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 # Creating copies as SetDiskID below will modify the objects and that can
2928 # lead to incorrect data returned from nodes
2929 devonly = [dev.Copy() for (_, dev) in disks]
2932 self.cfg.SetDiskID(dev, nname)
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 # Files which are optional, these must:
4314 # - be present in one other category as well
4315 # - either exist or not exist on all nodes of that category (mc, vm all)
4317 constants.RAPI_USERS_FILE,
4320 # Files which should only be on master candidates
4324 files_mc.add(constants.CLUSTER_CONF_FILE)
4326 # FIXME: this should also be replicated but Ganeti doesn't support files_mc
4328 files_mc.add(constants.DEFAULT_MASTER_SETUP_SCRIPT)
4330 # Files which should only be on VM-capable nodes
4331 files_vm = set(filename
4332 for hv_name in cluster.enabled_hypervisors
4333 for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[0])
4335 files_opt |= set(filename
4336 for hv_name in cluster.enabled_hypervisors
4337 for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[1])
4339 # Filenames in each category must be unique
4340 all_files_set = files_all | files_mc | files_vm
4341 assert (len(all_files_set) ==
4342 sum(map(len, [files_all, files_mc, files_vm]))), \
4343 "Found file listed in more than one file list"
4345 # Optional files must be present in one other category
4346 assert all_files_set.issuperset(files_opt), \
4347 "Optional file not in a different required list"
4349 return (files_all, files_opt, files_mc, files_vm)
4352 def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
4353 """Distribute additional files which are part of the cluster configuration.
4355 ConfigWriter takes care of distributing the config and ssconf files, but
4356 there are more files which should be distributed to all nodes. This function
4357 makes sure those are copied.
4359 @param lu: calling logical unit
4360 @param additional_nodes: list of nodes not in the config to distribute to
4361 @type additional_vm: boolean
4362 @param additional_vm: whether the additional nodes are vm-capable or not
4365 # Gather target nodes
4366 cluster = lu.cfg.GetClusterInfo()
4367 master_info = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
4369 online_nodes = lu.cfg.GetOnlineNodeList()
4370 vm_nodes = lu.cfg.GetVmCapableNodeList()
4372 if additional_nodes is not None:
4373 online_nodes.extend(additional_nodes)
4375 vm_nodes.extend(additional_nodes)
4377 # Never distribute to master node
4378 for nodelist in [online_nodes, vm_nodes]:
4379 if master_info.name in nodelist:
4380 nodelist.remove(master_info.name)
4383 (files_all, _, files_mc, files_vm) = \
4384 _ComputeAncillaryFiles(cluster, True)
4386 # Never re-distribute configuration file from here
4387 assert not (constants.CLUSTER_CONF_FILE in files_all or
4388 constants.CLUSTER_CONF_FILE in files_vm)
4389 assert not files_mc, "Master candidates not handled in this function"
4392 (online_nodes, files_all),
4393 (vm_nodes, files_vm),
4397 for (node_list, files) in filemap:
4399 _UploadHelper(lu, node_list, fname)
4402 class LUClusterRedistConf(NoHooksLU):
4403 """Force the redistribution of cluster configuration.
4405 This is a very simple LU.
4410 def ExpandNames(self):
4411 self.needed_locks = {
4412 locking.LEVEL_NODE: locking.ALL_SET,
4414 self.share_locks[locking.LEVEL_NODE] = 1
4416 def Exec(self, feedback_fn):
4417 """Redistribute the configuration.
4420 self.cfg.Update(self.cfg.GetClusterInfo(), feedback_fn)
4421 _RedistributeAncillaryFiles(self)
4424 class LUClusterActivateMasterIp(NoHooksLU):
4425 """Activate the master IP on the master node.
4428 def Exec(self, feedback_fn):
4429 """Activate the master IP.
4432 master_params = self.cfg.GetMasterNetworkParameters()
4433 ems = self.cfg.GetUseExternalMipScript()
4434 result = self.rpc.call_node_activate_master_ip(master_params.name,
4436 result.Raise("Could not activate the master IP")
4439 class LUClusterDeactivateMasterIp(NoHooksLU):
4440 """Deactivate the master IP on the master node.
4443 def Exec(self, feedback_fn):
4444 """Deactivate the master IP.
4447 master_params = self.cfg.GetMasterNetworkParameters()
4448 ems = self.cfg.GetUseExternalMipScript()
4449 result = self.rpc.call_node_deactivate_master_ip(master_params.name,
4451 result.Raise("Could not deactivate the master IP")
4454 def _WaitForSync(lu, instance, disks=None, oneshot=False):
4455 """Sleep and poll for an instance's disk to sync.
4458 if not instance.disks or disks is not None and not disks:
4461 disks = _ExpandCheckDisks(instance, disks)
4464 lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
4466 node = instance.primary_node
4469 lu.cfg.SetDiskID(dev, node)
4471 # TODO: Convert to utils.Retry
4474 degr_retries = 10 # in seconds, as we sleep 1 second each time
4478 cumul_degraded = False
4479 rstats = lu.rpc.call_blockdev_getmirrorstatus(node, (disks, instance))
4480 msg = rstats.fail_msg
4482 lu.LogWarning("Can't get any data from node %s: %s", node, msg)
4485 raise errors.RemoteError("Can't contact node %s for mirror data,"
4486 " aborting." % node)
4489 rstats = rstats.payload
4491 for i, mstat in enumerate(rstats):
4493 lu.LogWarning("Can't compute data for node %s/%s",
4494 node, disks[i].iv_name)
4497 cumul_degraded = (cumul_degraded or
4498 (mstat.is_degraded and mstat.sync_percent is None))
4499 if mstat.sync_percent is not None:
4501 if mstat.estimated_time is not None:
4502 rem_time = ("%s remaining (estimated)" %
4503 utils.FormatSeconds(mstat.estimated_time))
4504 max_time = mstat.estimated_time
4506 rem_time = "no time estimate"
4507 lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
4508 (disks[i].iv_name, mstat.sync_percent, rem_time))
4510 # if we're done but degraded, let's do a few small retries, to
4511 # make sure we see a stable and not transient situation; therefore
4512 # we force restart of the loop
4513 if (done or oneshot) and cumul_degraded and degr_retries > 0:
4514 logging.info("Degraded disks found, %d retries left", degr_retries)
4522 time.sleep(min(60, max_time))
4525 lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
4526 return not cumul_degraded
4529 def _BlockdevFind(lu, node, dev, instance):
4530 """Wrapper around call_blockdev_find to annotate diskparams.
4532 @param lu: A reference to the lu object
4533 @param node: The node to call out
4534 @param dev: The device to find
4535 @param instance: The instance object the device belongs to
4536 @returns The result of the rpc call
4539 (disk,) = _AnnotateDiskParams(instance, [dev], lu.cfg)
4540 return lu.rpc.call_blockdev_find(node, disk)
4543 def _CheckDiskConsistency(lu, instance, dev, node, on_primary, ldisk=False):
4544 """Wrapper around L{_CheckDiskConsistencyInner}.
4547 (disk,) = _AnnotateDiskParams(instance, [dev], lu.cfg)
4548 return _CheckDiskConsistencyInner(lu, instance, disk, node, on_primary,
4552 def _CheckDiskConsistencyInner(lu, instance, dev, node, on_primary,
4554 """Check that mirrors are not degraded.
4556 @attention: The device has to be annotated already.
4558 The ldisk parameter, if True, will change the test from the
4559 is_degraded attribute (which represents overall non-ok status for
4560 the device(s)) to the ldisk (representing the local storage status).
4563 lu.cfg.SetDiskID(dev, node)
4567 if on_primary or dev.AssembleOnSecondary():
4568 rstats = lu.rpc.call_blockdev_find(node, dev)
4569 msg = rstats.fail_msg
4571 lu.LogWarning("Can't find disk on node %s: %s", node, msg)
4573 elif not rstats.payload:
4574 lu.LogWarning("Can't find disk on node %s", node)
4578 result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
4580 result = result and not rstats.payload.is_degraded
4583 for child in dev.children:
4584 result = result and _CheckDiskConsistencyInner(lu, instance, child, node,
4590 class LUOobCommand(NoHooksLU):
4591 """Logical unit for OOB handling.
4595 _SKIP_MASTER = (constants.OOB_POWER_OFF, constants.OOB_POWER_CYCLE)
4597 def ExpandNames(self):
4598 """Gather locks we need.
4601 if self.op.node_names:
4602 self.op.node_names = _GetWantedNodes(self, self.op.node_names)
4603 lock_names = self.op.node_names
4605 lock_names = locking.ALL_SET
4607 self.needed_locks = {
4608 locking.LEVEL_NODE: lock_names,
4611 def CheckPrereq(self):
4612 """Check prerequisites.
4615 - the node exists in the configuration
4618 Any errors are signaled by raising errors.OpPrereqError.
4622 self.master_node = self.cfg.GetMasterNode()
4624 assert self.op.power_delay >= 0.0
4626 if self.op.node_names:
4627 if (self.op.command in self._SKIP_MASTER and
4628 self.master_node in self.op.node_names):
4629 master_node_obj = self.cfg.GetNodeInfo(self.master_node)
4630 master_oob_handler = _SupportsOob(self.cfg, master_node_obj)
4632 if master_oob_handler:
4633 additional_text = ("run '%s %s %s' if you want to operate on the"
4634 " master regardless") % (master_oob_handler,
4638 additional_text = "it does not support out-of-band operations"
4640 raise errors.OpPrereqError(("Operating on the master node %s is not"
4641 " allowed for %s; %s") %
4642 (self.master_node, self.op.command,
4643 additional_text), errors.ECODE_INVAL)
4645 self.op.node_names = self.cfg.GetNodeList()
4646 if self.op.command in self._SKIP_MASTER:
4647 self.op.node_names.remove(self.master_node)
4649 if self.op.command in self._SKIP_MASTER:
4650 assert self.master_node not in self.op.node_names
4652 for (node_name, node) in self.cfg.GetMultiNodeInfo(self.op.node_names):
4654 raise errors.OpPrereqError("Node %s not found" % node_name,
4657 self.nodes.append(node)
4659 if (not self.op.ignore_status and
4660 (self.op.command == constants.OOB_POWER_OFF and not node.offline)):
4661 raise errors.OpPrereqError(("Cannot power off node %s because it is"
4662 " not marked offline") % node_name,
4665 def Exec(self, feedback_fn):
4666 """Execute OOB and return result if we expect any.
4669 master_node = self.master_node
4672 for idx, node in enumerate(utils.NiceSort(self.nodes,
4673 key=lambda node: node.name)):
4674 node_entry = [(constants.RS_NORMAL, node.name)]
4675 ret.append(node_entry)
4677 oob_program = _SupportsOob(self.cfg, node)
4680 node_entry.append((constants.RS_UNAVAIL, None))
4683 logging.info("Executing out-of-band command '%s' using '%s' on %s",
4684 self.op.command, oob_program, node.name)
4685 result = self.rpc.call_run_oob(master_node, oob_program,
4686 self.op.command, node.name,
4690 self.LogWarning("Out-of-band RPC failed on node '%s': %s",
4691 node.name, result.fail_msg)
4692 node_entry.append((constants.RS_NODATA, None))
4695 self._CheckPayload(result)
4696 except errors.OpExecError, err:
4697 self.LogWarning("Payload returned by node '%s' is not valid: %s",
4699 node_entry.append((constants.RS_NODATA, None))
4701 if self.op.command == constants.OOB_HEALTH:
4702 # For health we should log important events
4703 for item, status in result.payload:
4704 if status in [constants.OOB_STATUS_WARNING,
4705 constants.OOB_STATUS_CRITICAL]:
4706 self.LogWarning("Item '%s' on node '%s' has status '%s'",
4707 item, node.name, status)
4709 if self.op.command == constants.OOB_POWER_ON:
4711 elif self.op.command == constants.OOB_POWER_OFF:
4712 node.powered = False
4713 elif self.op.command == constants.OOB_POWER_STATUS:
4714 powered = result.payload[constants.OOB_POWER_STATUS_POWERED]
4715 if powered != node.powered:
4716 logging.warning(("Recorded power state (%s) of node '%s' does not"
4717 " match actual power state (%s)"), node.powered,
4720 # For configuration changing commands we should update the node
4721 if self.op.command in (constants.OOB_POWER_ON,
4722 constants.OOB_POWER_OFF):
4723 self.cfg.Update(node, feedback_fn)
4725 node_entry.append((constants.RS_NORMAL, result.payload))
4727 if (self.op.command == constants.OOB_POWER_ON and
4728 idx < len(self.nodes) - 1):
4729 time.sleep(self.op.power_delay)
4733 def _CheckPayload(self, result):
4734 """Checks if the payload is valid.
4736 @param result: RPC result
4737 @raises errors.OpExecError: If payload is not valid
4741 if self.op.command == constants.OOB_HEALTH:
4742 if not isinstance(result.payload, list):
4743 errs.append("command 'health' is expected to return a list but got %s" %
4744 type(result.payload))
4746 for item, status in result.payload:
4747 if status not in constants.OOB_STATUSES:
4748 errs.append("health item '%s' has invalid status '%s'" %
4751 if self.op.command == constants.OOB_POWER_STATUS:
4752 if not isinstance(result.payload, dict):
4753 errs.append("power-status is expected to return a dict but got %s" %
4754 type(result.payload))
4756 if self.op.command in [
4757 constants.OOB_POWER_ON,
4758 constants.OOB_POWER_OFF,
4759 constants.OOB_POWER_CYCLE,
4761 if result.payload is not None:
4762 errs.append("%s is expected to not return payload but got '%s'" %
4763 (self.op.command, result.payload))
4766 raise errors.OpExecError("Check of out-of-band payload failed due to %s" %
4767 utils.CommaJoin(errs))
4770 class _OsQuery(_QueryBase):
4771 FIELDS = query.OS_FIELDS
4773 def ExpandNames(self, lu):
4774 # Lock all nodes in shared mode
4775 # Temporary removal of locks, should be reverted later
4776 # TODO: reintroduce locks when they are lighter-weight
4777 lu.needed_locks = {}
4778 #self.share_locks[locking.LEVEL_NODE] = 1
4779 #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4781 # The following variables interact with _QueryBase._GetNames
4783 self.wanted = self.names
4785 self.wanted = locking.ALL_SET
4787 self.do_locking = self.use_locking
4789 def DeclareLocks(self, lu, level):
4793 def _DiagnoseByOS(rlist):
4794 """Remaps a per-node return list into an a per-os per-node dictionary
4796 @param rlist: a map with node names as keys and OS objects as values
4799 @return: a dictionary with osnames as keys and as value another
4800 map, with nodes as keys and tuples of (path, status, diagnose,
4801 variants, parameters, api_versions) as values, eg::
4803 {"debian-etch": {"node1": [(/usr/lib/..., True, "", [], []),
4804 (/srv/..., False, "invalid api")],
4805 "node2": [(/srv/..., True, "", [], [])]}
4810 # we build here the list of nodes that didn't fail the RPC (at RPC
4811 # level), so that nodes with a non-responding node daemon don't
4812 # make all OSes invalid
4813 good_nodes = [node_name for node_name in rlist
4814 if not rlist[node_name].fail_msg]
4815 for node_name, nr in rlist.items():
4816 if nr.fail_msg or not nr.payload:
4818 for (name, path, status, diagnose, variants,
4819 params, api_versions) in nr.payload:
4820 if name not in all_os:
4821 # build a list of nodes for this os containing empty lists
4822 # for each node in node_list
4824 for nname in good_nodes:
4825 all_os[name][nname] = []
4826 # convert params from [name, help] to (name, help)
4827 params = [tuple(v) for v in params]
4828 all_os[name][node_name].append((path, status, diagnose,
4829 variants, params, api_versions))
4832 def _GetQueryData(self, lu):
4833 """Computes the list of nodes and their attributes.
4836 # Locking is not used
4837 assert not (compat.any(lu.glm.is_owned(level)
4838 for level in locking.LEVELS
4839 if level != locking.LEVEL_CLUSTER) or
4840 self.do_locking or self.use_locking)
4842 valid_nodes = [node.name
4843 for node in lu.cfg.GetAllNodesInfo().values()
4844 if not node.offline and node.vm_capable]
4845 pol = self._DiagnoseByOS(lu.rpc.call_os_diagnose(valid_nodes))
4846 cluster = lu.cfg.GetClusterInfo()
4850 for (os_name, os_data) in pol.items():
4851 info = query.OsInfo(name=os_name, valid=True, node_status=os_data,
4852 hidden=(os_name in cluster.hidden_os),
4853 blacklisted=(os_name in cluster.blacklisted_os))
4857 api_versions = set()
4859 for idx, osl in enumerate(os_data.values()):
4860 info.valid = bool(info.valid and osl and osl[0][1])
4864 (node_variants, node_params, node_api) = osl[0][3:6]
4867 variants.update(node_variants)
4868 parameters.update(node_params)
4869 api_versions.update(node_api)
4871 # Filter out inconsistent values
4872 variants.intersection_update(node_variants)
4873 parameters.intersection_update(node_params)
4874 api_versions.intersection_update(node_api)
4876 info.variants = list(variants)
4877 info.parameters = list(parameters)
4878 info.api_versions = list(api_versions)
4880 data[os_name] = info
4882 # Prepare data in requested order
4883 return [data[name] for name in self._GetNames(lu, pol.keys(), None)
4887 class LUOsDiagnose(NoHooksLU):
4888 """Logical unit for OS diagnose/query.
4894 def _BuildFilter(fields, names):
4895 """Builds a filter for querying OSes.
4898 name_filter = qlang.MakeSimpleFilter("name", names)
4900 # Legacy behaviour: Hide hidden, blacklisted or invalid OSes if the
4901 # respective field is not requested
4902 status_filter = [[qlang.OP_NOT, [qlang.OP_TRUE, fname]]
4903 for fname in ["hidden", "blacklisted"]
4904 if fname not in fields]
4905 if "valid" not in fields:
4906 status_filter.append([qlang.OP_TRUE, "valid"])
4909 status_filter.insert(0, qlang.OP_AND)
4911 status_filter = None
4913 if name_filter and status_filter:
4914 return [qlang.OP_AND, name_filter, status_filter]
4918 return status_filter
4920 def CheckArguments(self):
4921 self.oq = _OsQuery(self._BuildFilter(self.op.output_fields, self.op.names),
4922 self.op.output_fields, False)
4924 def ExpandNames(self):
4925 self.oq.ExpandNames(self)
4927 def Exec(self, feedback_fn):
4928 return self.oq.OldStyleQuery(self)
4931 class LUNodeRemove(LogicalUnit):
4932 """Logical unit for removing a node.
