(op, params)
12122 raise errors.ProgrammerError("Unhandled operation '%s'" % op)
12125 def _VerifyDiskModification(op, params):
12126 """Verifies a disk modification.
12129 if op == constants.DDM_ADD:
12130 mode = params.setdefault(constants.IDISK_MODE, constants.DISK_RDWR)
12131 if mode not in constants.DISK_ACCESS_SET:
12132 raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode,
12133 errors.ECODE_INVAL)
12135 size = params.get(constants.IDISK_SIZE, None)
12137 raise errors.OpPrereqError("Required disk parameter '%s' missing" %
12138 constants.IDISK_SIZE, errors.ECODE_INVAL)
12142 except (TypeError, ValueError), err:
12143 raise errors.OpPrereqError("Invalid disk size parameter: %s" % err,
12144 errors.ECODE_INVAL)
12146 params[constants.IDISK_SIZE] = size
12148 elif op == constants.DDM_MODIFY and constants.IDISK_SIZE in params:
12149 raise errors.OpPrereqError("Disk size change not possible, use"
12150 " grow-disk", errors.ECODE_INVAL)
12153 def _VerifyNicModification(op, params):
12154 """Verifies a network interface modification.
12157 if op in (constants.DDM_ADD, constants.DDM_MODIFY):
12158 ip = params.get(constants.INIC_IP, None)
12161 elif ip.lower() == constants.VALUE_NONE:
12162 params[constants.INIC_IP] = None
12163 elif not netutils.IPAddress.IsValid(ip):
12164 raise errors.OpPrereqError("Invalid IP address '%s'" % ip,
12165 errors.ECODE_INVAL)
12167 bridge = params.get("bridge", None)
12168 link = params.get(constants.INIC_LINK, None)
12169 if bridge and link:
12170 raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
12171 " at the same time", errors.ECODE_INVAL)
12172 elif bridge and bridge.lower() == constants.VALUE_NONE:
12173 params["bridge"] = None
12174 elif link and link.lower() == constants.VALUE_NONE:
12175 params[constants.INIC_LINK] = None
12177 if op == constants.DDM_ADD:
12178 macaddr = params.get(constants.INIC_MAC, None)
12179 if macaddr is None:
12180 params[constants.INIC_MAC] = constants.VALUE_AUTO
12182 if constants.INIC_MAC in params:
12183 macaddr = params[constants.INIC_MAC]
12184 if macaddr not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
12185 macaddr = utils.NormalizeAndValidateMac(macaddr)
12187 if op == constants.DDM_MODIFY and macaddr == constants.VALUE_AUTO:
12188 raise errors.OpPrereqError("'auto' is not a valid MAC address when"
12189 " modifying an existing NIC",
12190 errors.ECODE_INVAL)
12192 def CheckArguments(self):
12193 if not (self.op.nics or self.op.disks or self.op.disk_template or
12194 self.op.hvparams or self.op.beparams or self.op.os_name or
12195 self.op.offline is not None or self.op.runtime_mem):
12196 raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
12198 if self.op.hvparams:
12199 _CheckGlobalHvParams(self.op.hvparams)
12202 self._UpgradeDiskNicMods("disk", self.op.disks,
12203 opcodes.OpInstanceSetParams.TestDiskModifications)
12205 self._UpgradeDiskNicMods("NIC", self.op.nics,
12206 opcodes.OpInstanceSetParams.TestNicModifications)
12208 # Check disk modifications
12209 self._CheckMods("disk", self.op.disks, constants.IDISK_PARAMS_TYPES,
12210 self._VerifyDiskModification)
12212 if self.op.disks and self.op.disk_template is not None:
12213 raise errors.OpPrereqError("Disk template conversion and other disk"
12214 " changes not supported at the same time",
12215 errors.ECODE_INVAL)
12217 if (self.op.disk_template and
12218 self.op.disk_template in constants.DTS_INT_MIRROR and
12219 self.op.remote_node is None):
12220 raise errors.OpPrereqError("Changing the disk template to a mirrored"
12221 " one requires specifying a secondary node",
12222 errors.ECODE_INVAL)
12224 # Check NIC modifications
12225 self._CheckMods("NIC", self.op.nics, constants.INIC_PARAMS_TYPES,
12226 self._VerifyNicModification)
12228 def ExpandNames(self):
12229 self._ExpandAndLockInstance()
12230 # Can't even acquire node locks in shared mode as upcoming changes in
12231 # Ganeti 2.6 will start to modify the node object on disk conversion
12232 self.needed_locks[locking.LEVEL_NODE] = []
12233 self.needed_locks[locking.LEVEL_NODE_RES] = []
12234 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
12236 def DeclareLocks(self, level):
12237 # TODO: Acquire group lock in shared mode (disk parameters)
12238 if level == locking.LEVEL_NODE:
12239 self._LockInstancesNodes()
12240 if self.op.disk_template and self.op.remote_node:
12241 self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
12242 self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
12243 elif level == locking.LEVEL_NODE_RES and self.op.disk_template:
12245 self.needed_locks[locking.LEVEL_NODE_RES] = \
12246 self.needed_locks[locking.LEVEL_NODE][:]
12248 def BuildHooksEnv(self):
12249 """Build hooks env.
12251 This runs on the master, primary and secondaries.
12255 if constants.BE_MINMEM in self.be_new:
12256 args["minmem"] = self.be_new[constants.BE_MINMEM]
12257 if constants.BE_MAXMEM in self.be_new:
12258 args["maxmem"] = self.be_new[constants.BE_MAXMEM]
12259 if constants.BE_VCPUS in self.be_new:
12260 args["vcpus"] = self.be_new[constants.BE_VCPUS]
12261 # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
12262 # information at all.
12264 if self._new_nics is not None:
12267 for nic in self._new_nics:
12268 nicparams = self.cluster.SimpleFillNIC(nic.nicparams)
12269 mode = nicparams[constants.NIC_MODE]
12270 link = nicparams[constants.NIC_LINK]
12271 nics.append((nic.ip, nic.mac, mode, link))
12273 args["nics"] = nics
12275 env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
12276 if self.op.disk_template:
12277 env["NEW_DISK_TEMPLATE"] = self.op.disk_template
12278 if self.op.runtime_mem:
12279 env["RUNTIME_MEMORY"] = self.op.runtime_mem
12283 def BuildHooksNodes(self):
12284 """Build hooks nodes.
12287 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
12290 def _PrepareNicModification(self, params, private, old_ip, old_params,
12292 update_params_dict = dict([(key, params[key])
12293 for key in constants.NICS_PARAMETERS
12296 if "bridge" in params:
12297 update_params_dict[constants.NIC_LINK] = params["bridge"]
12299 new_params = _GetUpdatedParams(old_params, update_params_dict)
12300 utils.ForceDictType(new_params, constants.NICS_PARAMETER_TYPES)
12302 new_filled_params = cluster.SimpleFillNIC(new_params)
12303 objects.NIC.CheckParameterSyntax(new_filled_params)
12305 new_mode = new_filled_params[constants.NIC_MODE]
12306 if new_mode == constants.NIC_MODE_BRIDGED:
12307 bridge = new_filled_params[constants.NIC_LINK]
12308 msg = self.rpc.call_bridges_exist(pnode, [bridge]).fail_msg
12310 msg = "Error checking bridges on node '%s': %s" % (pnode, msg)
12312 self.warn.append(msg)
12314 raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
12316 elif new_mode == constants.NIC_MODE_ROUTED:
12317 ip = params.get(constants.INIC_IP, old_ip)
12319 raise errors.OpPrereqError("Cannot set the NIC IP address to None"
12320 " on a routed NIC", errors.ECODE_INVAL)
12322 if constants.INIC_MAC in params:
12323 mac = params[constants.INIC_MAC]
12325 raise errors.OpPrereqError("Cannot unset the NIC MAC address",
12326 errors.ECODE_INVAL)
12327 elif mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
12328 # otherwise generate the MAC address
12329 params[constants.INIC_MAC] = \
12330 self.cfg.GenerateMAC(self.proc.GetECId())
12332 # or validate/reserve the current one
12334 self.cfg.ReserveMAC(mac, self.proc.GetECId())
12335 except errors.ReservationError:
12336 raise errors.OpPrereqError("MAC address '%s' already in use"
12337 " in cluster" % mac,
12338 errors.ECODE_NOTUNIQUE)
12340 private.params = new_params
12341 private.filled = new_filled_params
12343 def CheckPrereq(self):
12344 """Check prerequisites.
12346 This only checks the instance list against the existing names.
12349 # checking the new params on the primary/secondary nodes
12351 instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
12352 cluster = self.cluster = self.cfg.GetClusterInfo()
12353 assert self.instance is not None, \
12354 "Cannot retrieve locked instance %s" % self.op.instance_name
12355 pnode = instance.primary_node
12356 nodelist = list(instance.all_nodes)
12357 pnode_info = self.cfg.GetNodeInfo(pnode)
12358 self.diskparams = self.cfg.GetInstanceDiskParams(instance)
12360 # Prepare disk/NIC modifications
12361 self.diskmod = PrepareContainerMods(self.op.disks, None)
12362 self.nicmod = PrepareContainerMods(self.op.nics, _InstNicModPrivate)
12365 if self.op.os_name and not self.op.force:
12366 _CheckNodeHasOS(self, instance.primary_node, self.op.os_name,
12367 self.op.force_variant)
12368 instance_os = self.op.os_name
12370 instance_os = instance.os
12372 assert not (self.op.disk_template and self.op.disks), \
12373 "Can't modify disk template and apply disk changes at the same time"
12375 if self.op.disk_template:
12376 if instance.disk_template == self.op.disk_template:
12377 raise errors.OpPrereqError("Instance already has disk template %s" %
12378 instance.disk_template, errors.ECODE_INVAL)
12380 if (instance.disk_template,
12381 self.op.disk_template) not in self._DISK_CONVERSIONS:
12382 raise errors.OpPrereqError("Unsupported disk template conversion from"
12383 " %s to %s" % (instance.disk_template,
12384 self.op.disk_template),
12385 errors.ECODE_INVAL)
12386 _CheckInstanceState(self, instance, INSTANCE_DOWN,
12387 msg="cannot change disk template")
12388 if self.op.disk_template in constants.DTS_INT_MIRROR:
12389 if self.op.remote_node == pnode:
12390 raise errors.OpPrereqError("Given new secondary node %s is the same"
12391 " as the primary node of the instance" %
12392 self.op.remote_node, errors.ECODE_STATE)
12393 _CheckNodeOnline(self, self.op.remote_node)
12394 _CheckNodeNotDrained(self, self.op.remote_node)
12395 # FIXME: here we assume that the old instance type is DT_PLAIN
12396 assert instance.disk_template == constants.DT_PLAIN
12397 disks = [{constants.IDISK_SIZE: d.size,
12398 constants.IDISK_VG: d.logical_id[0]}
12399 for d in instance.disks]
12400 required = _ComputeDiskSizePerVG(self.op.disk_template, disks)
12401 _CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required)
12403 snode_info = self.cfg.GetNodeInfo(self.op.remote_node)
12404 snode_group = self.cfg.GetNodeGroup(snode_info.group)
12405 ipolicy = _CalculateGroupIPolicy(cluster, snode_group)
12406 _CheckTargetNodeIPolicy(self, ipolicy, instance, snode_info,
12407 ignore=self.op.ignore_ipolicy)
12408 if pnode_info.group != snode_info.group:
12409 self.LogWarning("The primary and secondary nodes are in two"
12410 " different node groups; the disk parameters"
12411 " from the first disk's node group will be"
12414 # hvparams processing
12415 if self.op.hvparams:
12416 hv_type = instance.hypervisor
12417 i_hvdict = _GetUpdatedParams(instance.hvparams, self.op.hvparams)
12418 utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
12419 hv_new = cluster.SimpleFillHV(hv_type, instance.os, i_hvdict)
12422 hypervisor.GetHypervisor(hv_type).CheckParameterSyntax(hv_new)
12423 _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
12424 self.hv_proposed = self.hv_new = hv_new # the new actual values
12425 self.hv_inst = i_hvdict # the new dict (without defaults)
12427 self.hv_proposed = cluster.SimpleFillHV(instance.hypervisor, instance.os,
12429 self.hv_new = self.hv_inst = {}
12431 # beparams processing
12432 if self.op.beparams:
12433 i_bedict = _GetUpdatedParams(instance.beparams, self.op.beparams,
12435 objects.UpgradeBeParams(i_bedict)
12436 utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
12437 be_new = cluster.SimpleFillBE(i_bedict)
12438 self.be_proposed = self.be_new = be_new # the new actual values
12439 self.be_inst = i_bedict # the new dict (without defaults)
12441 self.be_new = self.be_inst = {}
12442 self.be_proposed = cluster.SimpleFillBE(instance.beparams)
12443 be_old = cluster.FillBE(instance)
12445 # CPU param validation -- checking every time a parameter is
12446 # changed to cover all cases where either CPU mask or vcpus have
12448 if (constants.BE_VCPUS in self.be_proposed and
12449 constants.HV_CPU_MASK in self.hv_proposed):
12451 utils.ParseMultiCpuMask(self.hv_proposed[constants.HV_CPU_MASK])
12452 # Verify mask is consistent with number of vCPUs. Can skip this
12453 # test if only 1 entry in the CPU mask, which means same mask
12454 # is applied to all vCPUs.