4935 HPATH = "node-remove"
4936 HTYPE = constants.HTYPE_NODE
4938 def BuildHooksEnv(self):
4943 "OP_TARGET": self.op.node_name,
4944 "NODE_NAME": self.op.node_name,
4947 def BuildHooksNodes(self):
4948 """Build hooks nodes.
4950 This doesn't run on the target node in the pre phase as a failed
4951 node would then be impossible to remove.
4954 all_nodes = self.cfg.GetNodeList()
4956 all_nodes.remove(self.op.node_name)
4959 return (all_nodes, all_nodes)
4961 def CheckPrereq(self):
4962 """Check prerequisites.
4965 - the node exists in the configuration
4966 - it does not have primary or secondary instances
4967 - it's not the master
4969 Any errors are signaled by raising errors.OpPrereqError.
4972 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4973 node = self.cfg.GetNodeInfo(self.op.node_name)
4974 assert node is not None
4976 masternode = self.cfg.GetMasterNode()
4977 if node.name == masternode:
4978 raise errors.OpPrereqError("Node is the master node, failover to another"
4979 " node is required", errors.ECODE_INVAL)
4981 for instance_name, instance in self.cfg.GetAllInstancesInfo().items():
4982 if node.name in instance.all_nodes:
4983 raise errors.OpPrereqError("Instance %s is still running on the node,"
4984 " please remove first" % instance_name,
4986 self.op.node_name = node.name
4989 def Exec(self, feedback_fn):
4990 """Removes the node from the cluster.
4994 logging.info("Stopping the node daemon and removing configs from node %s",
4997 modify_ssh_setup = self.cfg.GetClusterInfo().modify_ssh_setup
4999 assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER), \
5002 # Promote nodes to master candidate as needed
5003 _AdjustCandidatePool(self, exceptions=[node.name])
5004 self.context.RemoveNode(node.name)
5006 # Run post hooks on the node before it's removed
5007 _RunPostHook(self, node.name)
5009 result = self.rpc.call_node_leave_cluster(node.name, modify_ssh_setup)
5010 msg = result.fail_msg
5012 self.LogWarning("Errors encountered on the remote node while leaving"
5013 " the cluster: %s", msg)
5015 # Remove node from our /etc/hosts
5016 if self.cfg.GetClusterInfo().modify_etc_hosts:
5017 master_node = self.cfg.GetMasterNode()
5018 result = self.rpc.call_etc_hosts_modify(master_node,
5019 constants.ETC_HOSTS_REMOVE,
5021 result.Raise("Can't update hosts file with new host data")
5022 _RedistributeAncillaryFiles(self)
5025 class _NodeQuery(_QueryBase):
5026 FIELDS = query.NODE_FIELDS
5028 def ExpandNames(self, lu):
5029 lu.needed_locks = {}
5030 lu.share_locks = _ShareAll()
5033 self.wanted = _GetWantedNodes(lu, self.names)
5035 self.wanted = locking.ALL_SET
5037 self.do_locking = (self.use_locking and
5038 query.NQ_LIVE in self.requested_data)
5041 # If any non-static field is requested we need to lock the nodes
5042 lu.needed_locks[locking.LEVEL_NODE] = self.wanted
5044 def DeclareLocks(self, lu, level):
5047 def _GetQueryData(self, lu):
5048 """Computes the list of nodes and their attributes.
5051 all_info = lu.cfg.GetAllNodesInfo()
5053 nodenames = self._GetNames(lu, all_info.keys(), locking.LEVEL_NODE)
5055 # Gather data as requested
5056 if query.NQ_LIVE in self.requested_data:
5057 # filter out non-vm_capable nodes
5058 toquery_nodes = [name for name in nodenames if all_info[name].vm_capable]
5060 node_data = lu.rpc.call_node_info(toquery_nodes, [lu.cfg.GetVGName()],
5061 [lu.cfg.GetHypervisorType()])
5062 live_data = dict((name, _MakeLegacyNodeInfo(nresult.payload))
5063 for (name, nresult) in node_data.items()
5064 if not nresult.fail_msg and nresult.payload)
5068 if query.NQ_INST in self.requested_data:
5069 node_to_primary = dict([(name, set()) for name in nodenames])
5070 node_to_secondary = dict([(name, set()) for name in nodenames])
5072 inst_data = lu.cfg.GetAllInstancesInfo()
5074 for inst in inst_data.values():
5075 if inst.primary_node in node_to_primary:
5076 node_to_primary[inst.primary_node].add(inst.name)
5077 for secnode in inst.secondary_nodes:
5078 if secnode in node_to_secondary:
5079 node_to_secondary[secnode].add(inst.name)
5081 node_to_primary = None
5082 node_to_secondary = None
5084 if query.NQ_OOB in self.requested_data:
5085 oob_support = dict((name, bool(_SupportsOob(lu.cfg, node)))
5086 for name, node in all_info.iteritems())
5090 if query.NQ_GROUP in self.requested_data:
5091 groups = lu.cfg.GetAllNodeGroupsInfo()
5095 return query.NodeQueryData([all_info[name] for name in nodenames],
5096 live_data, lu.cfg.GetMasterNode(),
5097 node_to_primary, node_to_secondary, groups,
5098 oob_support, lu.cfg.GetClusterInfo())
5101 class LUNodeQuery(NoHooksLU):
5102 """Logical unit for querying nodes.
5105 # pylint: disable=W0142
5108 def CheckArguments(self):
5109 self.nq = _NodeQuery(qlang.MakeSimpleFilter("name", self.op.names),
5110 self.op.output_fields, self.op.use_locking)
5112 def ExpandNames(self):
5113 self.nq.ExpandNames(self)
5115 def DeclareLocks(self, level):
5116 self.nq.DeclareLocks(self, level)
5118 def Exec(self, feedback_fn):
5119 return self.nq.OldStyleQuery(self)
5122 class LUNodeQueryvols(NoHooksLU):
5123 """Logical unit for getting volumes on node(s).
5127 _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
5128 _FIELDS_STATIC = utils.FieldSet("node")
5130 def CheckArguments(self):
5131 _CheckOutputFields(static=self._FIELDS_STATIC,
5132 dynamic=self._FIELDS_DYNAMIC,
5133 selected=self.op.output_fields)
5135 def ExpandNames(self):
5136 self.share_locks = _ShareAll()
5137 self.needed_locks = {}
5139 if not self.op.nodes:
5140 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
5142 self.needed_locks[locking.LEVEL_NODE] = \
5143 _GetWantedNodes(self, self.op.nodes)
5145 def Exec(self, feedback_fn):
5146 """Computes the list of nodes and their attributes.
5149 nodenames = self.owned_locks(locking.LEVEL_NODE)
5150 volumes = self.rpc.call_node_volumes(nodenames)
5152 ilist = self.cfg.GetAllInstancesInfo()
5153 vol2inst = _MapInstanceDisksToNodes(ilist.values())
5156 for node in nodenames:
5157 nresult = volumes[node]
5160 msg = nresult.fail_msg
5162 self.LogWarning("Can't compute volume data on node %s: %s", node, msg)
5165 node_vols = sorted(nresult.payload,
5166 key=operator.itemgetter("dev"))
5168 for vol in node_vols:
5170 for field in self.op.output_fields:
5173 elif field == "phys":
5177 elif field == "name":
5179 elif field == "size":
5180 val = int(float(vol["size"]))
5181 elif field == "instance":
5182 val = vol2inst.get((node, vol["vg"] + "/" + vol["name"]), "-")
5184 raise errors.ParameterError(field)
5185 node_output.append(str(val))
5187 output.append(node_output)
5192 class LUNodeQueryStorage(NoHooksLU):
5193 """Logical unit for getting information on storage units on node(s).
5196 _FIELDS_STATIC = utils.FieldSet(constants.SF_NODE)
5199 def CheckArguments(self):
5200 _CheckOutputFields(static=self._FIELDS_STATIC,
5201 dynamic=utils.FieldSet(*constants.VALID_STORAGE_FIELDS),
5202 selected=self.op.output_fields)
5204 def ExpandNames(self):
5205 self.share_locks = _ShareAll()
5206 self.needed_locks = {}
5209 self.needed_locks[locking.LEVEL_NODE] = \
5210 _GetWantedNodes(self, self.op.nodes)
5212 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
5214 def Exec(self, feedback_fn):
5215 """Computes the list of nodes and their attributes.
5218 self.nodes = self.owned_locks(locking.LEVEL_NODE)
5220 # Always get name to sort by
5221 if constants.SF_NAME in self.op.output_fields:
5222 fields = self.op.output_fields[:]
5224 fields = [constants.SF_NAME] + self.op.output_fields
5226 # Never ask for node or type as it's only known to the LU
5227 for extra in [constants.SF_NODE, constants.SF_TYPE]:
5228 while extra in fields:
5229 fields.remove(extra)
5231 field_idx = dict([(name, idx) for (idx, name) in enumerate(fields)])
5232 name_idx = field_idx[constants.SF_NAME]
5234 st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
5235 data = self.rpc.call_storage_list(self.nodes,
5236 self.op.storage_type, st_args,
5237 self.op.name, fields)
5241 for node in utils.NiceSort(self.nodes):
5242 nresult = data[node]
5246 msg = nresult.fail_msg
5248 self.LogWarning("Can't get storage data from node %s: %s", node, msg)
5251 rows = dict([(row[name_idx], row) for row in nresult.payload])
5253 for name in utils.NiceSort(rows.keys()):
5258 for field in self.op.output_fields:
5259 if field == constants.SF_NODE:
5261 elif field == constants.SF_TYPE:
5262 val = self.op.storage_type
5263 elif field in field_idx:
5264 val = row[field_idx[field]]
5266 raise errors.ParameterError(field)
5275 class _InstanceQuery(_QueryBase):
5276 FIELDS = query.INSTANCE_FIELDS
5278 def ExpandNames(self, lu):
5279 lu.needed_locks = {}
5280 lu.share_locks = _ShareAll()
5283 self.wanted = _GetWantedInstances(lu, self.names)
5285 self.wanted = locking.ALL_SET
5287 self.do_locking = (self.use_locking and
5288 query.IQ_LIVE in self.requested_data)
5290 lu.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
5291 lu.needed_locks[locking.LEVEL_NODEGROUP] = []
5292 lu.needed_locks[locking.LEVEL_NODE] = []
5293 lu.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5295 self.do_grouplocks = (self.do_locking and
5296 query.IQ_NODES in self.requested_data)
5298 def DeclareLocks(self, lu, level):
5300 if level == locking.LEVEL_NODEGROUP and self.do_grouplocks:
5301 assert not lu.needed_locks[locking.LEVEL_NODEGROUP]
5303 # Lock all groups used by instances optimistically; this requires going
5304 # via the node before it's locked, requiring verification later on
5305 lu.needed_locks[locking.LEVEL_NODEGROUP] = \
5307 for instance_name in lu.owned_locks(locking.LEVEL_INSTANCE)
5308 for group_uuid in lu.cfg.GetInstanceNodeGroups(instance_name))
5309 elif level == locking.LEVEL_NODE:
5310 lu._LockInstancesNodes() # pylint: disable=W0212
5313 def _CheckGroupLocks(lu):
5314 owned_instances = frozenset(lu.owned_locks(locking.LEVEL_INSTANCE))
5315 owned_groups = frozenset(lu.owned_locks(locking.LEVEL_NODEGROUP))
5317 # Check if node groups for locked instances are still correct
5318 for instance_name in owned_instances:
5319 _CheckInstanceNodeGroups(lu.cfg, instance_name, owned_groups)
5321 def _GetQueryData(self, lu):
5322 """Computes the list of instances and their attributes.
5325 if self.do_grouplocks:
5326 self._CheckGroupLocks(lu)
5328 cluster = lu.cfg.GetClusterInfo()
5329 all_info = lu.cfg.GetAllInstancesInfo()
5331 instance_names = self._GetNames(lu, all_info.keys(), locking.LEVEL_INSTANCE)
5333 instance_list = [all_info[name] for name in instance_names]
5334 nodes = frozenset(itertools.chain(*(inst.all_nodes
5335 for inst in instance_list)))
5336 hv_list = list(set([inst.hypervisor for inst in instance_list]))
5339 wrongnode_inst = set()
5341 # Gather data as requested
5342 if self.requested_data & set([query.IQ_LIVE, query.IQ_CONSOLE]):
5344 node_data = lu.rpc.call_all_instances_info(nodes, hv_list)
5346 result = node_data[name]
5348 # offline nodes will be in both lists
5349 assert result.fail_msg
5350 offline_nodes.append(name)
5352 bad_nodes.append(name)
5353 elif result.payload:
5354 for inst in result.payload:
5355 if inst in all_info:
5356 if all_info[inst].primary_node == name:
5357 live_data.update(result.payload)
5359 wrongnode_inst.add(inst)
5361 # orphan instance; we don't list it here as we don't
5362 # handle this case yet in the output of instance listing
5363 logging.warning("Orphan instance '%s' found on node %s",
5365 # else no instance is alive
5369 if query.IQ_DISKUSAGE in self.requested_data:
5370 disk_usage = dict((inst.name,
5371 _ComputeDiskSize(inst.disk_template,
5372 [{constants.IDISK_SIZE: disk.size}
5373 for disk in inst.disks]))
5374 for inst in instance_list)
5378 if query.IQ_CONSOLE in self.requested_data:
5380 for inst in instance_list:
5381 if inst.name in live_data:
5382 # Instance is running
5383 consinfo[inst.name] = _GetInstanceConsole(cluster, inst)
5385 consinfo[inst.name] = None
5386 assert set(consinfo.keys()) == set(instance_names)
5390 if query.IQ_NODES in self.requested_data:
5391 node_names = set(itertools.chain(*map(operator.attrgetter("all_nodes"),
5393 nodes = dict(lu.cfg.GetMultiNodeInfo(node_names))
5394 groups = dict((uuid, lu.cfg.GetNodeGroup(uuid))
5395 for uuid in set(map(operator.attrgetter("group"),
5401 return query.InstanceQueryData(instance_list, lu.cfg.GetClusterInfo(),
5402 disk_usage, offline_nodes, bad_nodes,
5403 live_data, wrongnode_inst, consinfo,
5407 class LUQuery(NoHooksLU):
5408 """Query for resources/items of a certain kind.
5411 # pylint: disable=W0142
5414 def CheckArguments(self):
5415 qcls = _GetQueryImplementation(self.op.what)
5417 self.impl = qcls(self.op.qfilter, self.op.fields, self.op.use_locking)
5419 def ExpandNames(self):
5420 self.impl.ExpandNames(self)
5422 def DeclareLocks(self, level):
5423 self.impl.DeclareLocks(self, level)
5425 def Exec(self, feedback_fn):
5426 return self.impl.NewStyleQuery(self)
5429 class LUQueryFields(NoHooksLU):
5430 """Query for resources/items of a certain kind.
5433 # pylint: disable=W0142
5436 def CheckArguments(self):
5437 self.qcls = _GetQueryImplementation(self.op.what)
5439 def ExpandNames(self):
5440 self.needed_locks = {}
5442 def Exec(self, feedback_fn):
5443 return query.QueryFields(self.qcls.FIELDS, self.op.fields)
5446 class LUNodeModifyStorage(NoHooksLU):
5447 """Logical unit for modifying a storage volume on a node.
5452 def CheckArguments(self):
5453 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
5455 storage_type = self.op.storage_type
5458 modifiable = constants.MODIFIABLE_STORAGE_FIELDS[storage_type]
5460 raise errors.OpPrereqError("Storage units of type '%s' can not be"
5461 " modified" % storage_type,
5464 diff = set(self.op.changes.keys()) - modifiable
5466 raise errors.OpPrereqError("The following fields can not be modified for"
5467 " storage units of type '%s': %r" %
5468 (storage_type, list(diff)),
5471 def ExpandNames(self):
5472 self.needed_locks = {
5473 locking.LEVEL_NODE: self.op.node_name,
5476 def Exec(self, feedback_fn):
5477 """Computes the list of nodes and their attributes.
5480 st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
5481 result = self.rpc.call_storage_modify(self.op.node_name,
5482 self.op.storage_type, st_args,
5483 self.op.name, self.op.changes)
5484 result.Raise("Failed to modify storage unit '%s' on %s" %
5485 (self.op.name, self.op.node_name))
5488 class LUNodeAdd(LogicalUnit):
5489 """Logical unit for adding node to the cluster.
5493 HTYPE = constants.HTYPE_NODE
5494 _NFLAGS = ["master_capable", "vm_capable"]
5496 def CheckArguments(self):
5497 self.primary_ip_family = self.cfg.GetPrimaryIPFamily()
5498 # validate/normalize the node name
5499 self.hostname = netutils.GetHostname(name=self.op.node_name,
5500 family=self.primary_ip_family)
5501 self.op.node_name = self.hostname.name
5503 if self.op.readd and self.op.node_name == self.cfg.GetMasterNode():
5504 raise errors.OpPrereqError("Cannot readd the master node",
5507 if self.op.readd and self.op.group:
5508 raise errors.OpPrereqError("Cannot pass a node group when a node is"
5509 " being readded", errors.ECODE_INVAL)
5511 def BuildHooksEnv(self):
5514 This will run on all nodes before, and on all nodes + the new node after.
5518 "OP_TARGET": self.op.node_name,
5519 "NODE_NAME": self.op.node_name,
5520 "NODE_PIP": self.op.primary_ip,
5521 "NODE_SIP": self.op.secondary_ip,
5522 "MASTER_CAPABLE": str(self.op.master_capable),
5523 "VM_CAPABLE": str(self.op.vm_capable),
5526 def BuildHooksNodes(self):
5527 """Build hooks nodes.
5530 # Exclude added node
5531 pre_nodes = list(set(self.cfg.GetNodeList()) - set([self.op.node_name]))
5532 post_nodes = pre_nodes + [self.op.node_name, ]
5534 return (pre_nodes, post_nodes)
5536 def CheckPrereq(self):
5537 """Check prerequisites.
5540 - the new node is not already in the config
5542 - its parameters (single/dual homed) matches the cluster
5544 Any errors are signaled by raising errors.OpPrereqError.
5548 hostname = self.hostname
5549 node = hostname.name
5550 primary_ip = self.op.primary_ip = hostname.ip
5551 if self.op.secondary_ip is None:
5552 if self.primary_ip_family == netutils.IP6Address.family:
5553 raise errors.OpPrereqError("When using a IPv6 primary address, a valid"
5554 " IPv4 address must be given as secondary",
5556 self.op.secondary_ip = primary_ip
5558 secondary_ip = self.op.secondary_ip
5559 if not netutils.IP4Address.IsValid(secondary_ip):
5560 raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
5561 " address" % secondary_ip, errors.ECODE_INVAL)
5563 node_list = cfg.GetNodeList()
5564 if not self.op.readd and node in node_list:
5565 raise errors.OpPrereqError("Node %s is already in the configuration" %
5566 node, errors.ECODE_EXISTS)
5567 elif self.op.readd and node not in node_list:
5568 raise errors.OpPrereqError("Node %s is not in the configuration" % node,
5571 self.changed_primary_ip = False
5573 for existing_node_name, existing_node in cfg.GetMultiNodeInfo(node_list):
5574 if self.op.readd and node == existing_node_name:
5575 if existing_node.secondary_ip != secondary_ip:
5576 raise errors.OpPrereqError("Readded node doesn't have the same IP"
5577 " address configuration as before",
5579 if existing_node.primary_ip != primary_ip:
5580 self.changed_primary_ip = True
5584 if (existing_node.primary_ip == primary_ip or
5585 existing_node.secondary_ip == primary_ip or
5586 existing_node.primary_ip == secondary_ip or
5587 existing_node.secondary_ip == secondary_ip):
5588 raise errors.OpPrereqError("New node ip address(es) conflict with"
5589 " existing node %s" % existing_node.name,
5590 errors.ECODE_NOTUNIQUE)
5592 # After this 'if' block, None is no longer a valid value for the
5593 # _capable op attributes
5595 old_node = self.cfg.GetNodeInfo(node)
5596 assert old_node is not None, "Can't retrieve locked node %s" % node
5597 for attr in self._NFLAGS:
5598 if getattr(self.op, attr) is None:
5599 setattr(self.op, attr, getattr(old_node, attr))
5601 for attr in self._NFLAGS:
5602 if getattr(self.op, attr) is None:
5603 setattr(self.op, attr, True)
5605 if self.op.readd and not self.op.vm_capable:
5606 pri, sec = cfg.GetNodeInstances(node)
5608 raise errors.OpPrereqError("Node %s being re-added with vm_capable"
5609 " flag set to false, but it already holds"
5610 " instances" % node,
5613 # check that the type of the node (single versus dual homed) is the
5614 # same as for the master
5615 myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
5616 master_singlehomed = myself.secondary_ip == myself.primary_ip
5617 newbie_singlehomed = secondary_ip == primary_ip
5618 if master_singlehomed != newbie_singlehomed:
5619 if master_singlehomed:
5620 raise errors.OpPrereqError("The master has no secondary ip but the"
5621 " new node has one",
5624 raise errors.OpPrereqError("The master has a secondary ip but the"
5625 " new node doesn't have one",
5628 # checks reachability
5629 if not netutils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
5630 raise errors.OpPrereqError("Node not reachable by ping",
5631 errors.ECODE_ENVIRON)
5633 if not newbie_singlehomed:
5634 # check reachability from my secondary ip to newbie's secondary ip
5635 if not netutils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
5636 source=myself.secondary_ip):
5637 raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
5638 " based ping to node daemon port",
5639 errors.ECODE_ENVIRON)
5646 if self.op.master_capable:
5647 self.master_candidate = _DecideSelfPromotion(self, exceptions=exceptions)
5649 self.master_candidate = False
5652 self.new_node = old_node
5654 node_group = cfg.LookupNodeGroup(self.op.group)
5655 self.new_node = objects.Node(name=node,
5656 primary_ip=primary_ip,
5657 secondary_ip=secondary_ip,
5658 master_candidate=self.master_candidate,
5659 offline=False, drained=False,
5662 if self.op.ndparams:
5663 utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
5665 if self.op.hv_state:
5666 self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state, None)
5668 if self.op.disk_state:
5669 self.new_disk_state = _MergeAndVerifyDiskState(self.op.disk_state, None)
5671 # TODO: If we need to have multiple DnsOnlyRunner we probably should make
5672 # it a property on the base class.
5673 result = rpc.DnsOnlyRunner().call_version([node])[node]
5674 result.Raise("Can't get version information from node %s" % node)
5675 if constants.PROTOCOL_VERSION == result.payload:
5676 logging.info("Communication to node %s fine, sw version %s match",
5677 node, result.payload)
5679 raise errors.OpPrereqError("Version mismatch master version %s,"
5680 " node version %s" %
5681 (constants.PROTOCOL_VERSION, result.payload),
5682 errors.ECODE_ENVIRON)
5684 def Exec(self, feedback_fn):
5685 """Adds the new node to the cluster.
5688 new_node = self.new_node
5689 node = new_node.name
5691 assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER), \
5694 # We adding a new node so we assume it's powered
5695 new_node.powered = True
5697 # for re-adds, reset the offline/drained/master-candidate flags;
5698 # we need to reset here, otherwise offline would prevent RPC calls
5699 # later in the procedure; this also means that if the re-add
5700 # fails, we are left with a non-offlined, broken node
5702 new_node.drained = new_node.offline = False # pylint: disable=W0201
5703 self.LogInfo("Readding a node, the offline/drained flags were reset")
5704 # if we demote the node, we do cleanup later in the procedure
5705 new_node.master_candidate = self.master_candidate
5706 if self.changed_primary_ip:
5707 new_node.primary_ip = self.op.primary_ip
5709 # copy the master/vm_capable flags
5710 for attr in self._NFLAGS:
5711 setattr(new_node, attr, getattr(self.op, attr))
5713 # notify the user about any possible mc promotion
5714 if new_node.master_candidate:
5715 self.LogInfo("Node will be a master candidate")
5717 if self.op.ndparams:
5718 new_node.ndparams = self.op.ndparams
5720 new_node.ndparams = {}
5722 if self.op.hv_state:
5723 new_node.hv_state_static = self.new_hv_state
5725 if self.op.disk_state:
5726 new_node.disk_state_static = self.new_disk_state
5728 # Add node to our /etc/hosts, and add key to known_hosts
5729 if self.cfg.GetClusterInfo().modify_etc_hosts:
5730 master_node = self.cfg.GetMasterNode()
5731 result = self.rpc.call_etc_hosts_modify(master_node,
5732 constants.ETC_HOSTS_ADD,
5735 result.Raise("Can't update hosts file with new host data")
5737 if new_node.secondary_ip != new_node.primary_ip:
5738 _CheckNodeHasSecondaryIP(self, new_node.name, new_node.secondary_ip,
5741 node_verify_list = [self.cfg.GetMasterNode()]
5742 node_verify_param = {
5743 constants.NV_NODELIST: ([node], {}),
5744 # TODO: do a node-net-test as well?
5747 result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
5748 self.cfg.GetClusterName())
5749 for verifier in node_verify_list:
5750 result[verifier].Raise("Cannot communicate with node %s" % verifier)
5751 nl_payload = result[verifier].payload[constants.NV_NODELIST]
5753 for failed in nl_payload:
5754 feedback_fn("ssh/hostname verification failed"
5755 " (checking from %s): %s" %
5756 (verifier, nl_payload[failed]))
5757 raise errors.OpExecError("ssh/hostname verification failed")
5760 _RedistributeAncillaryFiles(self)
5761 self.context.ReaddNode(new_node)
5762 # make sure we redistribute the config
5763 self.cfg.Update(new_node, feedback_fn)
5764 # and make sure the new node will not have old files around
5765 if not new_node.master_candidate:
5766 result = self.rpc.call_node_demote_from_mc(new_node.name)
5767 msg = result.fail_msg
5769 self.LogWarning("Node failed to demote itself from master"
5770 " candidate status: %s" % msg)
5772 _RedistributeAncillaryFiles(self, additional_nodes=[node],
5773 additional_vm=self.op.vm_capable)
5774 self.context.AddNode(new_node, self.proc.GetECId())
5777 class LUNodeSetParams(LogicalUnit):
5778 """Modifies the parameters of a node.
5780 @cvar _F2R: a dictionary from tuples of flags (mc, drained, offline)
5781 to the node role (as _ROLE_*)
5782 @cvar _R2F: a dictionary from node role to tuples of flags
5783 @cvar _FLAGS: a list of attribute names corresponding to the flags
5786 HPATH = "node-modify"
5787 HTYPE = constants.HTYPE_NODE
5789 (_ROLE_CANDIDATE, _ROLE_DRAINED, _ROLE_OFFLINE, _ROLE_REGULAR) = range(4)
5791 (True, False, False): _ROLE_CANDIDATE,
5792 (False, True, False): _ROLE_DRAINED,
5793 (False, False, True): _ROLE_OFFLINE,
5794 (False, False, False): _ROLE_REGULAR,
5796 _R2F = dict((v, k) for k, v in _F2R.items())
5797 _FLAGS = ["master_candidate", "drained", "offline"]
5799 def CheckArguments(self):
5800 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
5801 all_mods = [self.op.offline, self.op.master_candidate, self.op.drained,
5802 self.op.master_capable, self.op.vm_capable,
5803 self.op.secondary_ip, self.op.ndparams, self.op.hv_state,
5805 if all_mods.count(None) == len(all_mods):
5806 raise errors.OpPrereqError("Please pass at least one modification",
5808 if all_mods.count(True) > 1:
5809 raise errors.OpPrereqError("Can't set the node into more than one"
5810 " state at the same time",
5813 # Boolean value that tells us whether we might be demoting from MC
5814 self.might_demote = (self.op.master_candidate == False or
5815 self.op.offline == True or
5816 self.op.drained == True or
5817 self.op.master_capable == False)
5819 if self.op.secondary_ip:
5820 if not netutils.IP4Address.IsValid(self.op.secondary_ip):
5821 raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
5822 " address" % self.op.secondary_ip,
5825 self.lock_all = self.op.auto_promote and self.might_demote
5826 self.lock_instances = self.op.secondary_ip is not None
5828 def _InstanceFilter(self, instance):
5829 """Filter for getting affected instances.
5832 return (instance.disk_template in constants.DTS_INT_MIRROR and
5833 self.op.node_name in instance.all_nodes)
5835 def ExpandNames(self):
5837 self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET}
5839 self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
5841 # Since modifying a node can have severe effects on currently running
5842 # operations the resource lock is at least acquired in shared mode
5843 self.needed_locks[locking.LEVEL_NODE_RES] = \
5844 self.needed_locks[locking.LEVEL_NODE]
5846 # Get node resource and instance locks in shared mode; they are not used
5847 # for anything but read-only access
5848 self.share_locks[locking.LEVEL_NODE_RES] = 1
5849 self.share_locks[locking.LEVEL_INSTANCE] = 1
5851 if self.lock_instances:
5852 self.needed_locks[locking.LEVEL_INSTANCE] = \
5853 frozenset(self.cfg.GetInstancesInfoByFilter(self._InstanceFilter))
5855 def BuildHooksEnv(self):
5858 This runs on the master node.
5862 "OP_TARGET": self.op.node_name,
5863 "MASTER_CANDIDATE": str(self.op.master_candidate),
5864 "OFFLINE": str(self.op.offline),
5865 "DRAINED": str(self.op.drained),
5866 "MASTER_CAPABLE": str(self.op.master_capable),
5867 "VM_CAPABLE": str(self.op.vm_capable),
5870 def BuildHooksNodes(self):
5871 """Build hooks nodes.
5874 nl = [self.cfg.GetMasterNode(), self.op.node_name]
5877 def CheckPrereq(self):
5878 """Check prerequisites.
5880 This only checks the instance list against the existing names.
5883 node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
5885 if self.lock_instances:
5886 affected_instances = \
5887 self.cfg.GetInstancesInfoByFilter(self._InstanceFilter)
5889 # Verify instance locks
5890 owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
5891 wanted_instances = frozenset(affected_instances.keys())
5892 if wanted_instances - owned_instances:
5893 raise errors.OpPrereqError("Instances affected by changing node %s's"
5894 " secondary IP address have changed since"
5895 " locks were acquired, wanted '%s', have"
5896 " '%s'; retry the operation" %
5898 utils.CommaJoin(wanted_instances),
5899 utils.CommaJoin(owned_instances)),
5902 affected_instances = None
5904 if (self.op.master_candidate is not None or
5905 self.op.drained is not None or
5906 self.op.offline is not None):
5907 # we can't change the master's node flags
5908 if self.op.node_name == self.cfg.GetMasterNode():
5909 raise errors.OpPrereqError("The master role can be changed"
5910 " only via master-failover",
5913 if self.op.master_candidate and not node.master_capable:
5914 raise errors.OpPrereqError("Node %s is not master capable, cannot make"
5915 " it a master candidate" % node.name,
5918 if self.op.vm_capable == False:
5919 (ipri, isec) = self.cfg.GetNodeInstances(self.op.node_name)
5921 raise errors.OpPrereqError("Node %s hosts instances, cannot unset"
5922 " the vm_capable flag" % node.name,
5925 if node.master_candidate and self.might_demote and not self.lock_all:
5926 assert not self.op.auto_promote, "auto_promote set but lock_all not"
5927 # check if after removing the current node, we're missing master
5929 (mc_remaining, mc_should, _) = \
5930 self.cfg.GetMasterCandidateStats(exceptions=[node.name])
5931 if mc_remaining < mc_should:
5932 raise errors.OpPrereqError("Not enough master candidates, please"
5933 " pass auto promote option to allow"
5934 " promotion (--auto-promote or RAPI"
5935 " auto_promote=True)", errors.ECODE_STATE)
5937 self.old_flags = old_flags = (node.master_candidate,
5938 node.drained, node.offline)
5939 assert old_flags in self._F2R, "Un-handled old flags %s" % str(old_flags)
5940 self.old_role = old_role = self._F2R[old_flags]
5942 # Check for ineffective changes
5943 for attr in self._FLAGS:
5944 if (getattr(self.op, attr) == False and getattr(node, attr) == False):
5945 self.LogInfo("Ignoring request to unset flag %s, already unset", attr)
5946 setattr(self.op, attr, None)
5948 # Past this point, any flag change to False means a transition
5949 # away from the respective state, as only real changes are kept
5951 # TODO: We might query the real power state if it supports OOB
5952 if _SupportsOob(self.cfg, node):
5953 if self.op.offline is False and not (node.powered or
5954 self.op.powered == True):
5955 raise errors.OpPrereqError(("Node %s needs to be turned on before its"
5956 " offline status can be reset") %
5958 elif self.op.powered is not None:
5959 raise errors.OpPrereqError(("Unable to change powered state for node %s"
5960 " as it does not support out-of-band"
5961 " handling") % self.op.node_name)
5963 # If we're being deofflined/drained, we'll MC ourself if needed
5964 if (self.op.drained == False or self.op.offline == False or
5965 (self.op.master_capable and not node.master_capable)):
5966 if _DecideSelfPromotion(self):
5967 self.op.master_candidate = True
5968 self.LogInfo("Auto-promoting node to master candidate")
5970 # If we're no longer master capable, we'll demote ourselves from MC
5971 if self.op.master_capable == False and node.master_candidate:
5972 self.LogInfo("Demoting from master candidate")
5973 self.op.master_candidate = False
5976 assert [getattr(self.op, attr) for attr in self._FLAGS].count(True) <= 1
5977 if self.op.master_candidate:
5978 new_role = self._ROLE_CANDIDATE
5979 elif self.op.drained:
5980 new_role = self._ROLE_DRAINED
5981 elif self.op.offline:
5982 new_role = self._ROLE_OFFLINE
5983 elif False in [self.op.master_candidate, self.op.drained, self.op.offline]:
5984 # False is still in new flags, which means we're un-setting (the
5986 new_role = self._ROLE_REGULAR
5987 else: # no new flags, nothing, keep old role
5990 self.new_role = new_role
5992 if old_role == self._ROLE_OFFLINE and new_role != old_role:
5993 # Trying to transition out of offline status
5994 result = self.rpc.call_version([node.name])[node.name]
5996 raise errors.OpPrereqError("Node %s is being de-offlined but fails"
5997 " to report its version: %s" %
5998 (node.name, result.fail_msg),
6001 self.LogWarning("Transitioning node from offline to online state"
6002 " without using re-add. Please make sure the node"
6005 if self.op.secondary_ip:
6006 # Ok even without locking, because this can't be changed by any LU
6007 master = self.cfg.GetNodeInfo(self.cfg.GetMasterNode())
6008 master_singlehomed = master.secondary_ip == master.primary_ip
6009 if master_singlehomed and self.op.secondary_ip:
6010 raise errors.OpPrereqError("Cannot change the secondary ip on a single"
6011 " homed cluster", errors.ECODE_INVAL)
6013 assert not (frozenset(affected_instances) -
6014 self.owned_locks(locking.LEVEL_INSTANCE))
6017 if affected_instances:
6018 raise errors.OpPrereqError("Cannot change secondary IP address:"
6019 " offline node has instances (%s)"
6020 " configured to use it" %
6021 utils.CommaJoin(affected_instances.keys()))
6023 # On online nodes, check that no instances are running, and that
6024 # the node has the new ip and we can reach it.
6025 for instance in affected_instances.values():
6026 _CheckInstanceState(self, instance, INSTANCE_DOWN,
6027 msg="cannot change secondary ip")
6029 _CheckNodeHasSecondaryIP(self, node.name, self.op.secondary_ip, True)
6030 if master.name != node.name:
6031 # check reachability from master secondary ip to new secondary ip
6032 if not netutils.TcpPing(self.op.secondary_ip,
6033 constants.DEFAULT_NODED_PORT,
6034 source=master.secondary_ip):
6035 raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
6036 " based ping to node daemon port",
6037 errors.ECODE_ENVIRON)
6039 if self.op.ndparams:
6040 new_ndparams = _GetUpdatedParams(self.node.ndparams, self.op.ndparams)
6041 utils.ForceDictType(new_ndparams, constants.NDS_PARAMETER_TYPES)
6042 self.new_ndparams = new_ndparams
6044 if self.op.hv_state:
6045 self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state,
6046 self.node.hv_state_static)
6048 if self.op.disk_state:
6049 self.new_disk_state = \
6050 _MergeAndVerifyDiskState(self.op.disk_state,
6051 self.node.disk_state_static)
6053 def Exec(self, feedback_fn):
6058 old_role = self.old_role
6059 new_role = self.new_role
6063 if self.op.ndparams:
6064 node.ndparams = self.new_ndparams
6066 if self.op.powered is not None:
6067 node.powered = self.op.powered
6069 if self.op.hv_state:
6070 node.hv_state_static = self.new_hv_state
6072 if self.op.disk_state:
6073 node.disk_state_static = self.new_disk_state
6075 for attr in ["master_capable", "vm_capable"]:
6076 val = getattr(self.op, attr)
6078 setattr(node, attr, val)
6079 result.append((attr, str(val)))
6081 if new_role != old_role:
6082 # Tell the node to demote itself, if no longer MC and not offline
6083 if old_role == self._ROLE_CANDIDATE and new_role != self._ROLE_OFFLINE:
6084 msg = self.rpc.call_node_demote_from_mc(node.name).fail_msg
6086 self.LogWarning("Node failed to demote itself: %s", msg)
6088 new_flags = self._R2F[new_role]
6089 for of, nf, desc in zip(self.old_flags, new_flags, self._FLAGS):
6091 result.append((desc, str(nf)))
6092 (node.master_candidate, node.drained, node.offline) = new_flags
6094 # we locked all nodes, we adjust the CP before updating this node
6096 _AdjustCandidatePool(self, [node.name])
6098 if self.op.secondary_ip:
6099 node.secondary_ip = self.op.secondary_ip
6100 result.append(("secondary_ip", self.op.secondary_ip))
6102 # this will trigger configuration file update, if needed
6103 self.cfg.Update(node, feedback_fn)
6105 # this will trigger job queue propagation or cleanup if the mc
6107 if [old_role, new_role].count(self._ROLE_CANDIDATE) == 1:
6108 self.context.ReaddNode(node)
6113 class LUNodePowercycle(NoHooksLU):
6114 """Powercycles a node.