12455 if (len(cpu_list) > 1 and
12456 len(cpu_list) != self.be_proposed[constants.BE_VCPUS]):
12457 raise errors.OpPrereqError("Number of vCPUs [%d] does not match the"
12459 (self.be_proposed[constants.BE_VCPUS],
12460 self.hv_proposed[constants.HV_CPU_MASK]),
12461 errors.ECODE_INVAL)
12463 # Only perform this test if a new CPU mask is given
12464 if constants.HV_CPU_MASK in self.hv_new:
12465 # Calculate the largest CPU number requested
12466 max_requested_cpu = max(map(max, cpu_list))
12467 # Check that all of the instance's nodes have enough physical CPUs to
12468 # satisfy the requested CPU mask
12469 _CheckNodesPhysicalCPUs(self, instance.all_nodes,
12470 max_requested_cpu + 1, instance.hypervisor)
12472 # osparams processing
12473 if self.op.osparams:
12474 i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
12475 _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
12476 self.os_inst = i_osdict # the new dict (without defaults)
12482 #TODO(dynmem): do the appropriate check involving MINMEM
12483 if (constants.BE_MAXMEM in self.op.beparams and not self.op.force and
12484 be_new[constants.BE_MAXMEM] > be_old[constants.BE_MAXMEM]):
12485 mem_check_list = [pnode]
12486 if be_new[constants.BE_AUTO_BALANCE]:
12487 # either we changed auto_balance to yes or it was from before
12488 mem_check_list.extend(instance.secondary_nodes)
12489 instance_info = self.rpc.call_instance_info(pnode, instance.name,
12490 instance.hypervisor)
12491 nodeinfo = self.rpc.call_node_info(mem_check_list, None,
12492 [instance.hypervisor])
12493 pninfo = nodeinfo[pnode]
12494 msg = pninfo.fail_msg
12496 # Assume the primary node is unreachable and go ahead
12497 self.warn.append("Can't get info from primary node %s: %s" %
12500 (_, _, (pnhvinfo, )) = pninfo.payload
12501 if not isinstance(pnhvinfo.get("memory_free", None), int):
12502 self.warn.append("Node data from primary node %s doesn't contain"
12503 " free memory information" % pnode)
12504 elif instance_info.fail_msg:
12505 self.warn.append("Can't get instance runtime information: %s" %
12506 instance_info.fail_msg)
12508 if instance_info.payload:
12509 current_mem = int(instance_info.payload["memory"])
12511 # Assume instance not running
12512 # (there is a slight race condition here, but it's not very
12513 # probable, and we have no other way to check)
12514 # TODO: Describe race condition
12516 #TODO(dynmem): do the appropriate check involving MINMEM
12517 miss_mem = (be_new[constants.BE_MAXMEM] - current_mem -
12518 pnhvinfo["memory_free"])
12520 raise errors.OpPrereqError("This change will prevent the instance"
12521 " from starting, due to %d MB of memory"
12522 " missing on its primary node" %
12524 errors.ECODE_NORES)
12526 if be_new[constants.BE_AUTO_BALANCE]:
12527 for node, nres in nodeinfo.items():
12528 if node not in instance.secondary_nodes:
12530 nres.Raise("Can't get info from secondary node %s" % node,
12531 prereq=True, ecode=errors.ECODE_STATE)
12532 (_, _, (nhvinfo, )) = nres.payload
12533 if not isinstance(nhvinfo.get("memory_free", None), int):
12534 raise errors.OpPrereqError("Secondary node %s didn't return free"
12535 " memory information" % node,
12536 errors.ECODE_STATE)
12537 #TODO(dynmem): do the appropriate check involving MINMEM
12538 elif be_new[constants.BE_MAXMEM] > nhvinfo["memory_free"]:
12539 raise errors.OpPrereqError("This change will prevent the instance"
12540 " from failover to its secondary node"
12541 " %s, due to not enough memory" % node,
12542 errors.ECODE_STATE)
12544 if self.op.runtime_mem:
12545 remote_info = self.rpc.call_instance_info(instance.primary_node,
12547 instance.hypervisor)
12548 remote_info.Raise("Error checking node %s" % instance.primary_node)
12549 if not remote_info.payload: # not running already
12550 raise errors.OpPrereqError("Instance %s is not running" % instance.name,
12551 errors.ECODE_STATE)
12553 current_memory = remote_info.payload["memory"]
12554 if (not self.op.force and
12555 (self.op.runtime_mem > self.be_proposed[constants.BE_MAXMEM] or
12556 self.op.runtime_mem < self.be_proposed[constants.BE_MINMEM])):
12557 raise errors.OpPrereqError("Instance %s must have memory between %d"
12558 " and %d MB of memory unless --force is"
12559 " given" % (instance.name,
12560 self.be_proposed[constants.BE_MINMEM],
12561 self.be_proposed[constants.BE_MAXMEM]),
12562 errors.ECODE_INVAL)
12564 if self.op.runtime_mem > current_memory:
12565 _CheckNodeFreeMemory(self, instance.primary_node,
12566 "ballooning memory for instance %s" %
12568 self.op.memory - current_memory,
12569 instance.hypervisor)
12571 if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
12572 raise errors.OpPrereqError("Disk operations not supported for"
12573 " diskless instances",
12574 errors.ECODE_INVAL)
12576 def _PrepareNicCreate(_, params, private):
12577 self._PrepareNicModification(params, private, None, {}, cluster, pnode)
12578 return (None, None)
12580 def _PrepareNicMod(_, nic, params, private):
12581 self._PrepareNicModification(params, private, nic.ip,
12582 nic.nicparams, cluster, pnode)
12585 # Verify NIC changes (operating on copy)
12586 nics = instance.nics[:]
12587 ApplyContainerMods("NIC", nics, None, self.nicmod,
12588 _PrepareNicCreate, _PrepareNicMod, None)
12589 if len(nics) > constants.MAX_NICS:
12590 raise errors.OpPrereqError("Instance has too many network interfaces"
12591 " (%d), cannot add more" % constants.MAX_NICS,
12592 errors.ECODE_STATE)
12594 # Verify disk changes (operating on a copy)
12595 disks = instance.disks[:]
12596 ApplyContainerMods("disk", disks, None, self.diskmod, None, None, None)
12597 if len(disks) > constants.MAX_DISKS:
12598 raise errors.OpPrereqError("Instance has too many disks (%d), cannot add"
12599 " more" % constants.MAX_DISKS,
12600 errors.ECODE_STATE)
12602 if self.op.offline is not None:
12603 if self.op.offline:
12604 msg = "can't change to offline"
12606 msg = "can't change to online"
12607 _CheckInstanceState(self, instance, CAN_CHANGE_INSTANCE_OFFLINE, msg=msg)
12609 # Pre-compute NIC changes (necessary to use result in hooks)
12610 self._nic_chgdesc = []
12612 # Operate on copies as this is still in prereq
12613 nics = [nic.Copy() for nic in instance.nics]
12614 ApplyContainerMods("NIC", nics, self._nic_chgdesc, self.nicmod,
12615 self._CreateNewNic, self._ApplyNicMods, None)
12616 self._new_nics = nics
12618 self._new_nics = None
12620 def _ConvertPlainToDrbd(self, feedback_fn):
12621 """Converts an instance from plain to drbd.
12624 feedback_fn("Converting template to drbd")
12625 instance = self.instance
12626 pnode = instance.primary_node
12627 snode = self.op.remote_node
12629 assert instance.disk_template == constants.DT_PLAIN
12631 # create a fake disk info for _GenerateDiskTemplate
12632 disk_info = [{constants.IDISK_SIZE: d.size, constants.IDISK_MODE: d.mode,
12633 constants.IDISK_VG: d.logical_id[0]}
12634 for d in instance.disks]
12635 new_disks = _GenerateDiskTemplate(self, self.op.disk_template,
12636 instance.name, pnode, [snode],
12637 disk_info, None, None, 0, feedback_fn,
12639 anno_disks = rpc.AnnotateDiskParams(constants.DT_DRBD8, new_disks,
12641 info = _GetInstanceInfoText(instance)
12642 feedback_fn("Creating additional volumes...")
12643 # first, create the missing data and meta devices
12644 for disk in anno_disks:
12645 # unfortunately this is... not too nice
12646 _CreateSingleBlockDev(self, pnode, instance, disk.children[1],
12648 for child in disk.children:
12649 _CreateSingleBlockDev(self, snode, instance, child, info, True)
12650 # at this stage, all new LVs have been created, we can rename the
12652 feedback_fn("Renaming original volumes...")
12653 rename_list = [(o, n.children[0].logical_id)
12654 for (o, n) in zip(instance.disks, new_disks)]
12655 result = self.rpc.call_blockdev_rename(pnode, rename_list)
12656 result.Raise("Failed to rename original LVs")
12658 feedback_fn("Initializing DRBD devices...")
12659 # all child devices are in place, we can now create the DRBD devices
12660 for disk in anno_disks:
12661 for node in [pnode, snode]:
12662 f_create = node == pnode
12663 _CreateSingleBlockDev(self, node, instance, disk, info, f_create)
12665 # at this point, the instance has been modified
12666 instance.disk_template = constants.DT_DRBD8
12667 instance.disks = new_disks
12668 self.cfg.Update(instance, feedback_fn)
12670 # Release node locks while waiting for sync
12671 _ReleaseLocks(self, locking.LEVEL_NODE)
12673 # disks are created, waiting for sync
12674 disk_abort = not _WaitForSync(self, instance,
12675 oneshot=not self.op.wait_for_sync)
12677 raise errors.OpExecError("There are some degraded disks for"
12678 " this instance, please cleanup manually")
12680 # Node resource locks will be released by caller
12682 def _ConvertDrbdToPlain(self, feedback_fn):
12683 """Converts an instance from drbd to plain.
12686 instance = self.instance
12688 assert len(instance.secondary_nodes) == 1
12689 assert instance.disk_template == constants.DT_DRBD8
12691 pnode = instance.primary_node
12692 snode = instance.secondary_nodes[0]
12693 feedback_fn("Converting template to plain")
12695 old_disks = _AnnotateDiskParams(instance, instance.disks, self.cfg)
12696 new_disks = [d.children[0] for d in instance.disks]
12698 # copy over size and mode
12699 for parent, child in zip(old_disks, new_disks):
12700 child.size = parent.size
12701 child.mode = parent.mode
12703 # this is a DRBD disk, return its port to the pool
12704 # NOTE: this must be done right before the call to cfg.Update!
12705 for disk in old_disks:
12706 tcp_port = disk.logical_id[2]
12707 self.cfg.AddTcpUdpPort(tcp_port)
12709 # update instance structure
12710 instance.disks = new_disks
12711 instance.disk_template = constants.DT_PLAIN
12712 self.cfg.Update(instance, feedback_fn)
12714 # Release locks in case removing disks takes a while
12715 _ReleaseLocks(self, locking.LEVEL_NODE)
12717 feedback_fn("Removing volumes on the secondary node...")
12718 for disk in old_disks:
12719 self.cfg.SetDiskID(disk, snode)
12720 msg = self.rpc.call_blockdev_remove(snode, disk).fail_msg
12722 self.LogWarning("Could not remove block device %s on node %s,"
12723 " continuing anyway: %s", disk.iv_name, snode, msg)
12725 feedback_fn("Removing unneeded volumes on the primary node...")
12726 for idx, disk in enumerate(old_disks):
12727 meta = disk.children[1]
12728 self.cfg.SetDiskID(meta, pnode)
12729 msg = self.rpc.call_blockdev_remove(pnode, meta).fail_msg
12731 self.LogWarning("Could not remove metadata for disk %d on node %s,"
12732 " continuing anyway: %s", idx, pnode, msg)
12734 def _CreateNewDisk(self, idx, params, _):
12735 """Creates a new disk.
12738 instance = self.instance
12741 if instance.disk_template in constants.DTS_FILEBASED:
12742 (file_driver, file_path) = instance.disks[0].logical_id
12743 file_path = os.path.dirname(file_path)
12745 file_driver = file_path = None
12748 _GenerateDiskTemplate(self, instance.disk_template, instance.name,
12749 instance.primary_node, instance.secondary_nodes,
12750 [params], file_path, file_driver, idx,
12751 self.Log, self.diskparams)[0]
12753 info = _GetInstanceInfoText(instance)
12755 logging.info("Creating volume %s for instance %s",
12756 disk.iv_name, instance.name)
12757 # Note: this needs to be kept in sync with _CreateDisks
12759 for node in instance.all_nodes:
12760 f_create = (node == instance.primary_node)
12762 _CreateBlockDev(self, node, instance, disk, f_create, info, f_create)
12763 except errors.OpExecError, err:
12764 self.LogWarning("Failed to create volume %s (%s) on node '%s': %s",
12765 disk.iv_name, disk, node, err)
12768 ("disk/%d" % idx, "add:size=%s,mode=%s" % (disk.size, disk.mode)),
12772 def _ModifyDisk(idx, disk, params, _):
12773 """Modifies a disk.