6119 def CheckArguments(self):
6120 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
6121 if self.op.node_name == self.cfg.GetMasterNode() and not self.op.force:
6122 raise errors.OpPrereqError("The node is the master and the force"
6123 " parameter was not set",
6126 def ExpandNames(self):
6127 """Locking for PowercycleNode.
6129 This is a last-resort option and shouldn't block on other
6130 jobs. Therefore, we grab no locks.
6133 self.needed_locks = {}
6135 def Exec(self, feedback_fn):
6139 result = self.rpc.call_node_powercycle(self.op.node_name,
6140 self.cfg.GetHypervisorType())
6141 result.Raise("Failed to schedule the reboot")
6142 return result.payload
6145 class LUClusterQuery(NoHooksLU):
6146 """Query cluster configuration.
6151 def ExpandNames(self):
6152 self.needed_locks = {}
6154 def Exec(self, feedback_fn):
6155 """Return cluster config.
6158 cluster = self.cfg.GetClusterInfo()
6161 # Filter just for enabled hypervisors
6162 for os_name, hv_dict in cluster.os_hvp.items():
6163 os_hvp[os_name] = {}
6164 for hv_name, hv_params in hv_dict.items():
6165 if hv_name in cluster.enabled_hypervisors:
6166 os_hvp[os_name][hv_name] = hv_params
6168 # Convert ip_family to ip_version
6169 primary_ip_version = constants.IP4_VERSION
6170 if cluster.primary_ip_family == netutils.IP6Address.family:
6171 primary_ip_version = constants.IP6_VERSION
6174 "software_version": constants.RELEASE_VERSION,
6175 "protocol_version": constants.PROTOCOL_VERSION,
6176 "config_version": constants.CONFIG_VERSION,
6177 "os_api_version": max(constants.OS_API_VERSIONS),
6178 "export_version": constants.EXPORT_VERSION,
6179 "architecture": runtime.GetArchInfo(),
6180 "name": cluster.cluster_name,
6181 "master": cluster.master_node,
6182 "default_hypervisor": cluster.primary_hypervisor,
6183 "enabled_hypervisors": cluster.enabled_hypervisors,
6184 "hvparams": dict([(hypervisor_name, cluster.hvparams[hypervisor_name])
6185 for hypervisor_name in cluster.enabled_hypervisors]),
6187 "beparams": cluster.beparams,
6188 "osparams": cluster.osparams,
6189 "ipolicy": cluster.ipolicy,
6190 "nicparams": cluster.nicparams,
6191 "ndparams": cluster.ndparams,
6192 "diskparams": cluster.diskparams,
6193 "candidate_pool_size": cluster.candidate_pool_size,
6194 "master_netdev": cluster.master_netdev,
6195 "master_netmask": cluster.master_netmask,
6196 "use_external_mip_script": cluster.use_external_mip_script,
6197 "volume_group_name": cluster.volume_group_name,
6198 "drbd_usermode_helper": cluster.drbd_usermode_helper,
6199 "file_storage_dir": cluster.file_storage_dir,
6200 "shared_file_storage_dir": cluster.shared_file_storage_dir,
6201 "maintain_node_health": cluster.maintain_node_health,
6202 "ctime": cluster.ctime,
6203 "mtime": cluster.mtime,
6204 "uuid": cluster.uuid,
6205 "tags": list(cluster.GetTags()),
6206 "uid_pool": cluster.uid_pool,
6207 "default_iallocator": cluster.default_iallocator,
6208 "reserved_lvs": cluster.reserved_lvs,
6209 "primary_ip_version": primary_ip_version,
6210 "prealloc_wipe_disks": cluster.prealloc_wipe_disks,
6211 "hidden_os": cluster.hidden_os,
6212 "blacklisted_os": cluster.blacklisted_os,
6218 class LUClusterConfigQuery(NoHooksLU):
6219 """Return configuration values.
6224 def CheckArguments(self):
6225 self.cq = _ClusterQuery(None, self.op.output_fields, False)
6227 def ExpandNames(self):
6228 self.cq.ExpandNames(self)
6230 def DeclareLocks(self, level):
6231 self.cq.DeclareLocks(self, level)
6233 def Exec(self, feedback_fn):
6234 result = self.cq.OldStyleQuery(self)
6236 assert len(result) == 1
6241 class _ClusterQuery(_QueryBase):
6242 FIELDS = query.CLUSTER_FIELDS
6244 #: Do not sort (there is only one item)
6247 def ExpandNames(self, lu):
6248 lu.needed_locks = {}
6250 # The following variables interact with _QueryBase._GetNames
6251 self.wanted = locking.ALL_SET
6252 self.do_locking = self.use_locking
6255 raise errors.OpPrereqError("Can not use locking for cluster queries",
6258 def DeclareLocks(self, lu, level):
6261 def _GetQueryData(self, lu):
6262 """Computes the list of nodes and their attributes.
6265 # Locking is not used
6266 assert not (compat.any(lu.glm.is_owned(level)
6267 for level in locking.LEVELS
6268 if level != locking.LEVEL_CLUSTER) or
6269 self.do_locking or self.use_locking)
6271 if query.CQ_CONFIG in self.requested_data:
6272 cluster = lu.cfg.GetClusterInfo()
6274 cluster = NotImplemented
6276 if query.CQ_QUEUE_DRAINED in self.requested_data:
6277 drain_flag = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
6279 drain_flag = NotImplemented
6281 if query.CQ_WATCHER_PAUSE in self.requested_data:
6282 watcher_pause = utils.ReadWatcherPauseFile(constants.WATCHER_PAUSEFILE)
6284 watcher_pause = NotImplemented
6286 return query.ClusterQueryData(cluster, drain_flag, watcher_pause)
6289 class LUInstanceActivateDisks(NoHooksLU):
6290 """Bring up an instance's disks.
6295 def ExpandNames(self):
6296 self._ExpandAndLockInstance()
6297 self.needed_locks[locking.LEVEL_NODE] = []
6298 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6300 def DeclareLocks(self, level):
6301 if level == locking.LEVEL_NODE:
6302 self._LockInstancesNodes()
6304 def CheckPrereq(self):
6305 """Check prerequisites.
6307 This checks that the instance is in the cluster.
6310 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6311 assert self.instance is not None, \
6312 "Cannot retrieve locked instance %s" % self.op.instance_name
6313 _CheckNodeOnline(self, self.instance.primary_node)
6315 def Exec(self, feedback_fn):
6316 """Activate the disks.
6319 disks_ok, disks_info = \
6320 _AssembleInstanceDisks(self, self.instance,
6321 ignore_size=self.op.ignore_size)
6323 raise errors.OpExecError("Cannot activate block devices")
6328 def _AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False,
6330 """Prepare the block devices for an instance.
6332 This sets up the block devices on all nodes.
6334 @type lu: L{LogicalUnit}
6335 @param lu: the logical unit on whose behalf we execute
6336 @type instance: L{objects.Instance}
6337 @param instance: the instance for whose disks we assemble
6338 @type disks: list of L{objects.Disk} or None
6339 @param disks: which disks to assemble (or all, if None)
6340 @type ignore_secondaries: boolean
6341 @param ignore_secondaries: if true, errors on secondary nodes
6342 won't result in an error return from the function
6343 @type ignore_size: boolean
6344 @param ignore_size: if true, the current known size of the disk
6345 will not be used during the disk activation, useful for cases
6346 when the size is wrong
6347 @return: False if the operation failed, otherwise a list of
6348 (host, instance_visible_name, node_visible_name)
6349 with the mapping from node devices to instance devices
6354 iname = instance.name
6355 disks = _ExpandCheckDisks(instance, disks)
6357 # With the two passes mechanism we try to reduce the window of
6358 # opportunity for the race condition of switching DRBD to primary
6359 # before handshaking occured, but we do not eliminate it
6361 # The proper fix would be to wait (with some limits) until the
6362 # connection has been made and drbd transitions from WFConnection
6363 # into any other network-connected state (Connected, SyncTarget,
6366 # 1st pass, assemble on all nodes in secondary mode
6367 for idx, inst_disk in enumerate(disks):
6368 for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
6370 node_disk = node_disk.Copy()
6371 node_disk.UnsetSize()
6372 lu.cfg.SetDiskID(node_disk, node)
6373 result = lu.rpc.call_blockdev_assemble(node, (node_disk, instance), iname,
6375 msg = result.fail_msg
6377 lu.proc.LogWarning("Could not prepare block device %s on node %s"
6378 " (is_primary=False, pass=1): %s",
6379 inst_disk.iv_name, node, msg)
6380 if not ignore_secondaries:
6383 # FIXME: race condition on drbd migration to primary
6385 # 2nd pass, do only the primary node
6386 for idx, inst_disk in enumerate(disks):
6389 for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
6390 if node != instance.primary_node:
6393 node_disk = node_disk.Copy()
6394 node_disk.UnsetSize()
6395 lu.cfg.SetDiskID(node_disk, node)
6396 result = lu.rpc.call_blockdev_assemble(node, (node_disk, instance), iname,
6398 msg = result.fail_msg
6400 lu.proc.LogWarning("Could not prepare block device %s on node %s"
6401 " (is_primary=True, pass=2): %s",
6402 inst_disk.iv_name, node, msg)
6405 dev_path = result.payload
6407 device_info.append((instance.primary_node, inst_disk.iv_name, dev_path))
6409 # leave the disks configured for the primary node
6410 # this is a workaround that would be fixed better by
6411 # improving the logical/physical id handling
6413 lu.cfg.SetDiskID(disk, instance.primary_node)
6415 return disks_ok, device_info
6418 def _StartInstanceDisks(lu, instance, force):
6419 """Start the disks of an instance.
6422 disks_ok, _ = _AssembleInstanceDisks(lu, instance,
6423 ignore_secondaries=force)
6425 _ShutdownInstanceDisks(lu, instance)
6426 if force is not None and not force:
6427 lu.proc.LogWarning("", hint="If the message above refers to a"
6429 " you can retry the operation using '--force'.")
6430 raise errors.OpExecError("Disk consistency error")
6433 class LUInstanceDeactivateDisks(NoHooksLU):
6434 """Shutdown an instance's disks.
6439 def ExpandNames(self):
6440 self._ExpandAndLockInstance()
6441 self.needed_locks[locking.LEVEL_NODE] = []
6442 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6444 def DeclareLocks(self, level):
6445 if level == locking.LEVEL_NODE:
6446 self._LockInstancesNodes()
6448 def CheckPrereq(self):
6449 """Check prerequisites.
6451 This checks that the instance is in the cluster.
6454 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6455 assert self.instance is not None, \
6456 "Cannot retrieve locked instance %s" % self.op.instance_name
6458 def Exec(self, feedback_fn):
6459 """Deactivate the disks
6462 instance = self.instance
6464 _ShutdownInstanceDisks(self, instance)
6466 _SafeShutdownInstanceDisks(self, instance)
6469 def _SafeShutdownInstanceDisks(lu, instance, disks=None):
6470 """Shutdown block devices of an instance.
6472 This function checks if an instance is running, before calling
6473 _ShutdownInstanceDisks.
6476 _CheckInstanceState(lu, instance, INSTANCE_DOWN, msg="cannot shutdown disks")
6477 _ShutdownInstanceDisks(lu, instance, disks=disks)
6480 def _ExpandCheckDisks(instance, disks):
6481 """Return the instance disks selected by the disks list
6483 @type disks: list of L{objects.Disk} or None
6484 @param disks: selected disks
6485 @rtype: list of L{objects.Disk}
6486 @return: selected instance disks to act on
6490 return instance.disks
6492 if not set(disks).issubset(instance.disks):
6493 raise errors.ProgrammerError("Can only act on disks belonging to the"
6498 def _ShutdownInstanceDisks(lu, instance, disks=None, ignore_primary=False):
6499 """Shutdown block devices of an instance.
6501 This does the shutdown on all nodes of the instance.
6503 If the ignore_primary is false, errors on the primary node are
6508 disks = _ExpandCheckDisks(instance, disks)
6511 for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
6512 lu.cfg.SetDiskID(top_disk, node)
6513 result = lu.rpc.call_blockdev_shutdown(node, top_disk)
6514 msg = result.fail_msg
6516 lu.LogWarning("Could not shutdown block device %s on node %s: %s",
6517 disk.iv_name, node, msg)
6518 if ((node == instance.primary_node and not ignore_primary) or
6519 (node != instance.primary_node and not result.offline)):
6524 def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
6525 """Checks if a node has enough free memory.
6527 This function check if a given node has the needed amount of free
6528 memory. In case the node has less memory or we cannot get the
6529 information from the node, this function raise an OpPrereqError
6532 @type lu: C{LogicalUnit}
6533 @param lu: a logical unit from which we get configuration data
6535 @param node: the node to check
6536 @type reason: C{str}
6537 @param reason: string to use in the error message
6538 @type requested: C{int}
6539 @param requested: the amount of memory in MiB to check for
6540 @type hypervisor_name: C{str}
6541 @param hypervisor_name: the hypervisor to ask for memory stats
6543 @return: node current free memory
6544 @raise errors.OpPrereqError: if the node doesn't have enough memory, or
6545 we cannot check the node
6548 nodeinfo = lu.rpc.call_node_info([node], None, [hypervisor_name])
6549 nodeinfo[node].Raise("Can't get data from node %s" % node,
6550 prereq=True, ecode=errors.ECODE_ENVIRON)
6551 (_, _, (hv_info, )) = nodeinfo[node].payload
6553 free_mem = hv_info.get("memory_free", None)
6554 if not isinstance(free_mem, int):
6555 raise errors.OpPrereqError("Can't compute free memory on node %s, result"
6556 " was '%s'" % (node, free_mem),
6557 errors.ECODE_ENVIRON)
6558 if requested > free_mem:
6559 raise errors.OpPrereqError("Not enough memory on node %s for %s:"
6560 " needed %s MiB, available %s MiB" %
6561 (node, reason, requested, free_mem),
6566 def _CheckNodesFreeDiskPerVG(lu, nodenames, req_sizes):
6567 """Checks if nodes have enough free disk space in the all VGs.
6569 This function check if all given nodes have the needed amount of
6570 free disk. In case any node has less disk or we cannot get the
6571 information from the node, this function raise an OpPrereqError
6574 @type lu: C{LogicalUnit}
6575 @param lu: a logical unit from which we get configuration data
6576 @type nodenames: C{list}
6577 @param nodenames: the list of node names to check
6578 @type req_sizes: C{dict}
6579 @param req_sizes: the hash of vg and corresponding amount of disk in
6581 @raise errors.OpPrereqError: if the node doesn't have enough disk,
6582 or we cannot check the node
6585 for vg, req_size in req_sizes.items():
6586 _CheckNodesFreeDiskOnVG(lu, nodenames, vg, req_size)
6589 def _CheckNodesFreeDiskOnVG(lu, nodenames, vg, requested):
6590 """Checks if nodes have enough free disk space in the specified VG.
6592 This function check if all given nodes have the needed amount of
6593 free disk. In case any node has less disk or we cannot get the
6594 information from the node, this function raise an OpPrereqError
6597 @type lu: C{LogicalUnit}
6598 @param lu: a logical unit from which we get configuration data
6599 @type nodenames: C{list}
6600 @param nodenames: the list of node names to check
6602 @param vg: the volume group to check
6603 @type requested: C{int}
6604 @param requested: the amount of disk in MiB to check for
6605 @raise errors.OpPrereqError: if the node doesn't have enough disk,
6606 or we cannot check the node
6609 nodeinfo = lu.rpc.call_node_info(nodenames, [vg], None)
6610 for node in nodenames:
6611 info = nodeinfo[node]
6612 info.Raise("Cannot get current information from node %s" % node,
6613 prereq=True, ecode=errors.ECODE_ENVIRON)
6614 (_, (vg_info, ), _) = info.payload
6615 vg_free = vg_info.get("vg_free", None)
6616 if not isinstance(vg_free, int):
6617 raise errors.OpPrereqError("Can't compute free disk space on node"
6618 " %s for vg %s, result was '%s'" %
6619 (node, vg, vg_free), errors.ECODE_ENVIRON)
6620 if requested > vg_free:
6621 raise errors.OpPrereqError("Not enough disk space on target node %s"
6622 " vg %s: required %d MiB, available %d MiB" %
6623 (node, vg, requested, vg_free),
6627 def _CheckNodesPhysicalCPUs(lu, nodenames, requested, hypervisor_name):
6628 """Checks if nodes have enough physical CPUs
6630 This function checks if all given nodes have the needed number of
6631 physical CPUs. In case any node has less CPUs or we cannot get the
6632 information from the node, this function raises an OpPrereqError
6635 @type lu: C{LogicalUnit}
6636 @param lu: a logical unit from which we get configuration data
6637 @type nodenames: C{list}
6638 @param nodenames: the list of node names to check
6639 @type requested: C{int}
6640 @param requested: the minimum acceptable number of physical CPUs
6641 @raise errors.OpPrereqError: if the node doesn't have enough CPUs,
6642 or we cannot check the node
6645 nodeinfo = lu.rpc.call_node_info(nodenames, None, [hypervisor_name])
6646 for node in nodenames:
6647 info = nodeinfo[node]
6648 info.Raise("Cannot get current information from node %s" % node,
6649 prereq=True, ecode=errors.ECODE_ENVIRON)
6650 (_, _, (hv_info, )) = info.payload
6651 num_cpus = hv_info.get("cpu_total", None)
6652 if not isinstance(num_cpus, int):
6653 raise errors.OpPrereqError("Can't compute the number of physical CPUs"
6654 " on node %s, result was '%s'" %
6655 (node, num_cpus), errors.ECODE_ENVIRON)
6656 if requested > num_cpus:
6657 raise errors.OpPrereqError("Node %s has %s physical CPUs, but %s are "
6658 "required" % (node, num_cpus, requested),
6662 class LUInstanceStartup(LogicalUnit):
6663 """Starts an instance.
6666 HPATH = "instance-start"
6667 HTYPE = constants.HTYPE_INSTANCE
6670 def CheckArguments(self):
6672 if self.op.beparams:
6673 # fill the beparams dict
6674 objects.UpgradeBeParams(self.op.beparams)
6675 utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
6677 def ExpandNames(self):
6678 self._ExpandAndLockInstance()
6679 self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
6681 def DeclareLocks(self, level):
6682 if level == locking.LEVEL_NODE_RES:
6683 self._LockInstancesNodes(primary_only=True, level=locking.LEVEL_NODE_RES)
6685 def BuildHooksEnv(self):
6688 This runs on master, primary and secondary nodes of the instance.
6692 "FORCE": self.op.force,
6695 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6699 def BuildHooksNodes(self):
6700 """Build hooks nodes.
6703 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6706 def CheckPrereq(self):
6707 """Check prerequisites.
6709 This checks that the instance is in the cluster.
6712 self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6713 assert self.instance is not None, \
6714 "Cannot retrieve locked instance %s" % self.op.instance_name
6717 if self.op.hvparams:
6718 # check hypervisor parameter syntax (locally)
6719 cluster = self.cfg.GetClusterInfo()
6720 utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
6721 filled_hvp = cluster.FillHV(instance)
6722 filled_hvp.update(self.op.hvparams)
6723 hv_type = hypervisor.GetHypervisor(instance.hypervisor)
6724 hv_type.CheckParameterSyntax(filled_hvp)
6725 _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
6727 _CheckInstanceState(self, instance, INSTANCE_ONLINE)
6729 self.primary_offline = self.cfg.GetNodeInfo(instance.primary_node).offline
6731 if self.primary_offline and self.op.ignore_offline_nodes:
6732 self.proc.LogWarning("Ignoring offline primary node")
6734 if self.op.hvparams or self.op.beparams:
6735 self.proc.LogWarning("Overridden parameters are ignored")
6737 _CheckNodeOnline(self, instance.primary_node)
6739 bep = self.cfg.GetClusterInfo().FillBE(instance)
6740 bep.update(self.op.beparams)
6742 # check bridges existence
6743 _CheckInstanceBridgesExist(self, instance)
6745 remote_info = self.rpc.call_instance_info(instance.primary_node,
6747 instance.hypervisor)
6748 remote_info.Raise("Error checking node %s" % instance.primary_node,
6749 prereq=True, ecode=errors.ECODE_ENVIRON)
6750 if not remote_info.payload: # not running already
6751 _CheckNodeFreeMemory(self, instance.primary_node,
6752 "starting instance %s" % instance.name,
6753 bep[constants.BE_MINMEM], instance.hypervisor)
6755 def Exec(self, feedback_fn):
6756 """Start the instance.
6759 instance = self.instance
6760 force = self.op.force
6762 if not self.op.no_remember:
6763 self.cfg.MarkInstanceUp(instance.name)
6765 if self.primary_offline:
6766 assert self.op.ignore_offline_nodes
6767 self.proc.LogInfo("Primary node offline, marked instance as started")
6769 node_current = instance.primary_node
6771 _StartInstanceDisks(self, instance, force)
6774 self.rpc.call_instance_start(node_current,
6775 (instance, self.op.hvparams,
6777 self.op.startup_paused)
6778 msg = result.fail_msg
6780 _ShutdownInstanceDisks(self, instance)
6781 raise errors.OpExecError("Could not start instance: %s" % msg)
6784 class LUInstanceReboot(LogicalUnit):
6785 """Reboot an instance.
6788 HPATH = "instance-reboot"
6789 HTYPE = constants.HTYPE_INSTANCE
6792 def ExpandNames(self):
6793 self._ExpandAndLockInstance()
6795 def BuildHooksEnv(self):
6798 This runs on master, primary and secondary nodes of the instance.
6802 "IGNORE_SECONDARIES": self.op.ignore_secondaries,
6803 "REBOOT_TYPE": self.op.reboot_type,
6804 "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
6807 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6811 def BuildHooksNodes(self):
6812 """Build hooks nodes.
6815 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6818 def CheckPrereq(self):
6819 """Check prerequisites.
6821 This checks that the instance is in the cluster.
6824 self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6825 assert self.instance is not None, \
6826 "Cannot retrieve locked instance %s" % self.op.instance_name
6827 _CheckInstanceState(self, instance, INSTANCE_ONLINE)
6828 _CheckNodeOnline(self, instance.primary_node)
6830 # check bridges existence
6831 _CheckInstanceBridgesExist(self, instance)
6833 def Exec(self, feedback_fn):
6834 """Reboot the instance.
6837 instance = self.instance
6838 ignore_secondaries = self.op.ignore_secondaries
6839 reboot_type = self.op.reboot_type
6841 remote_info = self.rpc.call_instance_info(instance.primary_node,
6843 instance.hypervisor)
6844 remote_info.Raise("Error checking node %s" % instance.primary_node)
6845 instance_running = bool(remote_info.payload)
6847 node_current = instance.primary_node
6849 if instance_running and reboot_type in [constants.INSTANCE_REBOOT_SOFT,
6850 constants.INSTANCE_REBOOT_HARD]:
6851 for disk in instance.disks:
6852 self.cfg.SetDiskID(disk, node_current)
6853 result = self.rpc.call_instance_reboot(node_current, instance,
6855 self.op.shutdown_timeout)
6856 result.Raise("Could not reboot instance")
6858 if instance_running:
6859 result = self.rpc.call_instance_shutdown(node_current, instance,
6860 self.op.shutdown_timeout)
6861 result.Raise("Could not shutdown instance for full reboot")
6862 _ShutdownInstanceDisks(self, instance)
6864 self.LogInfo("Instance %s was already stopped, starting now",
6866 _StartInstanceDisks(self, instance, ignore_secondaries)
6867 result = self.rpc.call_instance_start(node_current,
6868 (instance, None, None), False)
6869 msg = result.fail_msg
6871 _ShutdownInstanceDisks(self, instance)
6872 raise errors.OpExecError("Could not start instance for"
6873 " full reboot: %s" % msg)
6875 self.cfg.MarkInstanceUp(instance.name)
6878 class LUInstanceShutdown(LogicalUnit):
6879 """Shutdown an instance.
6882 HPATH = "instance-stop"
6883 HTYPE = constants.HTYPE_INSTANCE
6886 def ExpandNames(self):
6887 self._ExpandAndLockInstance()
6889 def BuildHooksEnv(self):
6892 This runs on master, primary and secondary nodes of the instance.
6895 env = _BuildInstanceHookEnvByObject(self, self.instance)
6896 env["TIMEOUT"] = self.op.timeout
6899 def BuildHooksNodes(self):
6900 """Build hooks nodes.
6903 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6906 def CheckPrereq(self):
6907 """Check prerequisites.
6909 This checks that the instance is in the cluster.
6912 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6913 assert self.instance is not None, \
6914 "Cannot retrieve locked instance %s" % self.op.instance_name
6916 _CheckInstanceState(self, self.instance, INSTANCE_ONLINE)
6918 self.primary_offline = \
6919 self.cfg.GetNodeInfo(self.instance.primary_node).offline
6921 if self.primary_offline and self.op.ignore_offline_nodes:
6922 self.proc.LogWarning("Ignoring offline primary node")
6924 _CheckNodeOnline(self, self.instance.primary_node)
6926 def Exec(self, feedback_fn):
6927 """Shutdown the instance.
6930 instance = self.instance
6931 node_current = instance.primary_node
6932 timeout = self.op.timeout
6934 if not self.op.no_remember:
6935 self.cfg.MarkInstanceDown(instance.name)
6937 if self.primary_offline:
6938 assert self.op.ignore_offline_nodes
6939 self.proc.LogInfo("Primary node offline, marked instance as stopped")
6941 result = self.rpc.call_instance_shutdown(node_current, instance, timeout)
6942 msg = result.fail_msg
6944 self.proc.LogWarning("Could not shutdown instance: %s" % msg)
6946 _ShutdownInstanceDisks(self, instance)
6949 class LUInstanceReinstall(LogicalUnit):
6950 """Reinstall an instance.
6953 HPATH = "instance-reinstall"
6954 HTYPE = constants.HTYPE_INSTANCE
6957 def ExpandNames(self):
6958 self._ExpandAndLockInstance()
6960 def BuildHooksEnv(self):
6963 This runs on master, primary and secondary nodes of the instance.
6966 return _BuildInstanceHookEnvByObject(self, self.instance)
6968 def BuildHooksNodes(self):
6969 """Build hooks nodes.
6972 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6975 def CheckPrereq(self):
6976 """Check prerequisites.
6978 This checks that the instance is in the cluster and is not running.
6981 instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6982 assert instance is not None, \
6983 "Cannot retrieve locked instance %s" % self.op.instance_name
6984 _CheckNodeOnline(self, instance.primary_node, "Instance primary node"
6985 " offline, cannot reinstall")
6986 for node in instance.secondary_nodes:
6987 _CheckNodeOnline(self, node, "Instance secondary node offline,"
6988 " cannot reinstall")
6990 if instance.disk_template == constants.DT_DISKLESS:
6991 raise errors.OpPrereqError("Instance '%s' has no disks" %
6992 self.op.instance_name,
6994 _CheckInstanceState(self, instance, INSTANCE_DOWN, msg="cannot reinstall")
6996 if self.op.os_type is not None:
6998 pnode = _ExpandNodeName(self.cfg, instance.primary_node)
6999 _CheckNodeHasOS(self, pnode, self.op.os_type, self.op.force_variant)
7000 instance_os = self.op.os_type
7002 instance_os = instance.os
7004 nodelist = list(instance.all_nodes)
7006 if self.op.osparams:
7007 i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
7008 _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
7009 self.os_inst = i_osdict # the new dict (without defaults)
7013 self.instance = instance
7015 def Exec(self, feedback_fn):
7016 """Reinstall the instance.
7019 inst = self.instance
7021 if self.op.os_type is not None:
7022 feedback_fn("Changing OS to '%s'..." % self.op.os_type)
7023 inst.os = self.op.os_type
7024 # Write to configuration
7025 self.cfg.Update(inst, feedback_fn)
7027 _StartInstanceDisks(self, inst, None)
7029 feedback_fn("Running the instance OS create scripts...")
7030 # FIXME: pass debug option from opcode to backend
7031 result = self.rpc.call_instance_os_add(inst.primary_node,
7032 (inst, self.os_inst), True,
7033 self.op.debug_level)
7034 result.Raise("Could not install OS for instance %s on node %s" %
7035 (inst.name, inst.primary_node))
7037 _ShutdownInstanceDisks(self, inst)
7040 class LUInstanceRecreateDisks(LogicalUnit):
7041 """Recreate an instance's missing disks.
7044 HPATH = "instance-recreate-disks"
7045 HTYPE = constants.HTYPE_INSTANCE
7048 _MODIFYABLE = frozenset([
7049 constants.IDISK_SIZE,
7050 constants.IDISK_MODE,
7053 # New or changed disk parameters may have different semantics
7054 assert constants.IDISK_PARAMS == (_MODIFYABLE | frozenset([
7055 constants.IDISK_ADOPT,
7057 # TODO: Implement support changing VG while recreating
7059 constants.IDISK_METAVG,
7062 def CheckArguments(self):
7063 if self.op.disks and ht.TPositiveInt(self.op.disks[0]):
7064 # Normalize and convert deprecated list of disk indices
7065 self.op.disks = [(idx, {}) for idx in sorted(frozenset(self.op.disks))]
7067 duplicates = utils.FindDuplicates(map(compat.fst, self.op.disks))
7069 raise errors.OpPrereqError("Some disks have been specified more than"
7070 " once: %s" % utils.CommaJoin(duplicates),
7073 for (idx, params) in self.op.disks:
7074 utils.ForceDictType(params, constants.IDISK_PARAMS_TYPES)
7075 unsupported = frozenset(params.keys()) - self._MODIFYABLE
7077 raise errors.OpPrereqError("Parameters for disk %s try to change"
7078 " unmodifyable parameter(s): %s" %
7079 (idx, utils.CommaJoin(unsupported)),
7082 def ExpandNames(self):
7083 self._ExpandAndLockInstance()
7084 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
7086 self.op.nodes = [_ExpandNodeName(self.cfg, n) for n in self.op.nodes]
7087 self.needed_locks[locking.LEVEL_NODE] = list(self.op.nodes)
7089 self.needed_locks[locking.LEVEL_NODE] = []
7090 self.needed_locks[locking.LEVEL_NODE_RES] = []
7092 def DeclareLocks(self, level):
7093 if level == locking.LEVEL_NODE:
7094 # if we replace the nodes, we only need to lock the old primary,
7095 # otherwise we need to lock all nodes for disk re-creation
7096 primary_only = bool(self.op.nodes)
7097 self._LockInstancesNodes(primary_only=primary_only)
7098 elif level == locking.LEVEL_NODE_RES:
7100 self.needed_locks[locking.LEVEL_NODE_RES] = \
7101 self.needed_locks[locking.LEVEL_NODE][:]
7103 def BuildHooksEnv(self):
7106 This runs on master, primary and secondary nodes of the instance.
7109 return _BuildInstanceHookEnvByObject(self, self.instance)
7111 def BuildHooksNodes(self):
7112 """Build hooks nodes.
7115 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
7118 def CheckPrereq(self):
7119 """Check prerequisites.
7121 This checks that the instance is in the cluster and is not running.
7124 instance = self.cfg.GetInstanceInfo(self.op.instance_name)
7125 assert instance is not None, \
7126 "Cannot retrieve locked instance %s" % self.op.instance_name
7128 if len(self.op.nodes) != len(instance.all_nodes):
7129 raise errors.OpPrereqError("Instance %s currently has %d nodes, but"
7130 " %d replacement nodes were specified" %
7131 (instance.name, len(instance.all_nodes),
7132 len(self.op.nodes)),
7134 assert instance.disk_template != constants.DT_DRBD8 or \
7135 len(self.op.nodes) == 2
7136 assert instance.disk_template != constants.DT_PLAIN or \
7137 len(self.op.nodes) == 1
7138 primary_node = self.op.nodes[0]
7140 primary_node = instance.primary_node
7141 _CheckNodeOnline(self, primary_node)
7143 if instance.disk_template == constants.DT_DISKLESS:
7144 raise errors.OpPrereqError("Instance '%s' has no disks" %
7145 self.op.instance_name, errors.ECODE_INVAL)
7147 # if we replace nodes *and* the old primary is offline, we don't
7149 assert instance.primary_node in self.owned_locks(locking.LEVEL_NODE)
7150 assert instance.primary_node in self.owned_locks(locking.LEVEL_NODE_RES)
7151 old_pnode = self.cfg.GetNodeInfo(instance.primary_node)
7152 if not (self.op.nodes and old_pnode.offline):
7153 _CheckInstanceState(self, instance, INSTANCE_NOT_RUNNING,
7154 msg="cannot recreate disks")
7157 self.disks = dict(self.op.disks)
7159 self.disks = dict((idx, {}) for idx in range(len(instance.disks)))
7161 maxidx = max(self.disks.keys())
7162 if maxidx >= len(instance.disks):
7163 raise errors.OpPrereqError("Invalid disk index '%s'" % maxidx,
7166 if (self.op.nodes and
7167 sorted(self.disks.keys()) != range(len(instance.disks))):
7168 raise errors.OpPrereqError("Can't recreate disks partially and"
7169 " change the nodes at the same time",
7172 self.instance = instance
7174 def Exec(self, feedback_fn):
7175 """Recreate the disks.
7178 instance = self.instance
7180 assert (self.owned_locks(locking.LEVEL_NODE) ==
7181 self.owned_locks(locking.LEVEL_NODE_RES))
7184 mods = [] # keeps track of needed changes
7186 for idx, disk in enumerate(instance.disks):
7188 changes = self.disks[idx]
7190 # Disk should not be recreated
7194 # update secondaries for disks, if needed
7195 if self.op.nodes and disk.dev_type == constants.LD_DRBD8:
7196 # need to update the nodes and minors
7197 assert len(self.op.nodes) == 2
7198 assert len(disk.logical_id) == 6 # otherwise disk internals
7200 (_, _, old_port, _, _, old_secret) = disk.logical_id
7201 new_minors = self.cfg.AllocateDRBDMinor(self.op.nodes, instance.name)
7202 new_id = (self.op.nodes[0], self.op.nodes[1], old_port,
7203 new_minors[0], new_minors[1], old_secret)
7204 assert len(disk.logical_id) == len(new_id)
7208 mods.append((idx, new_id, changes))
7210 # now that we have passed all asserts above, we can apply the mods
7211 # in a single run (to avoid partial changes)
7212 for idx, new_id, changes in mods:
7213 disk = instance.disks[idx]
7214 if new_id is not None:
7215 assert disk.dev_type == constants.LD_DRBD8
7216 disk.logical_id = new_id
7218 disk.Update(size=changes.get(constants.IDISK_SIZE, None),
7219 mode=changes.get(constants.IDISK_MODE, None))
7221 # change primary node, if needed
7223 instance.primary_node = self.op.nodes[0]
7224 self.LogWarning("Changing the instance's nodes, you will have to"
7225 " remove any disks left on the older nodes manually")
7228 self.cfg.Update(instance, feedback_fn)
7230 _CreateDisks(self, instance, to_skip=to_skip)
7233 class LUInstanceRename(LogicalUnit):
7234 """Rename an instance.
7237 HPATH = "instance-rename"
7238 HTYPE = constants.HTYPE_INSTANCE
7240 def CheckArguments(self):
7244 if self.op.ip_check and not self.op.name_check:
7245 # TODO: make the ip check more flexible and not depend on the name check
7246 raise errors.OpPrereqError("IP address check requires a name check",
7249 def BuildHooksEnv(self):
7252 This runs on master, primary and secondary nodes of the instance.
7255 env = _BuildInstanceHookEnvByObject(self, self.instance)
7256 env["INSTANCE_NEW_NAME"] = self.op.new_name
7259 def BuildHooksNodes(self):
7260 """Build hooks nodes.
7263 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
7266 def CheckPrereq(self):
7267 """Check prerequisites.