12776 disk.mode = params[constants.IDISK_MODE]
12779 ("disk.mode/%d" % idx, disk.mode),
12782 def _RemoveDisk(self, idx, root, _):
12786 (anno_disk,) = _AnnotateDiskParams(self.instance, [root], self.cfg)
12787 for node, disk in anno_disk.ComputeNodeTree(self.instance.primary_node):
12788 self.cfg.SetDiskID(disk, node)
12789 msg = self.rpc.call_blockdev_remove(node, disk).fail_msg
12791 self.LogWarning("Could not remove disk/%d on node '%s': %s,"
12792 " continuing anyway", idx, node, msg)
12794 # if this is a DRBD disk, return its port to the pool
12795 if root.dev_type in constants.LDS_DRBD:
12796 self.cfg.AddTcpUdpPort(root.logical_id[2])
12799 def _CreateNewNic(idx, params, private):
12800 """Creates data structure for a new network interface.
12803 mac = params[constants.INIC_MAC]
12804 ip = params.get(constants.INIC_IP, None)
12805 nicparams = private.params
12807 return (objects.NIC(mac=mac, ip=ip, nicparams=nicparams), [
12809 "add:mac=%s,ip=%s,mode=%s,link=%s" %
12810 (mac, ip, private.filled[constants.NIC_MODE],
12811 private.filled[constants.NIC_LINK])),
12815 def _ApplyNicMods(idx, nic, params, private):
12816 """Modifies a network interface.
12821 for key in [constants.INIC_MAC, constants.INIC_IP]:
12823 changes.append(("nic.%s/%d" % (key, idx), params[key]))
12824 setattr(nic, key, params[key])
12827 nic.nicparams = private.params
12829 for (key, val) in params.items():
12830 changes.append(("nic.%s/%d" % (key, idx), val))
12834 def Exec(self, feedback_fn):
12835 """Modifies an instance.
12837 All parameters take effect only at the next restart of the instance.
12840 # Process here the warnings from CheckPrereq, as we don't have a
12841 # feedback_fn there.
12842 # TODO: Replace with self.LogWarning
12843 for warn in self.warn:
12844 feedback_fn("WARNING: %s" % warn)
12846 assert ((self.op.disk_template is None) ^
12847 bool(self.owned_locks(locking.LEVEL_NODE_RES))), \
12848 "Not owning any node resource locks"
12851 instance = self.instance
12854 if self.op.runtime_mem:
12855 rpcres = self.rpc.call_instance_balloon_memory(instance.primary_node,
12857 self.op.runtime_mem)
12858 rpcres.Raise("Cannot modify instance runtime memory")
12859 result.append(("runtime_memory", self.op.runtime_mem))
12861 # Apply disk changes
12862 ApplyContainerMods("disk", instance.disks, result, self.diskmod,
12863 self._CreateNewDisk, self._ModifyDisk, self._RemoveDisk)
12864 _UpdateIvNames(0, instance.disks)
12866 if self.op.disk_template:
12868 check_nodes = set(instance.all_nodes)
12869 if self.op.remote_node:
12870 check_nodes.add(self.op.remote_node)
12871 for level in [locking.LEVEL_NODE, locking.LEVEL_NODE_RES]:
12872 owned = self.owned_locks(level)
12873 assert not (check_nodes - owned), \
12874 ("Not owning the correct locks, owning %r, expected at least %r" %
12875 (owned, check_nodes))
12877 r_shut = _ShutdownInstanceDisks(self, instance)
12879 raise errors.OpExecError("Cannot shutdown instance disks, unable to"
12880 " proceed with disk template conversion")
12881 mode = (instance.disk_template, self.op.disk_template)
12883 self._DISK_CONVERSIONS[mode](self, feedback_fn)
12885 self.cfg.ReleaseDRBDMinors(instance.name)
12887 result.append(("disk_template", self.op.disk_template))
12889 assert instance.disk_template == self.op.disk_template, \
12890 ("Expected disk template '%s', found '%s'" %
12891 (self.op.disk_template, instance.disk_template))
12893 # Release node and resource locks if there are any (they might already have
12894 # been released during disk conversion)
12895 _ReleaseLocks(self, locking.LEVEL_NODE)
12896 _ReleaseLocks(self, locking.LEVEL_NODE_RES)
12898 # Apply NIC changes
12899 if self._new_nics is not None:
12900 instance.nics = self._new_nics
12901 result.extend(self._nic_chgdesc)
12904 if self.op.hvparams:
12905 instance.hvparams = self.hv_inst
12906 for key, val in self.op.hvparams.iteritems():
12907 result.append(("hv/%s" % key, val))
12910 if self.op.beparams:
12911 instance.beparams = self.be_inst
12912 for key, val in self.op.beparams.iteritems():
12913 result.append(("be/%s" % key, val))
12916 if self.op.os_name:
12917 instance.os = self.op.os_name
12920 if self.op.osparams:
12921 instance.osparams = self.os_inst
12922 for key, val in self.op.osparams.iteritems():
12923 result.append(("os/%s" % key, val))
12925 if self.op.offline is None:
12928 elif self.op.offline:
12929 # Mark instance as offline
12930 self.cfg.MarkInstanceOffline(instance.name)
12931 result.append(("admin_state", constants.ADMINST_OFFLINE))
12933 # Mark instance as online, but stopped
12934 self.cfg.MarkInstanceDown(instance.name)
12935 result.append(("admin_state", constants.ADMINST_DOWN))
12937 self.cfg.Update(instance, feedback_fn)
12939 assert not (self.owned_locks(locking.LEVEL_NODE_RES) or
12940 self.owned_locks(locking.LEVEL_NODE)), \
12941 "All node locks should have been released by now"
12945 _DISK_CONVERSIONS = {
12946 (constants.DT_PLAIN, constants.DT_DRBD8): _ConvertPlainToDrbd,
12947 (constants.DT_DRBD8, constants.DT_PLAIN): _ConvertDrbdToPlain,
12951 class LUInstanceChangeGroup(LogicalUnit):
12952 HPATH = "instance-change-group"
12953 HTYPE = constants.HTYPE_INSTANCE
12956 def ExpandNames(self):
12957 self.share_locks = _ShareAll()
12958 self.needed_locks = {
12959 locking.LEVEL_NODEGROUP: [],
12960 locking.LEVEL_NODE: [],
12963 self._ExpandAndLockInstance()
12965 if self.op.target_groups:
12966 self.req_target_uuids = map(self.cfg.LookupNodeGroup,
12967 self.op.target_groups)
12969 self.req_target_uuids = None
12971 self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
12973 def DeclareLocks(self, level):
12974 if level == locking.LEVEL_NODEGROUP:
12975 assert not self.needed_locks[locking.LEVEL_NODEGROUP]
12977 if self.req_target_uuids:
12978 lock_groups = set(self.req_target_uuids)
12980 # Lock all groups used by instance optimistically; this requires going
12981 # via the node before it's locked, requiring verification later on
12982 instance_groups = self.cfg.GetInstanceNodeGroups(self.op.instance_name)
12983 lock_groups.update(instance_groups)
12985 # No target groups, need to lock all of them
12986 lock_groups = locking.ALL_SET
12988 self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
12990 elif level == locking.LEVEL_NODE:
12991 if self.req_target_uuids:
12992 # Lock all nodes used by instances
12993 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
12994 self._LockInstancesNodes()
12996 # Lock all nodes in all potential target groups
12997 lock_groups = (frozenset(self.owned_locks(locking.LEVEL_NODEGROUP)) -
12998 self.cfg.GetInstanceNodeGroups(self.op.instance_name))
12999 member_nodes = [node_name
13000 for group in lock_groups
13001 for node_name in self.cfg.GetNodeGroup(group).members]
13002 self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
13004 # Lock all nodes as all groups are potential targets
13005 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
13007 def CheckPrereq(self):
13008 owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
13009 owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
13010 owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
13012 assert (self.req_target_uuids is None or
13013 owned_groups.issuperset(self.req_target_uuids))
13014 assert owned_instances == set([self.op.instance_name])
13016 # Get instance information
13017 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
13019 # Check if node groups for locked instance are still correct
13020 assert owned_nodes.issuperset(self.instance.all_nodes), \
13021 ("Instance %s's nodes changed while we kept the lock" %
13022 self.op.instance_name)
13024 inst_groups = _CheckInstanceNodeGroups(self.cfg, self.op.instance_name,
13027 if self.req_target_uuids:
13028 # User requested specific target groups
13029 self.target_uuids = frozenset(self.req_target_uuids)
13031 # All groups except those used by the instance are potential targets
13032 self.target_uuids = owned_groups - inst_groups
13034 conflicting_groups = self.target_uuids & inst_groups
13035 if conflicting_groups:
13036 raise errors.OpPrereqError("Can't use group(s) '%s' as targets, they are"
13037 " used by the instance '%s'" %
13038 (utils.CommaJoin(conflicting_groups),
13039 self.op.instance_name),
13040 errors.ECODE_INVAL)
13042 if not self.target_uuids:
13043 raise errors.OpPrereqError("There are no possible target groups",
13044 errors.ECODE_INVAL)
13046 def BuildHooksEnv(self):
13047 """Build hooks env.
13050 assert self.target_uuids
13053 "TARGET_GROUPS": " ".join(self.target_uuids),
13056 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
13060 def BuildHooksNodes(self):
13061 """Build hooks nodes.
13064 mn = self.cfg.GetMasterNode()
13065 return ([mn], [mn])
13067 def Exec(self, feedback_fn):
13068 instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
13070 assert instances == [self.op.instance_name], "Instance not locked"
13072 ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_CHG_GROUP,
13073 instances=instances, target_groups=list(self.target_uuids))
13075 ial.Run(self.op.iallocator)
13077 if not ial.success:
13078 raise errors.OpPrereqError("Can't compute solution for changing group of"
13079 " instance '%s' using iallocator '%s': %s" %
13080 (self.op.instance_name, self.op.iallocator,
13082 errors.ECODE_NORES)
13084 jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, False)
13086 self.LogInfo("Iallocator returned %s job(s) for changing group of"
13087 " instance '%s'", len(jobs), self.op.instance_name)
13089 return ResultWithJobs(jobs)
13092 class LUBackupQuery(NoHooksLU):
13093 """Query the exports list
13098 def CheckArguments(self):
13099 self.expq = _ExportQuery(qlang.MakeSimpleFilter("node", self.op.nodes),
13100 ["node", "export"], self.op.use_locking)
13102 def ExpandNames(self):
13103 self.expq.ExpandNames(self)
13105 def DeclareLocks(self, level):
13106 self.expq.DeclareLocks(self, level)
13108 def Exec(self, feedback_fn):
13111 for (node, expname) in self.expq.OldStyleQuery(self):
13112 if expname is None:
13113 result[node] = False
13115 result.setdefault(node, []).append(expname)
13120 class _ExportQuery(_QueryBase):
13121 FIELDS = query.EXPORT_FIELDS
13123 #: The node name is not a unique key for this query
13124 SORT_FIELD = "node"
13126 def ExpandNames(self, lu):
13127 lu.needed_locks = {}
13129 # The following variables interact with _QueryBase._GetNames
13131 self.wanted = _GetWantedNodes(lu, self.names)
13133 self.wanted = locking.ALL_SET
13135 self.do_locking = self.use_locking
13137 if self.do_locking:
13138 lu.share_locks = _ShareAll()
13139 lu.needed_locks = {
13140 locking.LEVEL_NODE: self.wanted,
13143 def DeclareLocks(self, lu, level):
13146 def _GetQueryData(self, lu):
13147 """Computes the list of nodes and their attributes.
13150 # Locking is not used
13152 assert not (compat.any(lu.glm.is_owned(level)
13153 for level in locking.LEVELS
13154 if level != locking.LEVEL_CLUSTER) or
13155 self.do_locking or self.use_locking)
13157 nodes = self._GetNames(lu, lu.cfg.GetNodeList(), locking.LEVEL_NODE)
13161 for (node, nres) in lu.rpc.call_export_list(nodes).items():
13163 result.append((node, None))
13165 result.extend((node, expname) for expname in nres.payload)
13170 class LUBackupPrepare(NoHooksLU):
13171 """Prepares an instance for an export and returns useful information.
13176 def ExpandNames(self):
13177 self._ExpandAndLockInstance()
13179 def CheckPrereq(self):
13180 """Check prerequisites.
13183 instance_name = self.op.instance_name
13185 self.instance = self.cfg.GetInstanceInfo(instance_name)
13186 assert self.instance is not None, \
13187 "Cannot retrieve locked instance %s" % self.op.instance_name
13188 _CheckNodeOnline(self, self.instance.primary_node)
13190 self._cds = _GetClusterDomainSecret()
13192 def Exec(self, feedback_fn):
13193 """Prepares an instance for an export.
13196 instance = self.instance
13198 if self.op.mode == constants.EXPORT_MODE_REMOTE:
13199 salt = utils.GenerateSecret(8)
13201 feedback_fn("Generating X509 certificate on %s" % instance.primary_node)
13202 result = self.rpc.call_x509_cert_create(instance.primary_node,
13203 constants.RIE_CERT_VALIDITY)
13204 result.Raise("Can't create X509 key and certificate on %s" % result.node)
13206 (name, cert_pem) = result.payload
13208 cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
13212 "handshake": masterd.instance.ComputeRemoteExportHandshake(self._cds),
13213 "x509_key_name": (name, utils.Sha1Hmac(self._cds, name, salt=salt),
13215 "x509_ca": utils.SignX509Certificate(cert, self._cds, salt),
13221 class LUBackupExport(LogicalUnit):
13222 """Export an instance to an image in the cluster.
13225 HPATH = "instance-export"
13226 HTYPE = constants.HTYPE_INSTANCE
13229 def CheckArguments(self):
13230 """Check the arguments.