7269 This checks that the instance is in the cluster and is not running.
7272 self.op.instance_name = _ExpandInstanceName(self.cfg,
7273 self.op.instance_name)
7274 instance = self.cfg.GetInstanceInfo(self.op.instance_name)
7275 assert instance is not None
7276 _CheckNodeOnline(self, instance.primary_node)
7277 _CheckInstanceState(self, instance, INSTANCE_NOT_RUNNING,
7278 msg="cannot rename")
7279 self.instance = instance
7281 new_name = self.op.new_name
7282 if self.op.name_check:
7283 hostname = netutils.GetHostname(name=new_name)
7284 if hostname.name != new_name:
7285 self.LogInfo("Resolved given name '%s' to '%s'", new_name,
7287 if not utils.MatchNameComponent(self.op.new_name, [hostname.name]):
7288 raise errors.OpPrereqError(("Resolved hostname '%s' does not look the"
7289 " same as given hostname '%s'") %
7290 (hostname.name, self.op.new_name),
7292 new_name = self.op.new_name = hostname.name
7293 if (self.op.ip_check and
7294 netutils.TcpPing(hostname.ip, constants.DEFAULT_NODED_PORT)):
7295 raise errors.OpPrereqError("IP %s of instance %s already in use" %
7296 (hostname.ip, new_name),
7297 errors.ECODE_NOTUNIQUE)
7299 instance_list = self.cfg.GetInstanceList()
7300 if new_name in instance_list and new_name != instance.name:
7301 raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
7302 new_name, errors.ECODE_EXISTS)
7304 def Exec(self, feedback_fn):
7305 """Rename the instance.
7308 inst = self.instance
7309 old_name = inst.name
7311 rename_file_storage = False
7312 if (inst.disk_template in constants.DTS_FILEBASED and
7313 self.op.new_name != inst.name):
7314 old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
7315 rename_file_storage = True
7317 self.cfg.RenameInstance(inst.name, self.op.new_name)
7318 # Change the instance lock. This is definitely safe while we hold the BGL.
7319 # Otherwise the new lock would have to be added in acquired mode.
7321 self.glm.remove(locking.LEVEL_INSTANCE, old_name)
7322 self.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
7324 # re-read the instance from the configuration after rename
7325 inst = self.cfg.GetInstanceInfo(self.op.new_name)
7327 if rename_file_storage:
7328 new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
7329 result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
7330 old_file_storage_dir,
7331 new_file_storage_dir)
7332 result.Raise("Could not rename on node %s directory '%s' to '%s'"
7333 " (but the instance has been renamed in Ganeti)" %
7334 (inst.primary_node, old_file_storage_dir,
7335 new_file_storage_dir))
7337 _StartInstanceDisks(self, inst, None)
7339 result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
7340 old_name, self.op.debug_level)
7341 msg = result.fail_msg
7343 msg = ("Could not run OS rename script for instance %s on node %s"
7344 " (but the instance has been renamed in Ganeti): %s" %
7345 (inst.name, inst.primary_node, msg))
7346 self.proc.LogWarning(msg)
7348 _ShutdownInstanceDisks(self, inst)
7353 class LUInstanceRemove(LogicalUnit):
7354 """Remove an instance.
7357 HPATH = "instance-remove"
7358 HTYPE = constants.HTYPE_INSTANCE
7361 def ExpandNames(self):
7362 self._ExpandAndLockInstance()
7363 self.needed_locks[locking.LEVEL_NODE] = []
7364 self.needed_locks[locking.LEVEL_NODE_RES] = []
7365 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
7367 def DeclareLocks(self, level):
7368 if level == locking.LEVEL_NODE:
7369 self._LockInstancesNodes()
7370 elif level == locking.LEVEL_NODE_RES:
7372 self.needed_locks[locking.LEVEL_NODE_RES] = \
7373 self.needed_locks[locking.LEVEL_NODE][:]
7375 def BuildHooksEnv(self):
7378 This runs on master, primary and secondary nodes of the instance.
7381 env = _BuildInstanceHookEnvByObject(self, self.instance)
7382 env["SHUTDOWN_TIMEOUT"] = self.op.shutdown_timeout
7385 def BuildHooksNodes(self):
7386 """Build hooks nodes.
7389 nl = [self.cfg.GetMasterNode()]
7390 nl_post = list(self.instance.all_nodes) + nl
7391 return (nl, nl_post)
7393 def CheckPrereq(self):
7394 """Check prerequisites.
7396 This checks that the instance is in the cluster.
7399 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
7400 assert self.instance is not None, \
7401 "Cannot retrieve locked instance %s" % self.op.instance_name
7403 def Exec(self, feedback_fn):
7404 """Remove the instance.
7407 instance = self.instance
7408 logging.info("Shutting down instance %s on node %s",
7409 instance.name, instance.primary_node)
7411 result = self.rpc.call_instance_shutdown(instance.primary_node, instance,
7412 self.op.shutdown_timeout)
7413 msg = result.fail_msg
7415 if self.op.ignore_failures:
7416 feedback_fn("Warning: can't shutdown instance: %s" % msg)
7418 raise errors.OpExecError("Could not shutdown instance %s on"
7420 (instance.name, instance.primary_node, msg))
7422 assert (self.owned_locks(locking.LEVEL_NODE) ==
7423 self.owned_locks(locking.LEVEL_NODE_RES))
7424 assert not (set(instance.all_nodes) -
7425 self.owned_locks(locking.LEVEL_NODE)), \
7426 "Not owning correct locks"
7428 _RemoveInstance(self, feedback_fn, instance, self.op.ignore_failures)
7431 def _RemoveInstance(lu, feedback_fn, instance, ignore_failures):
7432 """Utility function to remove an instance.
7435 logging.info("Removing block devices for instance %s", instance.name)
7437 if not _RemoveDisks(lu, instance, ignore_failures=ignore_failures):
7438 if not ignore_failures:
7439 raise errors.OpExecError("Can't remove instance's disks")
7440 feedback_fn("Warning: can't remove instance's disks")
7442 logging.info("Removing instance %s out of cluster config", instance.name)
7444 lu.cfg.RemoveInstance(instance.name)
7446 assert not lu.remove_locks.get(locking.LEVEL_INSTANCE), \
7447 "Instance lock removal conflict"
7449 # Remove lock for the instance
7450 lu.remove_locks[locking.LEVEL_INSTANCE] = instance.name
7453 class LUInstanceQuery(NoHooksLU):
7454 """Logical unit for querying instances.
7457 # pylint: disable=W0142
7460 def CheckArguments(self):
7461 self.iq = _InstanceQuery(qlang.MakeSimpleFilter("name", self.op.names),
7462 self.op.output_fields, self.op.use_locking)
7464 def ExpandNames(self):
7465 self.iq.ExpandNames(self)
7467 def DeclareLocks(self, level):
7468 self.iq.DeclareLocks(self, level)
7470 def Exec(self, feedback_fn):
7471 return self.iq.OldStyleQuery(self)
7474 class LUInstanceFailover(LogicalUnit):
7475 """Failover an instance.
7478 HPATH = "instance-failover"
7479 HTYPE = constants.HTYPE_INSTANCE
7482 def CheckArguments(self):
7483 """Check the arguments.
7486 self.iallocator = getattr(self.op, "iallocator", None)
7487 self.target_node = getattr(self.op, "target_node", None)
7489 def ExpandNames(self):
7490 self._ExpandAndLockInstance()
7492 if self.op.target_node is not None:
7493 self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
7495 self.needed_locks[locking.LEVEL_NODE] = []
7496 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
7498 self.needed_locks[locking.LEVEL_NODE_RES] = []
7499 self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
7501 ignore_consistency = self.op.ignore_consistency
7502 shutdown_timeout = self.op.shutdown_timeout
7503 self._migrater = TLMigrateInstance(self, self.op.instance_name,
7506 ignore_consistency=ignore_consistency,
7507 shutdown_timeout=shutdown_timeout,
7508 ignore_ipolicy=self.op.ignore_ipolicy)
7509 self.tasklets = [self._migrater]
7511 def DeclareLocks(self, level):
7512 if level == locking.LEVEL_NODE:
7513 instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
7514 if instance.disk_template in constants.DTS_EXT_MIRROR:
7515 if self.op.target_node is None:
7516 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7518 self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
7519 self.op.target_node]
7520 del self.recalculate_locks[locking.LEVEL_NODE]
7522 self._LockInstancesNodes()
7523 elif level == locking.LEVEL_NODE_RES:
7525 self.needed_locks[locking.LEVEL_NODE_RES] = \
7526 self.needed_locks[locking.LEVEL_NODE][:]
7528 def BuildHooksEnv(self):
7531 This runs on master, primary and secondary nodes of the instance.
7534 instance = self._migrater.instance
7535 source_node = instance.primary_node
7536 target_node = self.op.target_node
7538 "IGNORE_CONSISTENCY": self.op.ignore_consistency,
7539 "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
7540 "OLD_PRIMARY": source_node,
7541 "NEW_PRIMARY": target_node,
7544 if instance.disk_template in constants.DTS_INT_MIRROR:
7545 env["OLD_SECONDARY"] = instance.secondary_nodes[0]
7546 env["NEW_SECONDARY"] = source_node
7548 env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = ""
7550 env.update(_BuildInstanceHookEnvByObject(self, instance))
7554 def BuildHooksNodes(self):
7555 """Build hooks nodes.
7558 instance = self._migrater.instance
7559 nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
7560 return (nl, nl + [instance.primary_node])
7563 class LUInstanceMigrate(LogicalUnit):
7564 """Migrate an instance.
7566 This is migration without shutting down, compared to the failover,
7567 which is done with shutdown.
7570 HPATH = "instance-migrate"
7571 HTYPE = constants.HTYPE_INSTANCE
7574 def ExpandNames(self):
7575 self._ExpandAndLockInstance()
7577 if self.op.target_node is not None:
7578 self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
7580 self.needed_locks[locking.LEVEL_NODE] = []
7581 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
7583 self.needed_locks[locking.LEVEL_NODE] = []
7584 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
7587 TLMigrateInstance(self, self.op.instance_name,
7588 cleanup=self.op.cleanup,
7590 fallback=self.op.allow_failover,
7591 allow_runtime_changes=self.op.allow_runtime_changes,
7592 ignore_ipolicy=self.op.ignore_ipolicy)
7593 self.tasklets = [self._migrater]
7595 def DeclareLocks(self, level):
7596 if level == locking.LEVEL_NODE:
7597 instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
7598 if instance.disk_template in constants.DTS_EXT_MIRROR:
7599 if self.op.target_node is None:
7600 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7602 self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
7603 self.op.target_node]
7604 del self.recalculate_locks[locking.LEVEL_NODE]
7606 self._LockInstancesNodes()
7607 elif level == locking.LEVEL_NODE_RES:
7609 self.needed_locks[locking.LEVEL_NODE_RES] = \
7610 self.needed_locks[locking.LEVEL_NODE][:]
7612 def BuildHooksEnv(self):
7615 This runs on master, primary and secondary nodes of the instance.
7618 instance = self._migrater.instance
7619 source_node = instance.primary_node
7620 target_node = self.op.target_node
7621 env = _BuildInstanceHookEnvByObject(self, instance)
7623 "MIGRATE_LIVE": self._migrater.live,
7624 "MIGRATE_CLEANUP": self.op.cleanup,
7625 "OLD_PRIMARY": source_node,
7626 "NEW_PRIMARY": target_node,
7627 "ALLOW_RUNTIME_CHANGES": self.op.allow_runtime_changes,
7630 if instance.disk_template in constants.DTS_INT_MIRROR:
7631 env["OLD_SECONDARY"] = target_node
7632 env["NEW_SECONDARY"] = source_node
7634 env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = None
7638 def BuildHooksNodes(self):
7639 """Build hooks nodes.
7642 instance = self._migrater.instance
7643 nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
7644 return (nl, nl + [instance.primary_node])
7647 class LUInstanceMove(LogicalUnit):
7648 """Move an instance by data-copying.
7651 HPATH = "instance-move"
7652 HTYPE = constants.HTYPE_INSTANCE
7655 def ExpandNames(self):
7656 self._ExpandAndLockInstance()
7657 target_node = _ExpandNodeName(self.cfg, self.op.target_node)
7658 self.op.target_node = target_node
7659 self.needed_locks[locking.LEVEL_NODE] = [target_node]
7660 self.needed_locks[locking.LEVEL_NODE_RES] = []
7661 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
7663 def DeclareLocks(self, level):
7664 if level == locking.LEVEL_NODE:
7665 self._LockInstancesNodes(primary_only=True)
7666 elif level == locking.LEVEL_NODE_RES:
7668 self.needed_locks[locking.LEVEL_NODE_RES] = \
7669 self.needed_locks[locking.LEVEL_NODE][:]
7671 def BuildHooksEnv(self):
7674 This runs on master, primary and secondary nodes of the instance.
7678 "TARGET_NODE": self.op.target_node,
7679 "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
7681 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
7684 def BuildHooksNodes(self):
7685 """Build hooks nodes.
7689 self.cfg.GetMasterNode(),
7690 self.instance.primary_node,
7691 self.op.target_node,
7695 def CheckPrereq(self):
7696 """Check prerequisites.
7698 This checks that the instance is in the cluster.
7701 self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
7702 assert self.instance is not None, \
7703 "Cannot retrieve locked instance %s" % self.op.instance_name
7705 node = self.cfg.GetNodeInfo(self.op.target_node)
7706 assert node is not None, \
7707 "Cannot retrieve locked node %s" % self.op.target_node
7709 self.target_node = target_node = node.name
7711 if target_node == instance.primary_node:
7712 raise errors.OpPrereqError("Instance %s is already on the node %s" %
7713 (instance.name, target_node),
7716 bep = self.cfg.GetClusterInfo().FillBE(instance)
7718 for idx, dsk in enumerate(instance.disks):
7719 if dsk.dev_type not in (constants.LD_LV, constants.LD_FILE):
7720 raise errors.OpPrereqError("Instance disk %d has a complex layout,"
7721 " cannot copy" % idx, errors.ECODE_STATE)
7723 _CheckNodeOnline(self, target_node)
7724 _CheckNodeNotDrained(self, target_node)
7725 _CheckNodeVmCapable(self, target_node)
7726 ipolicy = _CalculateGroupIPolicy(self.cfg.GetClusterInfo(),
7727 self.cfg.GetNodeGroup(node.group))
7728 _CheckTargetNodeIPolicy(self, ipolicy, instance, node,
7729 ignore=self.op.ignore_ipolicy)
7731 if instance.admin_state == constants.ADMINST_UP:
7732 # check memory requirements on the secondary node
7733 _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
7734 instance.name, bep[constants.BE_MAXMEM],
7735 instance.hypervisor)
7737 self.LogInfo("Not checking memory on the secondary node as"
7738 " instance will not be started")
7740 # check bridge existance
7741 _CheckInstanceBridgesExist(self, instance, node=target_node)
7743 def Exec(self, feedback_fn):
7744 """Move an instance.
7746 The move is done by shutting it down on its present node, copying
7747 the data over (slow) and starting it on the new node.
7750 instance = self.instance
7752 source_node = instance.primary_node
7753 target_node = self.target_node
7755 self.LogInfo("Shutting down instance %s on source node %s",
7756 instance.name, source_node)
7758 assert (self.owned_locks(locking.LEVEL_NODE) ==
7759 self.owned_locks(locking.LEVEL_NODE_RES))
7761 result = self.rpc.call_instance_shutdown(source_node, instance,
7762 self.op.shutdown_timeout)
7763 msg = result.fail_msg
7765 if self.op.ignore_consistency:
7766 self.proc.LogWarning("Could not shutdown instance %s on node %s."
7767 " Proceeding anyway. Please make sure node"
7768 " %s is down. Error details: %s",
7769 instance.name, source_node, source_node, msg)
7771 raise errors.OpExecError("Could not shutdown instance %s on"
7773 (instance.name, source_node, msg))
7775 # create the target disks
7777 _CreateDisks(self, instance, target_node=target_node)
7778 except errors.OpExecError:
7779 self.LogWarning("Device creation failed, reverting...")
7781 _RemoveDisks(self, instance, target_node=target_node)
7783 self.cfg.ReleaseDRBDMinors(instance.name)
7786 cluster_name = self.cfg.GetClusterInfo().cluster_name
7789 # activate, get path, copy the data over
7790 for idx, disk in enumerate(instance.disks):
7791 self.LogInfo("Copying data for disk %d", idx)
7792 result = self.rpc.call_blockdev_assemble(target_node, (disk, instance),
7793 instance.name, True, idx)
7795 self.LogWarning("Can't assemble newly created disk %d: %s",
7796 idx, result.fail_msg)
7797 errs.append(result.fail_msg)
7799 dev_path = result.payload
7800 result = self.rpc.call_blockdev_export(source_node, (disk, instance),
7801 target_node, dev_path,
7804 self.LogWarning("Can't copy data over for disk %d: %s",
7805 idx, result.fail_msg)
7806 errs.append(result.fail_msg)
7810 self.LogWarning("Some disks failed to copy, aborting")
7812 _RemoveDisks(self, instance, target_node=target_node)
7814 self.cfg.ReleaseDRBDMinors(instance.name)
7815 raise errors.OpExecError("Errors during disk copy: %s" %
7818 instance.primary_node = target_node
7819 self.cfg.Update(instance, feedback_fn)
7821 self.LogInfo("Removing the disks on the original node")
7822 _RemoveDisks(self, instance, target_node=source_node)
7824 # Only start the instance if it's marked as up
7825 if instance.admin_state == constants.ADMINST_UP:
7826 self.LogInfo("Starting instance %s on node %s",
7827 instance.name, target_node)
7829 disks_ok, _ = _AssembleInstanceDisks(self, instance,
7830 ignore_secondaries=True)
7832 _ShutdownInstanceDisks(self, instance)
7833 raise errors.OpExecError("Can't activate the instance's disks")
7835 result = self.rpc.call_instance_start(target_node,
7836 (instance, None, None), False)
7837 msg = result.fail_msg
7839 _ShutdownInstanceDisks(self, instance)
7840 raise errors.OpExecError("Could not start instance %s on node %s: %s" %
7841 (instance.name, target_node, msg))
7844 class LUNodeMigrate(LogicalUnit):
7845 """Migrate all instances from a node.
7848 HPATH = "node-migrate"
7849 HTYPE = constants.HTYPE_NODE
7852 def CheckArguments(self):
7855 def ExpandNames(self):
7856 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
7858 self.share_locks = _ShareAll()
7859 self.needed_locks = {
7860 locking.LEVEL_NODE: [self.op.node_name],
7863 def BuildHooksEnv(self):
7866 This runs on the master, the primary and all the secondaries.
7870 "NODE_NAME": self.op.node_name,
7871 "ALLOW_RUNTIME_CHANGES": self.op.allow_runtime_changes,
7874 def BuildHooksNodes(self):
7875 """Build hooks nodes.
7878 nl = [self.cfg.GetMasterNode()]
7881 def CheckPrereq(self):
7884 def Exec(self, feedback_fn):
7885 # Prepare jobs for migration instances
7886 allow_runtime_changes = self.op.allow_runtime_changes
7888 [opcodes.OpInstanceMigrate(instance_name=inst.name,
7891 iallocator=self.op.iallocator,
7892 target_node=self.op.target_node,
7893 allow_runtime_changes=allow_runtime_changes,
7894 ignore_ipolicy=self.op.ignore_ipolicy)]
7895 for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name)
7898 # TODO: Run iallocator in this opcode and pass correct placement options to
7899 # OpInstanceMigrate. Since other jobs can modify the cluster between
7900 # running the iallocator and the actual migration, a good consistency model
7901 # will have to be found.