13233 self.x509_key_name = self.op.x509_key_name
13234 self.dest_x509_ca_pem = self.op.destination_x509_ca
13236 if self.op.mode == constants.EXPORT_MODE_REMOTE:
13237 if not self.x509_key_name:
13238 raise errors.OpPrereqError("Missing X509 key name for encryption",
13239 errors.ECODE_INVAL)
13241 if not self.dest_x509_ca_pem:
13242 raise errors.OpPrereqError("Missing destination X509 CA",
13243 errors.ECODE_INVAL)
13245 def ExpandNames(self):
13246 self._ExpandAndLockInstance()
13248 # Lock all nodes for local exports
13249 if self.op.mode == constants.EXPORT_MODE_LOCAL:
13250 # FIXME: lock only instance primary and destination node
13252 # Sad but true, for now we have do lock all nodes, as we don't know where
13253 # the previous export might be, and in this LU we search for it and
13254 # remove it from its current node. In the future we could fix this by:
13255 # - making a tasklet to search (share-lock all), then create the
13256 # new one, then one to remove, after
13257 # - removing the removal operation altogether
13258 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
13260 def DeclareLocks(self, level):
13261 """Last minute lock declaration."""
13262 # All nodes are locked anyway, so nothing to do here.
13264 def BuildHooksEnv(self):
13265 """Build hooks env.
13267 This will run on the master, primary node and target node.
13271 "EXPORT_MODE": self.op.mode,
13272 "EXPORT_NODE": self.op.target_node,
13273 "EXPORT_DO_SHUTDOWN": self.op.shutdown,
13274 "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
13275 # TODO: Generic function for boolean env variables
13276 "REMOVE_INSTANCE": str(bool(self.op.remove_instance)),
13279 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
13283 def BuildHooksNodes(self):
13284 """Build hooks nodes.
13287 nl = [self.cfg.GetMasterNode(), self.instance.primary_node]
13289 if self.op.mode == constants.EXPORT_MODE_LOCAL:
13290 nl.append(self.op.target_node)
13294 def CheckPrereq(self):
13295 """Check prerequisites.
13297 This checks that the instance and node names are valid.
13300 instance_name = self.op.instance_name
13302 self.instance = self.cfg.GetInstanceInfo(instance_name)
13303 assert self.instance is not None, \
13304 "Cannot retrieve locked instance %s" % self.op.instance_name
13305 _CheckNodeOnline(self, self.instance.primary_node)
13307 if (self.op.remove_instance and
13308 self.instance.admin_state == constants.ADMINST_UP and
13309 not self.op.shutdown):
13310 raise errors.OpPrereqError("Can not remove instance without shutting it"
13313 if self.op.mode == constants.EXPORT_MODE_LOCAL:
13314 self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
13315 self.dst_node = self.cfg.GetNodeInfo(self.op.target_node)
13316 assert self.dst_node is not None
13318 _CheckNodeOnline(self, self.dst_node.name)
13319 _CheckNodeNotDrained(self, self.dst_node.name)
13322 self.dest_disk_info = None
13323 self.dest_x509_ca = None
13325 elif self.op.mode == constants.EXPORT_MODE_REMOTE:
13326 self.dst_node = None
13328 if len(self.op.target_node) != len(self.instance.disks):
13329 raise errors.OpPrereqError(("Received destination information for %s"
13330 " disks, but instance %s has %s disks") %
13331 (len(self.op.target_node), instance_name,
13332 len(self.instance.disks)),
13333 errors.ECODE_INVAL)
13335 cds = _GetClusterDomainSecret()
13337 # Check X509 key name
13339 (key_name, hmac_digest, hmac_salt) = self.x509_key_name
13340 except (TypeError, ValueError), err:
13341 raise errors.OpPrereqError("Invalid data for X509 key name: %s" % err)
13343 if not utils.VerifySha1Hmac(cds, key_name, hmac_digest, salt=hmac_salt):
13344 raise errors.OpPrereqError("HMAC for X509 key name is wrong",
13345 errors.ECODE_INVAL)
13347 # Load and verify CA
13349 (cert, _) = utils.LoadSignedX509Certificate(self.dest_x509_ca_pem, cds)
13350 except OpenSSL.crypto.Error, err:
13351 raise errors.OpPrereqError("Unable to load destination X509 CA (%s)" %
13352 (err, ), errors.ECODE_INVAL)
13354 (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
13355 if errcode is not None:
13356 raise errors.OpPrereqError("Invalid destination X509 CA (%s)" %
13357 (msg, ), errors.ECODE_INVAL)
13359 self.dest_x509_ca = cert
13361 # Verify target information
13363 for idx, disk_data in enumerate(self.op.target_node):
13365 (host, port, magic) = \
13366 masterd.instance.CheckRemoteExportDiskInfo(cds, idx, disk_data)
13367 except errors.GenericError, err:
13368 raise errors.OpPrereqError("Target info for disk %s: %s" %
13369 (idx, err), errors.ECODE_INVAL)
13371 disk_info.append((host, port, magic))
13373 assert len(disk_info) == len(self.op.target_node)
13374 self.dest_disk_info = disk_info
13377 raise errors.ProgrammerError("Unhandled export mode %r" %
13380 # instance disk type verification
13381 # TODO: Implement export support for file-based disks
13382 for disk in self.instance.disks:
13383 if disk.dev_type == constants.LD_FILE:
13384 raise errors.OpPrereqError("Export not supported for instances with"
13385 " file-based disks", errors.ECODE_INVAL)
13387 def _CleanupExports(self, feedback_fn):
13388 """Removes exports of current instance from all other nodes.
13390 If an instance in a cluster with nodes A..D was exported to node C, its
13391 exports will be removed from the nodes A, B and D.
13394 assert self.op.mode != constants.EXPORT_MODE_REMOTE
13396 nodelist = self.cfg.GetNodeList()
13397 nodelist.remove(self.dst_node.name)
13399 # on one-node clusters nodelist will be empty after the removal
13400 # if we proceed the backup would be removed because OpBackupQuery
13401 # substitutes an empty list with the full cluster node list.
13402 iname = self.instance.name
13404 feedback_fn("Removing old exports for instance %s" % iname)
13405 exportlist = self.rpc.call_export_list(nodelist)
13406 for node in exportlist:
13407 if exportlist[node].fail_msg:
13409 if iname in exportlist[node].payload:
13410 msg = self.rpc.call_export_remove(node, iname).fail_msg
13412 self.LogWarning("Could not remove older export for instance %s"
13413 " on node %s: %s", iname, node, msg)
13415 def Exec(self, feedback_fn):
13416 """Export an instance to an image in the cluster.
13419 assert self.op.mode in constants.EXPORT_MODES
13421 instance = self.instance
13422 src_node = instance.primary_node
13424 if self.op.shutdown:
13425 # shutdown the instance, but not the disks
13426 feedback_fn("Shutting down instance %s" % instance.name)
13427 result = self.rpc.call_instance_shutdown(src_node, instance,
13428 self.op.shutdown_timeout)
13429 # TODO: Maybe ignore failures if ignore_remove_failures is set
13430 result.Raise("Could not shutdown instance %s on"
13431 " node %s" % (instance.name, src_node))
13433 # set the disks ID correctly since call_instance_start needs the
13434 # correct drbd minor to create the symlinks
13435 for disk in instance.disks:
13436 self.cfg.SetDiskID(disk, src_node)
13438 activate_disks = (instance.admin_state != constants.ADMINST_UP)
13441 # Activate the instance disks if we'exporting a stopped instance
13442 feedback_fn("Activating disks for %s" % instance.name)
13443 _StartInstanceDisks(self, instance, None)
13446 helper = masterd.instance.ExportInstanceHelper(self, feedback_fn,
13449 helper.CreateSnapshots()
13451 if (self.op.shutdown and
13452 instance.admin_state == constants.ADMINST_UP and
13453 not self.op.remove_instance):
13454 assert not activate_disks
13455 feedback_fn("Starting instance %s" % instance.name)
13456 result = self.rpc.call_instance_start(src_node,
13457 (instance, None, None), False)
13458 msg = result.fail_msg
13460 feedback_fn("Failed to start instance: %s" % msg)
13461 _ShutdownInstanceDisks(self, instance)
13462 raise errors.OpExecError("Could not start instance: %s" % msg)
13464 if self.op.mode == constants.EXPORT_MODE_LOCAL:
13465 (fin_resu, dresults) = helper.LocalExport(self.dst_node)
13466 elif self.op.mode == constants.EXPORT_MODE_REMOTE:
13467 connect_timeout = constants.RIE_CONNECT_TIMEOUT
13468 timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
13470 (key_name, _, _) = self.x509_key_name
13473 OpenSSL.crypto.dump_certificate(OpenSSL.crypto.FILETYPE_PEM,
13476 (fin_resu, dresults) = helper.RemoteExport(self.dest_disk_info,
13477 key_name, dest_ca_pem,
13482 # Check for backwards compatibility
13483 assert len(dresults) == len(instance.disks)
13484 assert compat.all(isinstance(i, bool) for i in dresults), \
13485 "Not all results are boolean: %r" % dresults
13489 feedback_fn("Deactivating disks for %s" % instance.name)
13490 _ShutdownInstanceDisks(self, instance)
13492 if not (compat.all(dresults) and fin_resu):
13495 failures.append("export finalization")
13496 if not compat.all(dresults):
13497 fdsk = utils.CommaJoin(idx for (idx, dsk) in enumerate(dresults)
13499 failures.append("disk export: disk(s) %s" % fdsk)
13501 raise errors.OpExecError("Export failed, errors in %s" %
13502 utils.CommaJoin(failures))
13504 # At this point, the export was successful, we can cleanup/finish
13506 # Remove instance if requested
13507 if self.op.remove_instance:
13508 feedback_fn("Removing instance %s" % instance.name)
13509 _RemoveInstance(self, feedback_fn, instance,
13510 self.op.ignore_remove_failures)
13512 if self.op.mode == constants.EXPORT_MODE_LOCAL:
13513 self._CleanupExports(feedback_fn)
13515 return fin_resu, dresults
13518 class LUBackupRemove(NoHooksLU):
13519 """Remove exports related to the named instance.
13524 def ExpandNames(self):
13525 self.needed_locks = {}
13526 # We need all nodes to be locked in order for RemoveExport to work, but we
13527 # don't need to lock the instance itself, as nothing will happen to it (and
13528 # we can remove exports also for a removed instance)
13529 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
13531 def Exec(self, feedback_fn):
13532 """Remove any export.
13535 instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
13536 # If the instance was not found we'll try with the name that was passed in.
13537 # This will only work if it was an FQDN, though.
13539 if not instance_name:
13541 instance_name = self.op.instance_name
13543 locked_nodes = self.owned_locks(locking.LEVEL_NODE)
13544 exportlist = self.rpc.call_export_list(locked_nodes)
13546 for node in exportlist:
13547 msg = exportlist[node].fail_msg
13549 self.LogWarning("Failed to query node %s (continuing): %s", node, msg)
13551 if instance_name in exportlist[node].payload:
13553 result = self.rpc.call_export_remove(node, instance_name)
13554 msg = result.fail_msg
13556 logging.error("Could not remove export for instance %s"
13557 " on node %s: %s", instance_name, node, msg)
13559 if fqdn_warn and not found:
13560 feedback_fn("Export not found. If trying to remove an export belonging"
13561 " to a deleted instance please use its Fully Qualified"
13565 class LUGroupAdd(LogicalUnit):
13566 """Logical unit for creating node groups.
13569 HPATH = "group-add"
13570 HTYPE = constants.HTYPE_GROUP
13573 def ExpandNames(self):
13574 # We need the new group's UUID here so that we can create and acquire the
13575 # corresponding lock. Later, in Exec(), we'll indicate to cfg.AddNodeGroup
13576 # that it should not check whether the UUID exists in the configuration.
13577 self.group_uuid = self.cfg.GenerateUniqueID(self.proc.GetECId())
13578 self.needed_locks = {}
13579 self.add_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
13581 def CheckPrereq(self):
13582 """Check prerequisites.
13584 This checks that the given group name is not an existing node group
13589 existing_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
13590 except errors.OpPrereqError:
13593 raise errors.OpPrereqError("Desired group name '%s' already exists as a"
13594 " node group (UUID: %s)" %
13595 (self.op.group_name, existing_uuid),
13596 errors.ECODE_EXISTS)
13598 if self.op.ndparams:
13599 utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
13601 if self.op.hv_state:
13602 self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state, None)
13604 self.new_hv_state = None
13606 if self.op.disk_state:
13607 self.new_disk_state = _MergeAndVerifyDiskState(self.op.disk_state, None)
13609 self.new_disk_state = None
13611 if self.op.diskparams:
13612 for templ in constants.DISK_TEMPLATES:
13613 if templ in self.op.diskparams:
13614 utils.ForceDictType(self.op.diskparams[templ],
13615 constants.DISK_DT_TYPES)
13616 self.new_diskparams = self.op.diskparams
13618 utils.VerifyDictOptions(self.new_diskparams, constants.DISK_DT_DEFAULTS)
13619 except errors.OpPrereqError, err:
13620 raise errors.OpPrereqError("While verify diskparams options: %s" % err,
13621 errors.ECODE_INVAL)
13623 self.new_diskparams = {}
13625 if self.op.ipolicy:
13626 cluster = self.cfg.GetClusterInfo()
13627 full_ipolicy = cluster.SimpleFillIPolicy(self.op.ipolicy)
13629 objects.InstancePolicy.CheckParameterSyntax(full_ipolicy)
13630 except errors.ConfigurationError, err:
13631 raise errors.OpPrereqError("Invalid instance policy: %s" % err,
13632 errors.ECODE_INVAL)
13634 def BuildHooksEnv(self):
13635 """Build hooks env.