7903 assert (frozenset(self.owned_locks(locking.LEVEL_NODE)) ==
7904 frozenset([self.op.node_name]))
7906 return ResultWithJobs(jobs)
7909 class TLMigrateInstance(Tasklet):
7910 """Tasklet class for instance migration.
7913 @ivar live: whether the migration will be done live or non-live;
7914 this variable is initalized only after CheckPrereq has run
7915 @type cleanup: boolean
7916 @ivar cleanup: Wheater we cleanup from a failed migration
7917 @type iallocator: string
7918 @ivar iallocator: The iallocator used to determine target_node
7919 @type target_node: string
7920 @ivar target_node: If given, the target_node to reallocate the instance to
7921 @type failover: boolean
7922 @ivar failover: Whether operation results in failover or migration
7923 @type fallback: boolean
7924 @ivar fallback: Whether fallback to failover is allowed if migration not
7926 @type ignore_consistency: boolean
7927 @ivar ignore_consistency: Wheter we should ignore consistency between source
7929 @type shutdown_timeout: int
7930 @ivar shutdown_timeout: In case of failover timeout of the shutdown
7931 @type ignore_ipolicy: bool
7932 @ivar ignore_ipolicy: If true, we can ignore instance policy when migrating
7937 _MIGRATION_POLL_INTERVAL = 1 # seconds
7938 _MIGRATION_FEEDBACK_INTERVAL = 10 # seconds
7940 def __init__(self, lu, instance_name, cleanup=False,
7941 failover=False, fallback=False,
7942 ignore_consistency=False,
7943 allow_runtime_changes=True,
7944 shutdown_timeout=constants.DEFAULT_SHUTDOWN_TIMEOUT,
7945 ignore_ipolicy=False):
7946 """Initializes this class.
7949 Tasklet.__init__(self, lu)
7952 self.instance_name = instance_name
7953 self.cleanup = cleanup
7954 self.live = False # will be overridden later
7955 self.failover = failover
7956 self.fallback = fallback
7957 self.ignore_consistency = ignore_consistency
7958 self.shutdown_timeout = shutdown_timeout
7959 self.ignore_ipolicy = ignore_ipolicy
7960 self.allow_runtime_changes = allow_runtime_changes
7962 def CheckPrereq(self):
7963 """Check prerequisites.
7965 This checks that the instance is in the cluster.
7968 instance_name = _ExpandInstanceName(self.lu.cfg, self.instance_name)
7969 instance = self.cfg.GetInstanceInfo(instance_name)
7970 assert instance is not None
7971 self.instance = instance
7972 cluster = self.cfg.GetClusterInfo()
7974 if (not self.cleanup and
7975 not instance.admin_state == constants.ADMINST_UP and
7976 not self.failover and self.fallback):
7977 self.lu.LogInfo("Instance is marked down or offline, fallback allowed,"
7978 " switching to failover")
7979 self.failover = True
7981 if instance.disk_template not in constants.DTS_MIRRORED:
7986 raise errors.OpPrereqError("Instance's disk layout '%s' does not allow"
7987 " %s" % (instance.disk_template, text),
7990 if instance.disk_template in constants.DTS_EXT_MIRROR:
7991 _CheckIAllocatorOrNode(self.lu, "iallocator", "target_node")
7993 if self.lu.op.iallocator:
7994 self._RunAllocator()
7996 # We set set self.target_node as it is required by
7998 self.target_node = self.lu.op.target_node
8000 # Check that the target node is correct in terms of instance policy
8001 nodeinfo = self.cfg.GetNodeInfo(self.target_node)
8002 group_info = self.cfg.GetNodeGroup(nodeinfo.group)
8003 ipolicy = _CalculateGroupIPolicy(cluster, group_info)
8004 _CheckTargetNodeIPolicy(self.lu, ipolicy, instance, nodeinfo,
8005 ignore=self.ignore_ipolicy)
8007 # self.target_node is already populated, either directly or by the
8009 target_node = self.target_node
8010 if self.target_node == instance.primary_node:
8011 raise errors.OpPrereqError("Cannot migrate instance %s"
8012 " to its primary (%s)" %
8013 (instance.name, instance.primary_node))
8015 if len(self.lu.tasklets) == 1:
8016 # It is safe to release locks only when we're the only tasklet
8018 _ReleaseLocks(self.lu, locking.LEVEL_NODE,
8019 keep=[instance.primary_node, self.target_node])
8022 secondary_nodes = instance.secondary_nodes
8023 if not secondary_nodes:
8024 raise errors.ConfigurationError("No secondary node but using"
8025 " %s disk template" %
8026 instance.disk_template)
8027 target_node = secondary_nodes[0]
8028 if self.lu.op.iallocator or (self.lu.op.target_node and
8029 self.lu.op.target_node != target_node):
8031 text = "failed over"
8034 raise errors.OpPrereqError("Instances with disk template %s cannot"
8035 " be %s to arbitrary nodes"
8036 " (neither an iallocator nor a target"
8037 " node can be passed)" %
8038 (instance.disk_template, text),
8040 nodeinfo = self.cfg.GetNodeInfo(target_node)
8041 group_info = self.cfg.GetNodeGroup(nodeinfo.group)
8042 ipolicy = _CalculateGroupIPolicy(cluster, group_info)
8043 _CheckTargetNodeIPolicy(self.lu, ipolicy, instance, nodeinfo,
8044 ignore=self.ignore_ipolicy)
8046 i_be = cluster.FillBE(instance)
8048 # check memory requirements on the secondary node
8049 if (not self.cleanup and
8050 (not self.failover or instance.admin_state == constants.ADMINST_UP)):
8051 self.tgt_free_mem = _CheckNodeFreeMemory(self.lu, target_node,
8052 "migrating instance %s" %
8054 i_be[constants.BE_MINMEM],
8055 instance.hypervisor)
8057 self.lu.LogInfo("Not checking memory on the secondary node as"
8058 " instance will not be started")
8060 # check if failover must be forced instead of migration
8061 if (not self.cleanup and not self.failover and
8062 i_be[constants.BE_ALWAYS_FAILOVER]):
8064 self.lu.LogInfo("Instance configured to always failover; fallback"
8066 self.failover = True
8068 raise errors.OpPrereqError("This instance has been configured to"
8069 " always failover, please allow failover",
8072 # check bridge existance
8073 _CheckInstanceBridgesExist(self.lu, instance, node=target_node)
8075 if not self.cleanup:
8076 _CheckNodeNotDrained(self.lu, target_node)
8077 if not self.failover:
8078 result = self.rpc.call_instance_migratable(instance.primary_node,
8080 if result.fail_msg and self.fallback:
8081 self.lu.LogInfo("Can't migrate, instance offline, fallback to"
8083 self.failover = True
8085 result.Raise("Can't migrate, please use failover",
8086 prereq=True, ecode=errors.ECODE_STATE)
8088 assert not (self.failover and self.cleanup)
8090 if not self.failover:
8091 if self.lu.op.live is not None and self.lu.op.mode is not None:
8092 raise errors.OpPrereqError("Only one of the 'live' and 'mode'"
8093 " parameters are accepted",
8095 if self.lu.op.live is not None:
8097 self.lu.op.mode = constants.HT_MIGRATION_LIVE
8099 self.lu.op.mode = constants.HT_MIGRATION_NONLIVE
8100 # reset the 'live' parameter to None so that repeated
8101 # invocations of CheckPrereq do not raise an exception
8102 self.lu.op.live = None
8103 elif self.lu.op.mode is None:
8104 # read the default value from the hypervisor
8105 i_hv = cluster.FillHV(self.instance, skip_globals=False)
8106 self.lu.op.mode = i_hv[constants.HV_MIGRATION_MODE]
8108 self.live = self.lu.op.mode == constants.HT_MIGRATION_LIVE
8110 # Failover is never live
8113 if not (self.failover or self.cleanup):
8114 remote_info = self.rpc.call_instance_info(instance.primary_node,
8116 instance.hypervisor)
8117 remote_info.Raise("Error checking instance on node %s" %
8118 instance.primary_node)
8119 instance_running = bool(remote_info.payload)
8120 if instance_running:
8121 self.current_mem = int(remote_info.payload["memory"])
8123 def _RunAllocator(self):
8124 """Run the allocator based on input opcode.
8127 # FIXME: add a self.ignore_ipolicy option
8128 ial = IAllocator(self.cfg, self.rpc,
8129 mode=constants.IALLOCATOR_MODE_RELOC,
8130 name=self.instance_name,
8131 relocate_from=[self.instance.primary_node],
8134 ial.Run(self.lu.op.iallocator)
8137 raise errors.OpPrereqError("Can't compute nodes using"
8138 " iallocator '%s': %s" %
8139 (self.lu.op.iallocator, ial.info),
8141 if len(ial.result) != ial.required_nodes:
8142 raise errors.OpPrereqError("iallocator '%s' returned invalid number"
8143 " of nodes (%s), required %s" %
8144 (self.lu.op.iallocator, len(ial.result),
8145 ial.required_nodes), errors.ECODE_FAULT)
8146 self.target_node = ial.result[0]
8147 self.lu.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
8148 self.instance_name, self.lu.op.iallocator,
8149 utils.CommaJoin(ial.result))
8151 def _WaitUntilSync(self):
8152 """Poll with custom rpc for disk sync.
8154 This uses our own step-based rpc call.
8157 self.feedback_fn("* wait until resync is done")
8161 result = self.rpc.call_drbd_wait_sync(self.all_nodes,
8163 (self.instance.disks,
8166 for node, nres in result.items():
8167 nres.Raise("Cannot resync disks on node %s" % node)
8168 node_done, node_percent = nres.payload
8169 all_done = all_done and node_done
8170 if node_percent is not None:
8171 min_percent = min(min_percent, node_percent)
8173 if min_percent < 100:
8174 self.feedback_fn(" - progress: %.1f%%" % min_percent)
8177 def _EnsureSecondary(self, node):
8178 """Demote a node to secondary.
8181 self.feedback_fn("* switching node %s to secondary mode" % node)
8183 for dev in self.instance.disks:
8184 self.cfg.SetDiskID(dev, node)
8186 result = self.rpc.call_blockdev_close(node, self.instance.name,
8187 self.instance.disks)
8188 result.Raise("Cannot change disk to secondary on node %s" % node)
8190 def _GoStandalone(self):
8191 """Disconnect from the network.
8194 self.feedback_fn("* changing into standalone mode")
8195 result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
8196 self.instance.disks)
8197 for node, nres in result.items():
8198 nres.Raise("Cannot disconnect disks node %s" % node)
8200 def _GoReconnect(self, multimaster):
8201 """Reconnect to the network.
8207 msg = "single-master"
8208 self.feedback_fn("* changing disks into %s mode" % msg)
8209 result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
8210 (self.instance.disks, self.instance),
8211 self.instance.name, multimaster)
8212 for node, nres in result.items():
8213 nres.Raise("Cannot change disks config on node %s" % node)
8215 def _ExecCleanup(self):
8216 """Try to cleanup after a failed migration.
8218 The cleanup is done by:
8219 - check that the instance is running only on one node
8220 (and update the config if needed)
8221 - change disks on its secondary node to secondary
8222 - wait until disks are fully synchronized
8223 - disconnect from the network
8224 - change disks into single-master mode
8225 - wait again until disks are fully synchronized
8228 instance = self.instance
8229 target_node = self.target_node
8230 source_node = self.source_node
8232 # check running on only one node
8233 self.feedback_fn("* checking where the instance actually runs"
8234 " (if this hangs, the hypervisor might be in"
8236 ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
8237 for node, result in ins_l.items():
8238 result.Raise("Can't contact node %s" % node)
8240 runningon_source = instance.name in ins_l[source_node].payload
8241 runningon_target = instance.name in ins_l[target_node].payload
8243 if runningon_source and runningon_target:
8244 raise errors.OpExecError("Instance seems to be running on two nodes,"
8245 " or the hypervisor is confused; you will have"
8246 " to ensure manually that it runs only on one"
8247 " and restart this operation")
8249 if not (runningon_source or runningon_target):
8250 raise errors.OpExecError("Instance does not seem to be running at all;"
8251 " in this case it's safer to repair by"
8252 " running 'gnt-instance stop' to ensure disk"
8253 " shutdown, and then restarting it")
8255 if runningon_target:
8256 # the migration has actually succeeded, we need to update the config
8257 self.feedback_fn("* instance running on secondary node (%s),"
8258 " updating config" % target_node)
8259 instance.primary_node = target_node
8260 self.cfg.Update(instance, self.feedback_fn)
8261 demoted_node = source_node
8263 self.feedback_fn("* instance confirmed to be running on its"
8264 " primary node (%s)" % source_node)
8265 demoted_node = target_node
8267 if instance.disk_template in constants.DTS_INT_MIRROR:
8268 self._EnsureSecondary(demoted_node)
8270 self._WaitUntilSync()
8271 except errors.OpExecError:
8272 # we ignore here errors, since if the device is standalone, it
8273 # won't be able to sync
8275 self._GoStandalone()
8276 self._GoReconnect(False)
8277 self._WaitUntilSync()
8279 self.feedback_fn("* done")
8281 def _RevertDiskStatus(self):
8282 """Try to revert the disk status after a failed migration.
8285 target_node = self.target_node
8286 if self.instance.disk_template in constants.DTS_EXT_MIRROR:
8290 self._EnsureSecondary(target_node)
8291 self._GoStandalone()
8292 self._GoReconnect(False)
8293 self._WaitUntilSync()
8294 except errors.OpExecError, err:
8295 self.lu.LogWarning("Migration failed and I can't reconnect the drives,"
8296 " please try to recover the instance manually;"
8297 " error '%s'" % str(err))
8299 def _AbortMigration(self):
8300 """Call the hypervisor code to abort a started migration.
8303 instance = self.instance
8304 target_node = self.target_node
8305 source_node = self.source_node
8306 migration_info = self.migration_info
8308 abort_result = self.rpc.call_instance_finalize_migration_dst(target_node,
8312 abort_msg = abort_result.fail_msg
8314 logging.error("Aborting migration failed on target node %s: %s",
8315 target_node, abort_msg)
8316 # Don't raise an exception here, as we stil have to try to revert the
8317 # disk status, even if this step failed.
8319 abort_result = self.rpc.call_instance_finalize_migration_src(source_node,
8320 instance, False, self.live)
8321 abort_msg = abort_result.fail_msg
8323 logging.error("Aborting migration failed on source node %s: %s",
8324 source_node, abort_msg)
8326 def _ExecMigration(self):
8327 """Migrate an instance.
8329 The migrate is done by:
8330 - change the disks into dual-master mode
8331 - wait until disks are fully synchronized again
8332 - migrate the instance
8333 - change disks on the new secondary node (the old primary) to secondary
8334 - wait until disks are fully synchronized
8335 - change disks into single-master mode
8338 instance = self.instance
8339 target_node = self.target_node
8340 source_node = self.source_node
8342 # Check for hypervisor version mismatch and warn the user.
8343 nodeinfo = self.rpc.call_node_info([source_node, target_node],
8344 None, [self.instance.hypervisor])
8345 for ninfo in nodeinfo.values():
8346 ninfo.Raise("Unable to retrieve node information from node '%s'" %
8348 (_, _, (src_info, )) = nodeinfo[source_node].payload
8349 (_, _, (dst_info, )) = nodeinfo[target_node].payload
8351 if ((constants.HV_NODEINFO_KEY_VERSION in src_info) and
8352 (constants.HV_NODEINFO_KEY_VERSION in dst_info)):
8353 src_version = src_info[constants.HV_NODEINFO_KEY_VERSION]
8354 dst_version = dst_info[constants.HV_NODEINFO_KEY_VERSION]
8355 if src_version != dst_version:
8356 self.feedback_fn("* warning: hypervisor version mismatch between"
8357 " source (%s) and target (%s) node" %
8358 (src_version, dst_version))
8360 self.feedback_fn("* checking disk consistency between source and target")
8361 for (idx, dev) in enumerate(instance.disks):
8362 if not _CheckDiskConsistency(self.lu, instance, dev, target_node, False):
8363 raise errors.OpExecError("Disk %s is degraded or not fully"
8364 " synchronized on target node,"
8365 " aborting migration" % idx)
8367 if self.current_mem > self.tgt_free_mem:
8368 if not self.allow_runtime_changes:
8369 raise errors.OpExecError("Memory ballooning not allowed and not enough"
8370 " free memory to fit instance %s on target"
8371 " node %s (have %dMB, need %dMB)" %
8372 (instance.name, target_node,
8373 self.tgt_free_mem, self.current_mem))
8374 self.feedback_fn("* setting instance memory to %s" % self.tgt_free_mem)
8375 rpcres = self.rpc.call_instance_balloon_memory(instance.primary_node,
8378 rpcres.Raise("Cannot modify instance runtime memory")
8380 # First get the migration information from the remote node
8381 result = self.rpc.call_migration_info(source_node, instance)
8382 msg = result.fail_msg
8384 log_err = ("Failed fetching source migration information from %s: %s" %
8386 logging.error(log_err)
8387 raise errors.OpExecError(log_err)
8389 self.migration_info = migration_info = result.payload
8391 if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
8392 # Then switch the disks to master/master mode
8393 self._EnsureSecondary(target_node)
8394 self._GoStandalone()
8395 self._GoReconnect(True)
8396 self._WaitUntilSync()
8398 self.feedback_fn("* preparing %s to accept the instance" % target_node)
8399 result = self.rpc.call_accept_instance(target_node,
8402 self.nodes_ip[target_node])
8404 msg = result.fail_msg
8406 logging.error("Instance pre-migration failed, trying to revert"
8407 " disk status: %s", msg)
8408 self.feedback_fn("Pre-migration failed, aborting")
8409 self._AbortMigration()
8410 self._RevertDiskStatus()
8411 raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
8412 (instance.name, msg))
8414 self.feedback_fn("* migrating instance to %s" % target_node)
8415 result = self.rpc.call_instance_migrate(source_node, instance,
8416 self.nodes_ip[target_node],
8418 msg = result.fail_msg
8420 logging.error("Instance migration failed, trying to revert"
8421 " disk status: %s", msg)
8422 self.feedback_fn("Migration failed, aborting")
8423 self._AbortMigration()
8424 self._RevertDiskStatus()
8425 raise errors.OpExecError("Could not migrate instance %s: %s" %
8426 (instance.name, msg))
8428 self.feedback_fn("* starting memory transfer")
8429 last_feedback = time.time()
8431 result = self.rpc.call_instance_get_migration_status(source_node,
8433 msg = result.fail_msg
8434 ms = result.payload # MigrationStatus instance
8435 if msg or (ms.status in constants.HV_MIGRATION_FAILED_STATUSES):
8436 logging.error("Instance migration failed, trying to revert"
8437 " disk status: %s", msg)
8438 self.feedback_fn("Migration failed, aborting")
8439 self._AbortMigration()
8440 self._RevertDiskStatus()
8441 raise errors.OpExecError("Could not migrate instance %s: %s" %
8442 (instance.name, msg))
8444 if result.payload.status != constants.HV_MIGRATION_ACTIVE:
8445 self.feedback_fn("* memory transfer complete")
8448 if (utils.TimeoutExpired(last_feedback,
8449 self._MIGRATION_FEEDBACK_INTERVAL) and
8450 ms.transferred_ram is not None):
8451 mem_progress = 100 * float(ms.transferred_ram) / float(ms.total_ram)
8452 self.feedback_fn("* memory transfer progress: %.2f %%" % mem_progress)
8453 last_feedback = time.time()
8455 time.sleep(self._MIGRATION_POLL_INTERVAL)
8457 result = self.rpc.call_instance_finalize_migration_src(source_node,
8461 msg = result.fail_msg
8463 logging.error("Instance migration succeeded, but finalization failed"
8464 " on the source node: %s", msg)
8465 raise errors.OpExecError("Could not finalize instance migration: %s" %
8468 instance.primary_node = target_node
8470 # distribute new instance config to the other nodes
8471 self.cfg.Update(instance, self.feedback_fn)
8473 result = self.rpc.call_instance_finalize_migration_dst(target_node,
8477 msg = result.fail_msg
8479 logging.error("Instance migration succeeded, but finalization failed"
8480 " on the target node: %s", msg)
8481 raise errors.OpExecError("Could not finalize instance migration: %s" %
8484 if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
8485 self._EnsureSecondary(source_node)
8486 self._WaitUntilSync()
8487 self._GoStandalone()
8488 self._GoReconnect(False)
8489 self._WaitUntilSync()
8491 # If the instance's disk template is `rbd' and there was a successful
8492 # migration, unmap the device from the source node.