13639 "GROUP_NAME": self.op.group_name,
13642 def BuildHooksNodes(self):
13643 """Build hooks nodes.
13646 mn = self.cfg.GetMasterNode()
13647 return ([mn], [mn])
13649 def Exec(self, feedback_fn):
13650 """Add the node group to the cluster.
13653 group_obj = objects.NodeGroup(name=self.op.group_name, members=[],
13654 uuid=self.group_uuid,
13655 alloc_policy=self.op.alloc_policy,
13656 ndparams=self.op.ndparams,
13657 diskparams=self.new_diskparams,
13658 ipolicy=self.op.ipolicy,
13659 hv_state_static=self.new_hv_state,
13660 disk_state_static=self.new_disk_state)
13662 self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False)
13663 del self.remove_locks[locking.LEVEL_NODEGROUP]
13666 class LUGroupAssignNodes(NoHooksLU):
13667 """Logical unit for assigning nodes to groups.
13672 def ExpandNames(self):
13673 # These raise errors.OpPrereqError on their own:
13674 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
13675 self.op.nodes = _GetWantedNodes(self, self.op.nodes)
13677 # We want to lock all the affected nodes and groups. We have readily
13678 # available the list of nodes, and the *destination* group. To gather the
13679 # list of "source" groups, we need to fetch node information later on.
13680 self.needed_locks = {
13681 locking.LEVEL_NODEGROUP: set([self.group_uuid]),
13682 locking.LEVEL_NODE: self.op.nodes,
13685 def DeclareLocks(self, level):
13686 if level == locking.LEVEL_NODEGROUP:
13687 assert len(self.needed_locks[locking.LEVEL_NODEGROUP]) == 1
13689 # Try to get all affected nodes' groups without having the group or node
13690 # lock yet. Needs verification later in the code flow.
13691 groups = self.cfg.GetNodeGroupsFromNodes(self.op.nodes)
13693 self.needed_locks[locking.LEVEL_NODEGROUP].update(groups)
13695 def CheckPrereq(self):
13696 """Check prerequisites.
13699 assert self.needed_locks[locking.LEVEL_NODEGROUP]
13700 assert (frozenset(self.owned_locks(locking.LEVEL_NODE)) ==
13701 frozenset(self.op.nodes))
13703 expected_locks = (set([self.group_uuid]) |
13704 self.cfg.GetNodeGroupsFromNodes(self.op.nodes))
13705 actual_locks = self.owned_locks(locking.LEVEL_NODEGROUP)
13706 if actual_locks != expected_locks:
13707 raise errors.OpExecError("Nodes changed groups since locks were acquired,"
13708 " current groups are '%s', used to be '%s'" %
13709 (utils.CommaJoin(expected_locks),
13710 utils.CommaJoin(actual_locks)))
13712 self.node_data = self.cfg.GetAllNodesInfo()
13713 self.group = self.cfg.GetNodeGroup(self.group_uuid)
13714 instance_data = self.cfg.GetAllInstancesInfo()
13716 if self.group is None:
13717 raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
13718 (self.op.group_name, self.group_uuid))
13720 (new_splits, previous_splits) = \
13721 self.CheckAssignmentForSplitInstances([(node, self.group_uuid)
13722 for node in self.op.nodes],
13723 self.node_data, instance_data)
13726 fmt_new_splits = utils.CommaJoin(utils.NiceSort(new_splits))
13728 if not self.op.force:
13729 raise errors.OpExecError("The following instances get split by this"
13730 " change and --force was not given: %s" %
13733 self.LogWarning("This operation will split the following instances: %s",
13736 if previous_splits:
13737 self.LogWarning("In addition, these already-split instances continue"
13738 " to be split across groups: %s",
13739 utils.CommaJoin(utils.NiceSort(previous_splits)))
13741 def Exec(self, feedback_fn):
13742 """Assign nodes to a new group.
13745 mods = [(node_name, self.group_uuid) for node_name in self.op.nodes]
13747 self.cfg.AssignGroupNodes(mods)
13750 def CheckAssignmentForSplitInstances(changes, node_data, instance_data):
13751 """Check for split instances after a node assignment.
13753 This method considers a series of node assignments as an atomic operation,
13754 and returns information about split instances after applying the set of
13757 In particular, it returns information about newly split instances, and
13758 instances that were already split, and remain so after the change.
13760 Only instances whose disk template is listed in constants.DTS_INT_MIRROR are
13763 @type changes: list of (node_name, new_group_uuid) pairs.
13764 @param changes: list of node assignments to consider.
13765 @param node_data: a dict with data for all nodes
13766 @param instance_data: a dict with all instances to consider
13767 @rtype: a two-tuple
13768 @return: a list of instances that were previously okay and result split as a
13769 consequence of this change, and a list of instances that were previously
13770 split and this change does not fix.
13773 changed_nodes = dict((node, group) for node, group in changes
13774 if node_data[node].group != group)
13776 all_split_instances = set()
13777 previously_split_instances = set()
13779 def InstanceNodes(instance):
13780 return [instance.primary_node] + list(instance.secondary_nodes)
13782 for inst in instance_data.values():
13783 if inst.disk_template not in constants.DTS_INT_MIRROR:
13786 instance_nodes = InstanceNodes(inst)
13788 if len(set(node_data[node].group for node in instance_nodes)) > 1:
13789 previously_split_instances.add(inst.name)
13791 if len(set(changed_nodes.get(node, node_data[node].group)
13792 for node in instance_nodes)) > 1:
13793 all_split_instances.add(inst.name)
13795 return (list(all_split_instances - previously_split_instances),
13796 list(previously_split_instances & all_split_instances))
13799 class _GroupQuery(_QueryBase):
13800 FIELDS = query.GROUP_FIELDS
13802 def ExpandNames(self, lu):
13803 lu.needed_locks = {}
13805 self._all_groups = lu.cfg.GetAllNodeGroupsInfo()
13806 self._cluster = lu.cfg.GetClusterInfo()
13807 name_to_uuid = dict((g.name, g.uuid) for g in self._all_groups.values())
13810 self.wanted = [name_to_uuid[name]
13811 for name in utils.NiceSort(name_to_uuid.keys())]
13813 # Accept names to be either names or UUIDs.
13816 all_uuid = frozenset(self._all_groups.keys())
13818 for name in self.names:
13819 if name in all_uuid:
13820 self.wanted.append(name)
13821 elif name in name_to_uuid:
13822 self.wanted.append(name_to_uuid[name])
13824 missing.append(name)
13827 raise errors.OpPrereqError("Some groups do not exist: %s" %
13828 utils.CommaJoin(missing),
13829 errors.ECODE_NOENT)
13831 def DeclareLocks(self, lu, level):
13834 def _GetQueryData(self, lu):
13835 """Computes the list of node groups and their attributes.
13838 do_nodes = query.GQ_NODE in self.requested_data
13839 do_instances = query.GQ_INST in self.requested_data
13841 group_to_nodes = None
13842 group_to_instances = None
13844 # For GQ_NODE, we need to map group->[nodes], and group->[instances] for
13845 # GQ_INST. The former is attainable with just GetAllNodesInfo(), but for the
13846 # latter GetAllInstancesInfo() is not enough, for we have to go through
13847 # instance->node. Hence, we will need to process nodes even if we only need
13848 # instance information.
13849 if do_nodes or do_instances:
13850 all_nodes = lu.cfg.GetAllNodesInfo()
13851 group_to_nodes = dict((uuid, []) for uuid in self.wanted)
13854 for node in all_nodes.values():
13855 if node.group in group_to_nodes:
13856 group_to_nodes[node.group].append(node.name)
13857 node_to_group[node.name] = node.group
13860 all_instances = lu.cfg.GetAllInstancesInfo()
13861 group_to_instances = dict((uuid, []) for uuid in self.wanted)
13863 for instance in all_instances.values():
13864 node = instance.primary_node
13865 if node in node_to_group:
13866 group_to_instances[node_to_group[node]].append(instance.name)
13869 # Do not pass on node information if it was not requested.
13870 group_to_nodes = None
13872 return query.GroupQueryData(self._cluster,
13873 [self._all_groups[uuid]
13874 for uuid in self.wanted],
13875 group_to_nodes, group_to_instances,
13876 query.GQ_DISKPARAMS in self.requested_data)
13879 class LUGroupQuery(NoHooksLU):
13880 """Logical unit for querying node groups.
13885 def CheckArguments(self):
13886 self.gq = _GroupQuery(qlang.MakeSimpleFilter("name", self.op.names),
13887 self.op.output_fields, False)
13889 def ExpandNames(self):
13890 self.gq.ExpandNames(self)
13892 def DeclareLocks(self, level):
13893 self.gq.DeclareLocks(self, level)
13895 def Exec(self, feedback_fn):
13896 return self.gq.OldStyleQuery(self)
13899 class LUGroupSetParams(LogicalUnit):
13900 """Modifies the parameters of a node group.
13903 HPATH = "group-modify"
13904 HTYPE = constants.HTYPE_GROUP
13907 def CheckArguments(self):
13910 self.op.diskparams,
13911 self.op.alloc_policy,
13913 self.op.disk_state,
13917 if all_changes.count(None) == len(all_changes):
13918 raise errors.OpPrereqError("Please pass at least one modification",
13919 errors.ECODE_INVAL)
13921 def ExpandNames(self):
13922 # This raises errors.OpPrereqError on its own:
13923 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
13925 self.needed_locks = {
13926 locking.LEVEL_INSTANCE: [],
13927 locking.LEVEL_NODEGROUP: [self.group_uuid],
13930 self.share_locks[locking.LEVEL_INSTANCE] = 1
13932 def DeclareLocks(self, level):
13933 if level == locking.LEVEL_INSTANCE:
13934 assert not self.needed_locks[locking.LEVEL_INSTANCE]
13936 # Lock instances optimistically, needs verification once group lock has
13938 self.needed_locks[locking.LEVEL_INSTANCE] = \
13939 self.cfg.GetNodeGroupInstances(self.group_uuid)
13942 def _UpdateAndVerifyDiskParams(old, new):
13943 """Updates and verifies disk parameters.
13946 new_params = _GetUpdatedParams(old, new)
13947 utils.ForceDictType(new_params, constants.DISK_DT_TYPES)
13950 def CheckPrereq(self):
13951 """Check prerequisites.
13954 owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
13956 # Check if locked instances are still correct
13957 _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
13959 self.group = self.cfg.GetNodeGroup(self.group_uuid)
13960 cluster = self.cfg.GetClusterInfo()
13962 if self.group is None:
13963 raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
13964 (self.op.group_name, self.group_uuid))
13966 if self.op.ndparams:
13967 new_ndparams = _GetUpdatedParams(self.group.ndparams, self.op.ndparams)
13968 utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
13969 self.new_ndparams = new_ndparams
13971 if self.op.diskparams:
13972 diskparams = self.group.diskparams
13973 uavdp = self._UpdateAndVerifyDiskParams
13974 # For each disktemplate subdict update and verify the values
13975 new_diskparams = dict((dt,
13976 uavdp(diskparams.get(dt, {}),
13977 self.op.diskparams[dt]))
13978 for dt in constants.DISK_TEMPLATES
13979 if dt in self.op.diskparams)
13980 # As we've all subdicts of diskparams ready, lets merge the actual
13981 # dict with all updated subdicts
13982 self.new_diskparams = objects.FillDict(diskparams, new_diskparams)
13984 utils.VerifyDictOptions(self.new_diskparams, constants.DISK_DT_DEFAULTS)
13985 except errors.OpPrereqError, err:
13986 raise errors.OpPrereqError("While verify diskparams options: %s" % err,
13987 errors.ECODE_INVAL)
13989 if self.op.hv_state:
13990 self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state,
13991 self.group.hv_state_static)
13993 if self.op.disk_state:
13994 self.new_disk_state = \
13995 _MergeAndVerifyDiskState(self.op.disk_state,
13996 self.group.disk_state_static)
13998 if self.op.ipolicy:
13999 self.new_ipolicy = _GetUpdatedIPolicy(self.group.ipolicy,
14003 new_ipolicy = cluster.SimpleFillIPolicy(self.new_ipolicy)
14004 inst_filter = lambda inst: inst.name in owned_instances
14005 instances = self.cfg.GetInstancesInfoByFilter(inst_filter).values()
14007 _ComputeNewInstanceViolations(_CalculateGroupIPolicy(cluster,
14009 new_ipolicy, instances)
14012 self.LogWarning("After the ipolicy change the following instances"
14013 " violate them: %s",
14014 utils.CommaJoin(violations))
14016 def BuildHooksEnv(self):
14017 """Build hooks env.
14021 "GROUP_NAME": self.op.group_name,
14022 "NEW_ALLOC_POLICY": self.op.alloc_policy,
14025 def BuildHooksNodes(self):
14026 """Build hooks nodes.
14029 mn = self.cfg.GetMasterNode()
14030 return ([mn], [mn])
14032 def Exec(self, feedback_fn):
14033 """Modifies the node group.
14038 if self.op.ndparams:
14039 self.group.ndparams = self.new_ndparams
14040 result.append(("ndparams", str(self.group.ndparams)))
14042 if self.op.diskparams:
14043 self.group.diskparams = self.new_diskparams
14044 result.append(("diskparams", str(self.group.diskparams)))
14046 if self.op.alloc_policy:
14047 self.group.alloc_policy = self.op.alloc_policy
14049 if self.op.hv_state:
14050 self.group.hv_state_static = self.new_hv_state
14052 if self.op.disk_state:
14053 self.group.disk_state_static = self.new_disk_state
14055 if self.op.ipolicy:
14056 self.group.ipolicy = self.new_ipolicy
14058 self.cfg.Update(self.group, feedback_fn)
14062 class LUGroupRemove(LogicalUnit):
14063 HPATH = "group-remove"
14064 HTYPE = constants.HTYPE_GROUP
14067 def ExpandNames(self):
14068 # This will raises errors.OpPrereqError on its own:
14069 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
14070 self.needed_locks = {
14071 locking.LEVEL_NODEGROUP: [self.group_uuid],
14074 def CheckPrereq(self):
14075 """Check prerequisites.
14077 This checks that the given group name exists as a node group, that is
14078 empty (i.e., contains no nodes), and that is not the last group of the
14082 # Verify that the group is empty.
14083 group_nodes = [node.name
14084 for node in self.cfg.GetAllNodesInfo().values()
14085 if node.group == self.group_uuid]
14088 raise errors.OpPrereqError("Group '%s' not empty, has the following"
14090 (self.op.group_name,
14091 utils.CommaJoin(utils.NiceSort(group_nodes))),
14092 errors.ECODE_STATE)
14094 # Verify the cluster would not be left group-less.
14095 if len(self.cfg.GetNodeGroupList()) == 1:
14096 raise errors.OpPrereqError("Group '%s' is the only group,"
14097 " cannot be removed" %
14098 self.op.group_name,
14099 errors.ECODE_STATE)
14101 def BuildHooksEnv(self):
14102 """Build hooks env.
14106 "GROUP_NAME": self.op.group_name,
14109 def BuildHooksNodes(self):
14110 """Build hooks nodes.
14113 mn = self.cfg.GetMasterNode()
14114 return ([mn], [mn])
14116 def Exec(self, feedback_fn):
14117 """Remove the node group.
14121 self.cfg.RemoveNodeGroup(self.group_uuid)
14122 except errors.ConfigurationError:
14123 raise errors.OpExecError("Group '%s' with UUID %s disappeared" %
14124 (self.op.group_name, self.group_uuid))
14126 self.remove_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
14129 class LUGroupRename(LogicalUnit):
14130 HPATH = "group-rename"
14131 HTYPE = constants.HTYPE_GROUP
14134 def ExpandNames(self):
14135 # This raises errors.OpPrereqError on its own:
14136 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
14138 self.needed_locks = {
14139 locking.LEVEL_NODEGROUP: [self.group_uuid],
14142 def CheckPrereq(self):
14143 """Check prerequisites.
14145 Ensures requested new name is not yet used.
14149 new_name_uuid = self.cfg.LookupNodeGroup(self.op.new_name)
14150 except errors.OpPrereqError:
14153 raise errors.OpPrereqError("Desired new name '%s' clashes with existing"
14154 " node group (UUID: %s)" %
14155 (self.op.new_name, new_name_uuid),
14156 errors.ECODE_EXISTS)
14158 def BuildHooksEnv(self):
14159 """Build hooks env.
14163 "OLD_NAME": self.op.group_name,
14164 "NEW_NAME": self.op.new_name,
14167 def BuildHooksNodes(self):
14168 """Build hooks nodes.
14171 mn = self.cfg.GetMasterNode()
14173 all_nodes = self.cfg.GetAllNodesInfo()
14174 all_nodes.pop(mn, None)
14177 run_nodes.extend(node.name for node in all_nodes.values()
14178 if node.group == self.group_uuid)
14180 return (run_nodes, run_nodes)
14182 def Exec(self, feedback_fn):
14183 """Rename the node group.
14186 group = self.cfg.GetNodeGroup(self.group_uuid)
14189 raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
14190 (self.op.group_name, self.group_uuid))
14192 group.name = self.op.new_name
14193 self.cfg.Update(group, feedback_fn)
14195 return self.op.new_name
14198 class LUGroupEvacuate(LogicalUnit):
14199 HPATH = "group-evacuate"
14200 HTYPE = constants.HTYPE_GROUP
14203 def ExpandNames(self):
14204 # This raises errors.OpPrereqError on its own:
14205 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
14207 if self.op.target_groups:
14208 self.req_target_uuids = map(self.cfg.LookupNodeGroup,
14209 self.op.target_groups)
14211 self.req_target_uuids = []
14213 if self.group_uuid in self.req_target_uuids:
14214 raise errors.OpPrereqError("Group to be evacuated (%s) can not be used"
14215 " as a target group (targets are %s)" %
14217 utils.CommaJoin(self.req_target_uuids)),
14218 errors.ECODE_INVAL)
14220 self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
14222 self.share_locks = _ShareAll()
14223 self.needed_locks = {
14224 locking.LEVEL_INSTANCE: [],
14225 locking.LEVEL_NODEGROUP: [],
14226 locking.LEVEL_NODE: [],
14229 def DeclareLocks(self, level):
14230 if level == locking.LEVEL_INSTANCE:
14231 assert not self.needed_locks[locking.LEVEL_INSTANCE]
14233 # Lock instances optimistically, needs verification once node and group
14234 # locks have been acquired
14235 self.needed_locks[locking.LEVEL_INSTANCE] = \
14236 self.cfg.GetNodeGroupInstances(self.group_uuid)
14238 elif level == locking.LEVEL_NODEGROUP:
14239 assert not self.needed_locks[locking.LEVEL_NODEGROUP]
14241 if self.req_target_uuids:
14242 lock_groups = set([self.group_uuid] + self.req_target_uuids)
14244 # Lock all groups used by instances optimistically; this requires going
14245 # via the node before it's locked, requiring verification later on
14246 lock_groups.update(group_uuid
14247 for instance_name in
14248 self.owned_locks(locking.LEVEL_INSTANCE)
14250 self.cfg.GetInstanceNodeGroups(instance_name))
14252 # No target groups, need to lock all of them
14253 lock_groups = locking.ALL_SET
14255 self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
14257 elif level == locking.LEVEL_NODE:
14258 # This will only lock the nodes in the group to be evacuated which
14259 # contain actual instances
14260 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
14261 self._LockInstancesNodes()
14263 # Lock all nodes in group to be evacuated and target groups
14264 owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
14265 assert self.group_uuid in owned_groups
14266 member_nodes = [node_name
14267 for group in owned_groups
14268 for node_name in self.cfg.GetNodeGroup(group).members]
14269 self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
14271 def CheckPrereq(self):
14272 owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
14273 owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
14274 owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
14276 assert owned_groups.issuperset(self.req_target_uuids)
14277 assert self.group_uuid in owned_groups
14279 # Check if locked instances are still correct
14280 _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
14282 # Get instance information
14283 self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
14285 # Check if node groups for locked instances are still correct
14286 _CheckInstancesNodeGroups(self.cfg, self.instances,
14287 owned_groups, owned_nodes, self.group_uuid)
14289 if self.req_target_uuids:
14290 # User requested specific target groups
14291 self.target_uuids = self.req_target_uuids
14293 # All groups except the one to be evacuated are potential targets
14294 self.target_uuids = [group_uuid for group_uuid in owned_groups
14295 if group_uuid != self.group_uuid]
14297 if not self.target_uuids:
14298 raise errors.OpPrereqError("There are no possible target groups",
14299 errors.ECODE_INVAL)
14301 def BuildHooksEnv(self):
14302 """Build hooks env.
14306 "GROUP_NAME": self.op.group_name,
14307 "TARGET_GROUPS": " ".join(self.target_uuids),
14310 def BuildHooksNodes(self):
14311 """Build hooks nodes.
14314 mn = self.cfg.GetMasterNode()
14316 assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
14318 run_nodes = [mn] + self.cfg.GetNodeGroup(self.group_uuid).members
14320 return (run_nodes, run_nodes)
14322 def Exec(self, feedback_fn):
14323 instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
14325 assert self.group_uuid not in self.target_uuids
14327 ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_CHG_GROUP,
14328 instances=instances, target_groups=self.target_uuids)
14330 ial.Run(self.op.iallocator)
14332 if not ial.success:
14333 raise errors.OpPrereqError("Can't compute group evacuation using"
14334 " iallocator '%s': %s" %
14335 (self.op.iallocator, ial.info),
14336 errors.ECODE_NORES)
14338 jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, False)
14340 self.LogInfo("Iallocator returned %s job(s) for evacuating node group %s",
14341 len(jobs), self.op.group_name)
14343 return ResultWithJobs(jobs)
14346 class TagsLU(NoHooksLU): # pylint: disable=W0223
14347 """Generic tags LU.
14349 This is an abstract class which is the parent of all the other tags LUs.
14352 def ExpandNames(self):
14353 self.group_uuid = None
14354 self.needed_locks = {}
14356 if self.op.kind == constants.TAG_NODE:
14357 self.op.name = _ExpandNodeName(self.cfg, self.op.name)
14358 lock_level = locking.LEVEL_NODE
14359 lock_name = self.op.name
14360 elif self.op.kind == constants.TAG_INSTANCE:
14361 self.op.name = _ExpandInstanceName(self.cfg, self.op.name)
14362 lock_level = locking.LEVEL_INSTANCE
14363 lock_name = self.op.name
14364 elif self.op.kind == constants.TAG_NODEGROUP:
14365 self.group_uuid = self.cfg.LookupNodeGroup(self.op.name)
14366 lock_level = locking.LEVEL_NODEGROUP
14367 lock_name = self.group_uuid
14372 if lock_level and getattr(self.op, "use_locking", True):
14373 self.needed_locks[lock_level] = lock_name
14375 # FIXME: Acquire BGL for cluster tag operations (as of this writing it's
14376 # not possible to acquire the BGL based on opcode parameters)
14378 def CheckPrereq(self):
14379 """Check prerequisites.
14382 if self.op.kind == constants.TAG_CLUSTER:
14383 self.target = self.cfg.GetClusterInfo()
14384 elif self.op.kind == constants.TAG_NODE:
14385 self.target = self.cfg.GetNodeInfo(self.op.name)
14386 elif self.op.kind == constants.TAG_INSTANCE:
14387 self.target = self.cfg.GetInstanceInfo(self.op.name)
14388 elif self.op.kind == constants.TAG_NODEGROUP:
14389 self.target = self.cfg.GetNodeGroup(self.group_uuid)
14391 raise errors.OpPrereqError("Wrong tag type requested (%s)" %
14392 str(self.op.kind), errors.ECODE_INVAL)
14395 class LUTagsGet(TagsLU):
14396 """Returns the tags of a given object.
14401 def ExpandNames(self):
14402 TagsLU.ExpandNames(self)
14404 # Share locks as this is only a read operation
14405 self.share_locks = _ShareAll()
14407 def Exec(self, feedback_fn):
14408 """Returns the tag list.
14411 return list(self.target.GetTags())
14414 class LUTagsSearch(NoHooksLU):
14415 """Searches the tags for a given pattern.
14420 def ExpandNames(self):
14421 self.needed_locks = {}
14423 def CheckPrereq(self):
14424 """Check prerequisites.
14426 This checks the pattern passed for validity by compiling it.
14430 self.re = re.compile(self.op.pattern)
14431 except re.error, err:
14432 raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
14433 (self.op.pattern, err), errors.ECODE_INVAL)
14435 def Exec(self, feedback_fn):
14436 """Returns the tag list.
14440 tgts = [("/cluster", cfg.GetClusterInfo())]
14441 ilist = cfg.GetAllInstancesInfo().values()
14442 tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
14443 nlist = cfg.GetAllNodesInfo().values()
14444 tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
14445 tgts.extend(("/nodegroup/%s" % n.name, n)
14446 for n in cfg.GetAllNodeGroupsInfo().values())
14448 for path, target in tgts:
14449 for tag in target.GetTags():
14450 if self.re.search(tag):
14451 results.append((path, tag))
14455 class LUTagsSet(TagsLU):
14456 """Sets a tag on a given object.
14461 def CheckPrereq(self):
14462 """Check prerequisites.
14464 This checks the type and length of the tag name and value.
14467 TagsLU.CheckPrereq(self)
14468 for tag in self.op.tags:
14469 objects.TaggableObject.ValidateTag(tag)
14471 def Exec(self, feedback_fn):
14476 for tag in self.op.tags:
14477 self.target.AddTag(tag)
14478 except errors.TagError, err:
14479 raise errors.OpExecError("Error while setting tag: %s" % str(err))
14480 self.cfg.Update(self.target, feedback_fn)
14483 class LUTagsDel(TagsLU):
14484 """Delete a list of tags from a given object.
14489 def CheckPrereq(self):
14490 """Check prerequisites.
14492 This checks that we have the given tag.
14495 TagsLU.CheckPrereq(self)
14496 for tag in self.op.tags:
14497 objects.TaggableObject.ValidateTag(tag)
14498 del_tags = frozenset(self.op.tags)
14499 cur_tags = self.target.GetTags()
14501 diff_tags = del_tags - cur_tags
14503 diff_names = ("'%s'" % i for i in sorted(diff_tags))
14504 raise errors.OpPrereqError("Tag(s) %s not found" %
14505 (utils.CommaJoin(diff_names), ),
14506 errors.ECODE_NOENT)
14508 def Exec(self, feedback_fn):
14509 """Remove the tag from the object.
14512 for tag in self.op.tags:
14513 self.target.RemoveTag(tag)
14514 self.cfg.Update(self.target, feedback_fn)
14517 class LUTestDelay(NoHooksLU):
14518 """Sleep for a specified amount of time.
14520 This LU sleeps on the master and/or nodes for a specified amount of
14526 def ExpandNames(self):
14527 """Expand names and set required locks.
14529 This expands the node list, if any.
14532 self.needed_locks = {}
14533 if self.op.on_nodes:
14534 # _GetWantedNodes can be used here, but is not always appropriate to use
14535 # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
14536 # more information.
14537 self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
14538 self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
14540 def _TestDelay(self):
14541 """Do the actual sleep.
14544 if self.op.on_master:
14545 if not utils.TestDelay(self.op.duration):
14546 raise errors.OpExecError("Error during master delay test")
14547 if self.op.on_nodes:
14548 result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
14549 for node, node_result in result.items():
14550 node_result.Raise("Failure during rpc call to node %s" % node)
14552 def Exec(self, feedback_fn):
14553 """Execute the test delay opcode, with the wanted repetitions.
14556 if self.op.repeat == 0:
14559 top_value = self.op.repeat - 1
14560 for i in range(self.op.repeat):
14561 self.LogInfo("Test delay iteration %d/%d" % (i, top_value))
14565 class LUTestJqueue(NoHooksLU):
14566 """Utility LU to test some aspects of the job queue.
14571 # Must be lower than default timeout for WaitForJobChange to see whether it
14572 # notices changed jobs
14573 _CLIENT_CONNECT_TIMEOUT = 20.0
14574 _CLIENT_CONFIRM_TIMEOUT = 60.0
14577 def _NotifyUsingSocket(cls, cb, errcls):
14578 """Opens a Unix socket and waits for another program to connect.
14581 @param cb: Callback to send socket name to client
14582 @type errcls: class
14583 @param errcls: Exception class to use for errors
14586 # Using a temporary directory as there's no easy way to create temporary
14587 # sockets without writing a custom loop around tempfile.mktemp and
14589 tmpdir = tempfile.mkdtemp()
14591 tmpsock = utils.PathJoin(tmpdir, "sock")
14593 logging.debug("Creating temporary socket at %s", tmpsock)
14594 sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
14599 # Send details to client
14602 # Wait for client to connect before continuing
14603 sock.settimeout(cls._CLIENT_CONNECT_TIMEOUT)
14605 (conn, _) = sock.accept()
14606 except socket.error, err:
14607 raise errcls("Client didn't connect in time (%s)" % err)
14611 # Remove as soon as client is connected
14612 shutil.rmtree(tmpdir)
14614 # Wait for client to close
14617 # pylint: disable=E1101
14618 # Instance of '_socketobject' has no ... member
14619 conn.settimeout(cls._CLIENT_CONFIRM_TIMEOUT)
14621 except socket.error, err:
14622 raise errcls("Client failed to confirm notification (%s)" % err)
14626 def _SendNotification(self, test, arg, sockname):
14627 """Sends a notification to the client.
14630 @param test: Test name
14631 @param arg: Test argument (depends on test)
14632 @type sockname: string
14633 @param sockname: Socket path
14636 self.Log(constants.ELOG_JQUEUE_TEST, (sockname, test, arg))
14638 def _Notify(self, prereq, test, arg):
14639 """Notifies the client of a test.
14642 @param prereq: Whether this is a prereq-phase test
14644 @param test: Test name
14645 @param arg: Test argument (depends on test)
14649 errcls = errors.OpPrereqError
14651 errcls = errors.OpExecError
14653 return self._NotifyUsingSocket(compat.partial(self._SendNotification,
14657 def CheckArguments(self):
14658 self.checkargs_calls = getattr(self, "checkargs_calls", 0) + 1
14659 self.expandnames_calls = 0
14661 def ExpandNames(self):
14662 checkargs_calls = getattr(self, "checkargs_calls", 0)
14663 if checkargs_calls < 1:
14664 raise errors.ProgrammerError("CheckArguments was not called")
14666 self.expandnames_calls += 1
14668 if self.op.notify_waitlock:
14669 self._Notify(True, constants.JQT_EXPANDNAMES, None)
14671 self.LogInfo("Expanding names")
14673 # Get lock on master node (just to get a lock, not for a particular reason)
14674 self.needed_locks = {
14675 locking.LEVEL_NODE: self.cfg.GetMasterNode(),
14678 def Exec(self, feedback_fn):
14679 if self.expandnames_calls < 1:
14680 raise errors.ProgrammerError("ExpandNames was not called")
14682 if self.op.notify_exec:
14683 self._Notify(False, constants.JQT_EXEC, None)
14685 self.LogInfo("Executing")
14687 if self.op.log_messages:
14688 self._Notify(False, constants.JQT_STARTMSG, len(self.op.log_messages))
14689 for idx, msg in enumerate(self.op.log_messages):
14690 self.LogInfo("Sending log message %s", idx + 1)
14691 feedback_fn(constants.JQT_MSGPREFIX + msg)
14692 # Report how many test messages have been sent
14693 self._Notify(False, constants.JQT_LOGMSG, idx + 1)
14696 raise errors.OpExecError("Opcode failure was requested")
14701 class IAllocator(object):
14702 """IAllocator framework.
14704 An IAllocator instance has three sets of attributes:
14705 - cfg that is needed to query the cluster
14706 - input data (all members of the _KEYS class attribute are required)
14707 - four buffer attributes (in|out_data|text), that represent the
14708 input (to the external script) in text and data structure format,
14709 and the output from it, again in two formats
14710 - the result variables from the script (success, info, nodes) for
14714 # pylint: disable=R0902
14715 # lots of instance attributes
14717 def __init__(self, cfg, rpc_runner, mode, **kwargs):
14719 self.rpc = rpc_runner
14720 # init buffer variables
14721 self.in_text = self.out_text = self.in_data = self.out_data = None
14722 # init all input fields so that pylint is happy
14724 self.memory = self.disks = self.disk_template = self.spindle_use = None
14725 self.os = self.tags = self.nics = self.vcpus = None
14726 self.hypervisor = None
14727 self.relocate_from = None
14729 self.instances = None
14730 self.evac_mode = None
14731 self.target_groups = []
14733 self.required_nodes = None
14734 # init result fields
14735 self.success = self.info = self.result = None
14738 (fn, keydata, self._result_check) = self._MODE_DATA[self.mode]
14740 raise errors.ProgrammerError("Unknown mode '%s' passed to the"
14741 " IAllocator" % self.mode)
14743 keyset = [n for (n, _) in keydata]
14746 if key not in keyset:
14747 raise errors.ProgrammerError("Invalid input parameter '%s' to"
14748 " IAllocator" % key)
14749 setattr(self, key, kwargs[key])
14752 if key not in kwargs:
14753 raise errors.ProgrammerError("Missing input parameter '%s' to"
14754 " IAllocator" % key)
14755 self._BuildInputData(compat.partial(fn, self), keydata)
14757 def _ComputeClusterData(self):
14758 """Compute the generic allocator input data.
14760 This is the data that is independent of the actual operation.
14764 cluster_info = cfg.GetClusterInfo()
14767 "version": constants.IALLOCATOR_VERSION,
14768 "cluster_name": cfg.GetClusterName(),
14769 "cluster_tags": list(cluster_info.GetTags()),
14770 "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
14771 "ipolicy": cluster_info.ipolicy,
14773 ninfo = cfg.GetAllNodesInfo()
14774 iinfo = cfg.GetAllInstancesInfo().values()
14775 i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
14778 node_list = [n.name for n in ninfo.values() if n.vm_capable]
14780 if self.mode == constants.IALLOCATOR_MODE_ALLOC:
14781 hypervisor_name = self.hypervisor
14782 elif self.mode == constants.IALLOCATOR_MODE_RELOC:
14783 hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
14785 hypervisor_name = cluster_info.primary_hypervisor
14787 node_data = self.rpc.call_node_info(node_list, [cfg.GetVGName()],
14790 self.rpc.call_all_instances_info(node_list,
14791 cluster_info.enabled_hypervisors)
14793 data["nodegroups"] = self._ComputeNodeGroupData(cfg)
14795 config_ndata = self._ComputeBasicNodeData(cfg, ninfo)
14796 data["nodes"] = self._ComputeDynamicNodeData(ninfo, node_data, node_iinfo,
14797 i_list, config_ndata)
14798 assert len(data["nodes"]) == len(ninfo), \
14799 "Incomplete node data computed"
14801 data["instances"] = self._ComputeInstanceData(cluster_info, i_list)
14803 self.in_data = data
14806 def _ComputeNodeGroupData(cfg):
14807 """Compute node groups data.
14810 cluster = cfg.GetClusterInfo()
14811 ng = dict((guuid, {
14812 "name": gdata.name,
14813 "alloc_policy": gdata.alloc_policy,
14814 "ipolicy": _CalculateGroupIPolicy(cluster, gdata),
14816 for guuid, gdata in cfg.GetAllNodeGroupsInfo().items())
14821 def _ComputeBasicNodeData(cfg, node_cfg):
14822 """Compute global node data.
14825 @returns: a dict of name: (node dict, node config)
14828 # fill in static (config-based) values
14829 node_results = dict((ninfo.name, {
14830 "tags": list(ninfo.GetTags()),
14831 "primary_ip": ninfo.primary_ip,
14832 "secondary_ip": ninfo.secondary_ip,
14833 "offline": ninfo.offline,
14834 "drained": ninfo.drained,
14835 "master_candidate": ninfo.master_candidate,
14836 "group": ninfo.group,
14837 "master_capable": ninfo.master_capable,
14838 "vm_capable": ninfo.vm_capable,
14839 "ndparams": cfg.GetNdParams(ninfo),
14841 for ninfo in node_cfg.values())
14843 return node_results
14846 def _ComputeDynamicNodeData(node_cfg, node_data, node_iinfo, i_list,
14848 """Compute global node data.
14850 @param node_results: the basic node structures as filled from the config
14853 #TODO(dynmem): compute the right data on MAX and MIN memory
14854 # make a copy of the current dict
14855 node_results = dict(node_results)
14856 for nname, nresult in node_data.items():
14857 assert nname in node_results, "Missing basic data for node %s" % nname
14858 ninfo = node_cfg[nname]
14860 if not (ninfo.offline or ninfo.drained):
14861 nresult.Raise("Can't get data for node %s" % nname)
14862 node_iinfo[nname].Raise("Can't get node instance info from node %s" %
14864 remote_info = _MakeLegacyNodeInfo(nresult.payload)
14866 for attr in ["memory_total", "memory_free", "memory_dom0",
14867 "vg_size", "vg_free", "cpu_total"]:
14868 if attr not in remote_info:
14869 raise errors.OpExecError("Node '%s' didn't return attribute"
14870 " '%s'" % (nname, attr))
14871 if not isinstance(remote_info[attr], int):
14872 raise errors.OpExecError("Node '%s' returned invalid value"
14874 (nname, attr, remote_info[attr]))
14875 # compute memory used by primary instances
14876 i_p_mem = i_p_up_mem = 0
14877 for iinfo, beinfo in i_list:
14878 if iinfo.primary_node == nname:
14879 i_p_mem += beinfo[constants.BE_MAXMEM]
14880 if iinfo.name not in node_iinfo[nname].payload:
14883 i_used_mem = int(node_iinfo[nname].payload[iinfo.name]["memory"])
14884 i_mem_diff = beinfo[constants.BE_MAXMEM] - i_used_mem
14885 remote_info["memory_free"] -= max(0, i_mem_diff)
14887 if iinfo.admin_state == constants.ADMINST_UP:
14888 i_p_up_mem += beinfo[constants.BE_MAXMEM]
14890 # compute memory used by instances
14892 "total_memory": remote_info["memory_total"],
14893 "reserved_memory": remote_info["memory_dom0"],
14894 "free_memory": remote_info["memory_free"],
14895 "total_disk": remote_info["vg_size"],
14896 "free_disk": remote_info["vg_free"],
14897 "total_cpus": remote_info["cpu_total"],
14898 "i_pri_memory": i_p_mem,
14899 "i_pri_up_memory": i_p_up_mem,
14901 pnr_dyn.update(node_results[nname])
14902 node_results[nname] = pnr_dyn
14904 return node_results
14907 def _ComputeInstanceData(cluster_info, i_list):
14908 """Compute global instance data.
14912 for iinfo, beinfo in i_list:
14914 for nic in iinfo.nics:
14915 filled_params = cluster_info.SimpleFillNIC(nic.nicparams)
14919 "mode": filled_params[constants.NIC_MODE],
14920 "link": filled_params[constants.NIC_LINK],
14922 if filled_params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
14923 nic_dict["bridge"] = filled_params[constants.NIC_LINK]
14924 nic_data.append(nic_dict)
14926 "tags": list(iinfo.GetTags()),
14927 "admin_state": iinfo.admin_state,
14928 "vcpus": beinfo[constants.BE_VCPUS],
14929 "memory": beinfo[constants.BE_MAXMEM],
14930 "spindle_use": beinfo[constants.BE_SPINDLE_USE],
14932 "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
14934 "disks": [{constants.IDISK_SIZE: dsk.size,
14935 constants.IDISK_MODE: dsk.mode}
14936 for dsk in iinfo.disks],
14937 "disk_template": iinfo.disk_template,
14938 "hypervisor": iinfo.hypervisor,
14940 pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
14942 instance_data[iinfo.name] = pir
14944 return instance_data
14946 def _AddNewInstance(self):
14947 """Add new instance data to allocator structure.
14949 This in combination with _AllocatorGetClusterData will create the
14950 correct structure needed as input for the allocator.
14952 The checks for the completeness of the opcode must have already been
14956 disk_space = _ComputeDiskSize(self.disk_template, self.disks)
14958 if self.disk_template in constants.DTS_INT_MIRROR:
14959 self.required_nodes = 2
14961 self.required_nodes = 1
14965 "disk_template": self.disk_template,
14968 "vcpus": self.vcpus,
14969 "memory": self.memory,
14970 "spindle_use": self.spindle_use,
14971 "disks": self.disks,
14972 "disk_space_total": disk_space,
14974 "required_nodes": self.required_nodes,
14975 "hypervisor": self.hypervisor,
14980 def _AddRelocateInstance(self):
14981 """Add relocate instance data to allocator structure.
14983 This in combination with _IAllocatorGetClusterData will create the
14984 correct structure needed as input for the allocator.
14986 The checks for the completeness of the opcode must have already been
14990 instance = self.cfg.GetInstanceInfo(self.name)
14991 if instance is None:
14992 raise errors.ProgrammerError("Unknown instance '%s' passed to"
14993 " IAllocator" % self.name)
14995 if instance.disk_template not in constants.DTS_MIRRORED:
14996 raise errors.OpPrereqError("Can't relocate non-mirrored instances",
14997 errors.ECODE_INVAL)
14999 if instance.disk_template in constants.DTS_INT_MIRROR and \
15000 len(instance.secondary_nodes) != 1:
15001 raise errors.OpPrereqError("Instance has not exactly one secondary node",
15002 errors.ECODE_STATE)
15004 self.required_nodes = 1
15005 disk_sizes = [{constants.IDISK_SIZE: disk.size} for disk in instance.disks]
15006 disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
15010 "disk_space_total": disk_space,
15011 "required_nodes": self.required_nodes,
15012 "relocate_from": self.relocate_from,
15016 def _AddNodeEvacuate(self):
15017 """Get data for node-evacuate requests.
15021 "instances": self.instances,
15022 "evac_mode": self.evac_mode,
15025 def _AddChangeGroup(self):
15026 """Get data for node-evacuate requests.
15030 "instances": self.instances,
15031 "target_groups": self.target_groups,
15034 def _BuildInputData(self, fn, keydata):
15035 """Build input data structures.
15038 self._ComputeClusterData()
15041 request["type"] = self.mode
15042 for keyname, keytype in keydata:
15043 if keyname not in request:
15044 raise errors.ProgrammerError("Request parameter %s is missing" %
15046 val = request[keyname]
15047 if not keytype(val):
15048 raise errors.ProgrammerError("Request parameter %s doesn't pass"
15049 " validation, value %s, expected"
15050 " type %s" % (keyname, val, keytype))
15051 self.in_data["request"] = request
15053 self.in_text = serializer.Dump(self.in_data)
15055 _STRING_LIST = ht.TListOf(ht.TString)
15056 _JOB_LIST = ht.TListOf(ht.TListOf(ht.TStrictDict(True, False, {
15057 # pylint: disable=E1101
15058 # Class '...' has no 'OP_ID' member
15059 "OP_ID": ht.TElemOf([opcodes.OpInstanceFailover.OP_ID,
15060 opcodes.OpInstanceMigrate.OP_ID,
15061 opcodes.OpInstanceReplaceDisks.OP_ID])
15065 ht.TListOf(ht.TAnd(ht.TIsLength(3),
15066 ht.TItems([ht.TNonEmptyString,
15067 ht.TNonEmptyString,
15068 ht.TListOf(ht.TNonEmptyString),
15071 ht.TListOf(ht.TAnd(ht.TIsLength(2),
15072 ht.TItems([ht.TNonEmptyString,
15075 _NEVAC_RESULT = ht.TAnd(ht.TIsLength(3),
15076 ht.TItems([_NEVAC_MOVED, _NEVAC_FAILED, _JOB_LIST]))
15079 constants.IALLOCATOR_MODE_ALLOC:
15082 ("name", ht.TString),
15083 ("memory", ht.TInt),
15084 ("spindle_use", ht.TInt),
15085 ("disks", ht.TListOf(ht.TDict)),
15086 ("disk_template", ht.TString),
15087 ("os", ht.TString),
15088 ("tags", _STRING_LIST),
15089 ("nics", ht.TListOf(ht.TDict)),
15090 ("vcpus", ht.TInt),
15091 ("hypervisor", ht.TString),
15093 constants.IALLOCATOR_MODE_RELOC:
15094 (_AddRelocateInstance,
15095 [("name", ht.TString), ("relocate_from", _STRING_LIST)],
15097 constants.IALLOCATOR_MODE_NODE_EVAC:
15098 (_AddNodeEvacuate, [
15099 ("instances", _STRING_LIST),
15100 ("evac_mode", ht.TElemOf(constants.IALLOCATOR_NEVAC_MODES)),
15102 constants.IALLOCATOR_MODE_CHG_GROUP:
15103 (_AddChangeGroup, [
15104 ("instances", _STRING_LIST),
15105 ("target_groups", _STRING_LIST),
15109 def Run(self, name, validate=True, call_fn=None):
15110 """Run an instance allocator and return the results.
15113 if call_fn is None:
15114 call_fn = self.rpc.call_iallocator_runner
15116 result = call_fn(self.cfg.GetMasterNode(), name, self.in_text)
15117 result.Raise("Failure while running the iallocator script")
15119 self.out_text = result.payload
15121 self._ValidateResult()
15123 def _ValidateResult(self):
15124 """Process the allocator results.
15126 This will process and if successful save the result in
15127 self.out_data and the other parameters.
15131 rdict = serializer.Load(self.out_text)
15132 except Exception, err:
15133 raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
15135 if not isinstance(rdict, dict):
15136 raise errors.OpExecError("Can't parse iallocator results: not a dict")
15138 # TODO: remove backwards compatiblity in later versions
15139 if "nodes" in rdict and "result" not in rdict:
15140 rdict["result"] = rdict["nodes"]
15143 for key in "success", "info", "result":
15144 if key not in rdict:
15145 raise errors.OpExecError("Can't parse iallocator results:"
15146 " missing key '%s'" % key)
15147 setattr(self, key, rdict[key])
15149 if not self._result_check(self.result):
15150 raise errors.OpExecError("Iallocator returned invalid result,"
15151 " expected %s, got %s" %
15152 (self._result_check, self.result),
15153 errors.ECODE_INVAL)
15155 if self.mode == constants.IALLOCATOR_MODE_RELOC:
15156 assert self.relocate_from is not None
15157 assert self.required_nodes == 1
15159 node2group = dict((name, ndata["group"])
15160 for (name, ndata) in self.in_data["nodes"].items())
15162 fn = compat.partial(self._NodesToGroups, node2group,
15163 self.in_data["nodegroups"])
15165 instance = self.cfg.GetInstanceInfo(self.name)
15166 request_groups = fn(self.relocate_from + [instance.primary_node])
15167 result_groups = fn(rdict["result"] + [instance.primary_node])
15169 if self.success and not set(result_groups).issubset(request_groups):
15170 raise errors.OpExecError("Groups of nodes returned by iallocator (%s)"
15171 " differ from original groups (%s)" %
15172 (utils.CommaJoin(result_groups),
15173 utils.CommaJoin(request_groups)))
15175 elif self.mode == constants.IALLOCATOR_MODE_NODE_EVAC:
15176 assert self.evac_mode in constants.IALLOCATOR_NEVAC_MODES
15178 self.out_data = rdict
15181 def _NodesToGroups(node2group, groups, nodes):
15182 """Returns a list of unique group names for a list of nodes.
15184 @type node2group: dict
15185 @param node2group: Map from node name to group UUID
15187 @param groups: Group information
15189 @param nodes: Node names
15196 group_uuid = node2group[node]
15198 # Ignore unknown node
15202 group = groups[group_uuid]
15204 # Can't find group, let's use UUID
15205 group_name = group_uuid
15207 group_name = group["name"]
15209 result.add(group_name)
15211 return sorted(result)
15214 class LUTestAllocator(NoHooksLU):
15215 """Run allocator tests.
15217 This LU runs the allocator tests
15220 def CheckPrereq(self):
15221 """Check prerequisites.
15223 This checks the opcode parameters depending on the director and mode test.
15226 if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
15227 for attr in ["memory", "disks", "disk_template",
15228 "os", "tags", "nics", "vcpus"]:
15229 if not hasattr(self.op, attr):
15230 raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
15231 attr, errors.ECODE_INVAL)
15232 iname = self.cfg.ExpandInstanceName(self.op.name)
15233 if iname is not None:
15234 raise errors.OpPrereqError("Instance '%s' already in the cluster" %
15235 iname, errors.ECODE_EXISTS)
15236 if not isinstance(self.op.nics, list):
15237 raise errors.OpPrereqError("Invalid parameter 'nics'",
15238 errors.ECODE_INVAL)
15239 if not isinstance(self.op.disks, list):
15240 raise errors.OpPrereqError("Invalid parameter 'disks'",
15241 errors.ECODE_INVAL)
15242 for row in self.op.disks:
15243 if (not isinstance(row, dict) or
15244 constants.IDISK_SIZE not in row or
15245 not isinstance(row[constants.IDISK_SIZE], int) or
15246 constants.IDISK_MODE not in row or
15247 row[constants.IDISK_MODE] not in constants.DISK_ACCESS_SET):
15248 raise errors.OpPrereqError("Invalid contents of the 'disks'"
15249 " parameter", errors.ECODE_INVAL)
15250 if self.op.hypervisor is None:
15251 self.op.hypervisor = self.cfg.GetHypervisorType()
15252 elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
15253 fname = _ExpandInstanceName(self.cfg, self.op.name)
15254 self.op.name = fname
15255 self.relocate_from = \
15256 list(self.cfg.GetInstanceInfo(fname).secondary_nodes)
15257 elif self.op.mode in (constants.IALLOCATOR_MODE_CHG_GROUP,
15258 constants.IALLOCATOR_MODE_NODE_EVAC):
15259 if not self.op.instances:
15260 raise errors.OpPrereqError("Missing instances", errors.ECODE_INVAL)
15261 self.op.instances = _GetWantedInstances(self, self.op.instances)
15263 raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
15264 self.op.mode, errors.ECODE_INVAL)
15266 if self.op.direction == constants.IALLOCATOR_DIR_OUT:
15267 if self.op.allocator is None:
15268 raise errors.OpPrereqError("Missing allocator name",
15269 errors.ECODE_INVAL)
15270 elif self.op.direction != constants.IALLOCATOR_DIR_IN:
15271 raise errors.OpPrereqError("Wrong allocator test '%s'" %
15272 self.op.direction, errors.ECODE_INVAL)
15274 def Exec(self, feedback_fn):
15275 """Run the allocator test.
15278 if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
15279 ial = IAllocator(self.cfg, self.rpc,
15282 memory=self.op.memory,
15283 disks=self.op.disks,
15284 disk_template=self.op.disk_template,
15288 vcpus=self.op.vcpus,
15289 hypervisor=self.op.hypervisor,
15291 elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
15292 ial = IAllocator(self.cfg, self.rpc,
15295 relocate_from=list(self.relocate_from),
15297 elif self.op.mode == constants.IALLOCATOR_MODE_CHG_GROUP:
15298 ial = IAllocator(self.cfg, self.rpc,
15300 instances=self.op.instances,
15301 target_groups=self.op.target_groups)
15302 elif self.op.mode == constants.IALLOCATOR_MODE_NODE_EVAC:
15303 ial = IAllocator(self.cfg, self.rpc,
15305 instances=self.op.instances,
15306 evac_mode=self.op.evac_mode)
15308 raise errors.ProgrammerError("Uncatched mode %s in"
15309 " LUTestAllocator.Exec", self.op.mode)
15311 if self.op.direction == constants.IALLOCATOR_DIR_IN:
15312 result = ial.in_text
15314 ial.Run(self.op.allocator, validate=False)
15315 result = ial.out_text
15319 #: Query type implementations
15321 constants.QR_CLUSTER: _ClusterQuery,
15322 constants.QR_INSTANCE: _InstanceQuery,
15323 constants.QR_NODE: _NodeQuery,
15324 constants.QR_GROUP: _GroupQuery,
15325 constants.QR_OS: _OsQuery,
15326 constants.QR_EXPORT: _ExportQuery,
15329 assert set(_QUERY_IMPL.keys()) == constants.QR_VIA_OP
15332 def _GetQueryImplementation(name):
15333 """Returns the implemtnation for a query type.
15335 @param name: Query type, must be one of L{constants.QR_VIA_OP}
15339 return _QUERY_IMPL[name]
15341 raise errors.OpPrereqError("Unknown query resource '%s'" % name,
15342 errors.ECODE_INVAL)