8493 if self.instance.disk_template == constants.DT_RBD:
8494 disks = _ExpandCheckDisks(instance, instance.disks)
8495 self.feedback_fn("* unmapping instance's disks from %s" % source_node)
8497 result = self.rpc.call_blockdev_shutdown(source_node, disk)
8498 msg = result.fail_msg
8500 logging.error("Migration was successful, but couldn't unmap the"
8501 " block device %s on source node %s: %s",
8502 disk.iv_name, source_node, msg)
8503 logging.error("You need to unmap the device %s manually on %s",
8504 disk.iv_name, source_node)
8506 self.feedback_fn("* done")
8508 def _ExecFailover(self):
8509 """Failover an instance.
8511 The failover is done by shutting it down on its present node and
8512 starting it on the secondary.
8515 instance = self.instance
8516 primary_node = self.cfg.GetNodeInfo(instance.primary_node)
8518 source_node = instance.primary_node
8519 target_node = self.target_node
8521 if instance.admin_state == constants.ADMINST_UP:
8522 self.feedback_fn("* checking disk consistency between source and target")
8523 for (idx, dev) in enumerate(instance.disks):
8524 # for drbd, these are drbd over lvm
8525 if not _CheckDiskConsistency(self.lu, instance, dev, target_node,
8527 if primary_node.offline:
8528 self.feedback_fn("Node %s is offline, ignoring degraded disk %s on"
8530 (primary_node.name, idx, target_node))
8531 elif not self.ignore_consistency:
8532 raise errors.OpExecError("Disk %s is degraded on target node,"
8533 " aborting failover" % idx)
8535 self.feedback_fn("* not checking disk consistency as instance is not"
8538 self.feedback_fn("* shutting down instance on source node")
8539 logging.info("Shutting down instance %s on node %s",
8540 instance.name, source_node)
8542 result = self.rpc.call_instance_shutdown(source_node, instance,
8543 self.shutdown_timeout)
8544 msg = result.fail_msg
8546 if self.ignore_consistency or primary_node.offline:
8547 self.lu.LogWarning("Could not shutdown instance %s on node %s,"
8548 " proceeding anyway; please make sure node"
8549 " %s is down; error details: %s",
8550 instance.name, source_node, source_node, msg)
8552 raise errors.OpExecError("Could not shutdown instance %s on"
8554 (instance.name, source_node, msg))
8556 self.feedback_fn("* deactivating the instance's disks on source node")
8557 if not _ShutdownInstanceDisks(self.lu, instance, ignore_primary=True):
8558 raise errors.OpExecError("Can't shut down the instance's disks")
8560 instance.primary_node = target_node
8561 # distribute new instance config to the other nodes
8562 self.cfg.Update(instance, self.feedback_fn)
8564 # Only start the instance if it's marked as up
8565 if instance.admin_state == constants.ADMINST_UP:
8566 self.feedback_fn("* activating the instance's disks on target node %s" %
8568 logging.info("Starting instance %s on node %s",
8569 instance.name, target_node)
8571 disks_ok, _ = _AssembleInstanceDisks(self.lu, instance,
8572 ignore_secondaries=True)
8574 _ShutdownInstanceDisks(self.lu, instance)
8575 raise errors.OpExecError("Can't activate the instance's disks")
8577 self.feedback_fn("* starting the instance on the target node %s" %
8579 result = self.rpc.call_instance_start(target_node, (instance, None, None),
8581 msg = result.fail_msg
8583 _ShutdownInstanceDisks(self.lu, instance)
8584 raise errors.OpExecError("Could not start instance %s on node %s: %s" %
8585 (instance.name, target_node, msg))
8587 def Exec(self, feedback_fn):
8588 """Perform the migration.
8591 self.feedback_fn = feedback_fn
8592 self.source_node = self.instance.primary_node
8594 # FIXME: if we implement migrate-to-any in DRBD, this needs fixing
8595 if self.instance.disk_template in constants.DTS_INT_MIRROR:
8596 self.target_node = self.instance.secondary_nodes[0]
8597 # Otherwise self.target_node has been populated either
8598 # directly, or through an iallocator.
8600 self.all_nodes = [self.source_node, self.target_node]
8601 self.nodes_ip = dict((name, node.secondary_ip) for (name, node)
8602 in self.cfg.GetMultiNodeInfo(self.all_nodes))
8605 feedback_fn("Failover instance %s" % self.instance.name)
8606 self._ExecFailover()
8608 feedback_fn("Migrating instance %s" % self.instance.name)
8611 return self._ExecCleanup()
8613 return self._ExecMigration()
8616 def _CreateBlockDev(lu, node, instance, device, force_create, info,
8618 """Wrapper around L{_CreateBlockDevInner}.
8620 This method annotates the root device first.
8623 (disk,) = _AnnotateDiskParams(instance, [device], lu.cfg)
8624 return _CreateBlockDevInner(lu, node, instance, disk, force_create, info,
8628 def _CreateBlockDevInner(lu, node, instance, device, force_create,
8630 """Create a tree of block devices on a given node.
8632 If this device type has to be created on secondaries, create it and
8635 If not, just recurse to children keeping the same 'force' value.
8637 @attention: The device has to be annotated already.
8639 @param lu: the lu on whose behalf we execute
8640 @param node: the node on which to create the device
8641 @type instance: L{objects.Instance}
8642 @param instance: the instance which owns the device
8643 @type device: L{objects.Disk}
8644 @param device: the device to create
8645 @type force_create: boolean
8646 @param force_create: whether to force creation of this device; this
8647 will be change to True whenever we find a device which has
8648 CreateOnSecondary() attribute
8649 @param info: the extra 'metadata' we should attach to the device
8650 (this will be represented as a LVM tag)
8651 @type force_open: boolean
8652 @param force_open: this parameter will be passes to the
8653 L{backend.BlockdevCreate} function where it specifies
8654 whether we run on primary or not, and it affects both
8655 the child assembly and the device own Open() execution
8658 if device.CreateOnSecondary():
8662 for child in device.children:
8663 _CreateBlockDevInner(lu, node, instance, child, force_create,
8666 if not force_create:
8669 _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
8672 def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
8673 """Create a single block device on a given node.
8675 This will not recurse over children of the device, so they must be
8678 @param lu: the lu on whose behalf we execute
8679 @param node: the node on which to create the device
8680 @type instance: L{objects.Instance}
8681 @param instance: the instance which owns the device
8682 @type device: L{objects.Disk}
8683 @param device: the device to create
8684 @param info: the extra 'metadata' we should attach to the device
8685 (this will be represented as a LVM tag)
8686 @type force_open: boolean
8687 @param force_open: this parameter will be passes to the
8688 L{backend.BlockdevCreate} function where it specifies
8689 whether we run on primary or not, and it affects both
8690 the child assembly and the device own Open() execution
8693 lu.cfg.SetDiskID(device, node)
8694 result = lu.rpc.call_blockdev_create(node, device, device.size,
8695 instance.name, force_open, info)
8696 result.Raise("Can't create block device %s on"
8697 " node %s for instance %s" % (device, node, instance.name))
8698 if device.physical_id is None:
8699 device.physical_id = result.payload
8702 def _GenerateUniqueNames(lu, exts):
8703 """Generate a suitable LV name.
8705 This will generate a logical volume name for the given instance.
8710 new_id = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
8711 results.append("%s%s" % (new_id, val))
8715 def _GenerateDRBD8Branch(lu, primary, secondary, size, vgnames, names,
8716 iv_name, p_minor, s_minor):
8717 """Generate a drbd8 device complete with its children.
8720 assert len(vgnames) == len(names) == 2
8721 port = lu.cfg.AllocatePort()
8722 shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
8724 dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
8725 logical_id=(vgnames[0], names[0]),
8727 dev_meta = objects.Disk(dev_type=constants.LD_LV, size=DRBD_META_SIZE,
8728 logical_id=(vgnames[1], names[1]),
8730 drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
8731 logical_id=(primary, secondary, port,
8734 children=[dev_data, dev_meta],
8735 iv_name=iv_name, params={})
8739 _DISK_TEMPLATE_NAME_PREFIX = {
8740 constants.DT_PLAIN: "",
8741 constants.DT_RBD: ".rbd",
8745 _DISK_TEMPLATE_DEVICE_TYPE = {
8746 constants.DT_PLAIN: constants.LD_LV,
8747 constants.DT_FILE: constants.LD_FILE,
8748 constants.DT_SHARED_FILE: constants.LD_FILE,
8749 constants.DT_BLOCK: constants.LD_BLOCKDEV,
8750 constants.DT_RBD: constants.LD_RBD,
8754 def _GenerateDiskTemplate(lu, template_name, instance_name, primary_node,
8755 secondary_nodes, disk_info, file_storage_dir, file_driver, base_index,
8756 feedback_fn, full_disk_params, _req_file_storage=opcodes.RequireFileStorage,
8757 _req_shr_file_storage=opcodes.RequireSharedFileStorage):
8758 """Generate the entire disk layout for a given template type.
8761 #TODO: compute space requirements
8763 vgname = lu.cfg.GetVGName()
8764 disk_count = len(disk_info)
8767 if template_name == constants.DT_DISKLESS:
8769 elif template_name == constants.DT_DRBD8:
8770 if len(secondary_nodes) != 1:
8771 raise errors.ProgrammerError("Wrong template configuration")
8772 remote_node = secondary_nodes[0]
8773 minors = lu.cfg.AllocateDRBDMinor(
8774 [primary_node, remote_node] * len(disk_info), instance_name)
8776 (drbd_params, _, _) = objects.Disk.ComputeLDParams(template_name,
8778 drbd_default_metavg = drbd_params[constants.LDP_DEFAULT_METAVG]
8781 for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
8782 for i in range(disk_count)]):
8783 names.append(lv_prefix + "_data")
8784 names.append(lv_prefix + "_meta")
8785 for idx, disk in enumerate(disk_info):
8786 disk_index = idx + base_index
8787 data_vg = disk.get(constants.IDISK_VG, vgname)
8788 meta_vg = disk.get(constants.IDISK_METAVG, drbd_default_metavg)
8789 disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
8790 disk[constants.IDISK_SIZE],
8792 names[idx * 2:idx * 2 + 2],
8793 "disk/%d" % disk_index,
8794 minors[idx * 2], minors[idx * 2 + 1])
8795 disk_dev.mode = disk[constants.IDISK_MODE]
8796 disks.append(disk_dev)
8799 raise errors.ProgrammerError("Wrong template configuration")
8801 if template_name == constants.DT_FILE:
8803 elif template_name == constants.DT_SHARED_FILE:
8804 _req_shr_file_storage()
8806 name_prefix = _DISK_TEMPLATE_NAME_PREFIX.get(template_name, None)
8807 if name_prefix is None:
8810 names = _GenerateUniqueNames(lu, ["%s.disk%s" %
8811 (name_prefix, base_index + i)
8812 for i in range(disk_count)])
8814 if template_name == constants.DT_PLAIN:
8815 def logical_id_fn(idx, _, disk):
8816 vg = disk.get(constants.IDISK_VG, vgname)
8817 return (vg, names[idx])
8818 elif template_name in (constants.DT_FILE, constants.DT_SHARED_FILE):
8820 lambda _, disk_index, disk: (file_driver,
8821 "%s/disk%d" % (file_storage_dir,
8823 elif template_name == constants.DT_BLOCK:
8825 lambda idx, disk_index, disk: (constants.BLOCKDEV_DRIVER_MANUAL,
8826 disk[constants.IDISK_ADOPT])
8827 elif template_name == constants.DT_RBD:
8828 logical_id_fn = lambda idx, _, disk: ("rbd", names[idx])
8830 raise errors.ProgrammerError("Unknown disk template '%s'" % template_name)
8832 dev_type = _DISK_TEMPLATE_DEVICE_TYPE[template_name]
8834 for idx, disk in enumerate(disk_info):
8835 disk_index = idx + base_index
8836 size = disk[constants.IDISK_SIZE]
8837 feedback_fn("* disk %s, size %s" %
8838 (disk_index, utils.FormatUnit(size, "h")))
8839 disks.append(objects.Disk(dev_type=dev_type, size=size,
8840 logical_id=logical_id_fn(idx, disk_index, disk),
8841 iv_name="disk/%d" % disk_index,
8842 mode=disk[constants.IDISK_MODE],
8848 def _GetInstanceInfoText(instance):
8849 """Compute that text that should be added to the disk's metadata.
8852 return "originstname+%s" % instance.name
8855 def _CalcEta(time_taken, written, total_size):
8856 """Calculates the ETA based on size written and total size.
8858 @param time_taken: The time taken so far
8859 @param written: amount written so far
8860 @param total_size: The total size of data to be written
8861 @return: The remaining time in seconds
8864 avg_time = time_taken / float(written)
8865 return (total_size - written) * avg_time
8868 def _WipeDisks(lu, instance):
8869 """Wipes instance disks.
8871 @type lu: L{LogicalUnit}
8872 @param lu: the logical unit on whose behalf we execute
8873 @type instance: L{objects.Instance}
8874 @param instance: the instance whose disks we should create
8875 @return: the success of the wipe
8878 node = instance.primary_node
8880 for device in instance.disks:
8881 lu.cfg.SetDiskID(device, node)
8883 logging.info("Pause sync of instance %s disks", instance.name)
8884 result = lu.rpc.call_blockdev_pause_resume_sync(node,
8885 (instance.disks, instance),
8888 for idx, success in enumerate(result.payload):
8890 logging.warn("pause-sync of instance %s for disks %d failed",
8894 for idx, device in enumerate(instance.disks):
8895 # The wipe size is MIN_WIPE_CHUNK_PERCENT % of the instance disk but
8896 # MAX_WIPE_CHUNK at max
8897 wipe_chunk_size = min(constants.MAX_WIPE_CHUNK, device.size / 100.0 *
8898 constants.MIN_WIPE_CHUNK_PERCENT)
8899 # we _must_ make this an int, otherwise rounding errors will
8901 wipe_chunk_size = int(wipe_chunk_size)
8903 lu.LogInfo("* Wiping disk %d", idx)
8904 logging.info("Wiping disk %d for instance %s, node %s using"
8905 " chunk size %s", idx, instance.name, node, wipe_chunk_size)
8910 start_time = time.time()
8912 while offset < size:
8913 wipe_size = min(wipe_chunk_size, size - offset)
8914 logging.debug("Wiping disk %d, offset %s, chunk %s",
8915 idx, offset, wipe_size)
8916 result = lu.rpc.call_blockdev_wipe(node, (device, instance), offset,
8918 result.Raise("Could not wipe disk %d at offset %d for size %d" %
8919 (idx, offset, wipe_size))
8922 if now - last_output >= 60:
8923 eta = _CalcEta(now - start_time, offset, size)
8924 lu.LogInfo(" - done: %.1f%% ETA: %s" %
8925 (offset / float(size) * 100, utils.FormatSeconds(eta)))
8928 logging.info("Resume sync of instance %s disks", instance.name)
8930 result = lu.rpc.call_blockdev_pause_resume_sync(node,
8931 (instance.disks, instance),
8934 for idx, success in enumerate(result.payload):
8936 lu.LogWarning("Resume sync of disk %d failed, please have a"
8937 " look at the status and troubleshoot the issue", idx)
8938 logging.warn("resume-sync of instance %s for disks %d failed",
8942 def _CreateDisks(lu, instance, to_skip=None, target_node=None):
8943 """Create all disks for an instance.
8945 This abstracts away some work from AddInstance.
8947 @type lu: L{LogicalUnit}
8948 @param lu: the logical unit on whose behalf we execute
8949 @type instance: L{objects.Instance}
8950 @param instance: the instance whose disks we should create
8952 @param to_skip: list of indices to skip
8953 @type target_node: string
8954 @param target_node: if passed, overrides the target node for creation
8956 @return: the success of the creation
8959 info = _GetInstanceInfoText(instance)
8960 if target_node is None:
8961 pnode = instance.primary_node
8962 all_nodes = instance.all_nodes
8967 if instance.disk_template in constants.DTS_FILEBASED:
8968 file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
8969 result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
8971 result.Raise("Failed to create directory '%s' on"
8972 " node %s" % (file_storage_dir, pnode))
8974 # Note: this needs to be kept in sync with adding of disks in
8975 # LUInstanceSetParams
8976 for idx, device in enumerate(instance.disks):
8977 if to_skip and idx in to_skip:
8979 logging.info("Creating disk %s for instance '%s'", idx, instance.name)
8981 for node in all_nodes:
8982 f_create = node == pnode
8983 _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
8986 def _RemoveDisks(lu, instance, target_node=None, ignore_failures=False):
8987 """Remove all disks for an instance.
8989 This abstracts away some work from `AddInstance()` and
8990 `RemoveInstance()`. Note that in case some of the devices couldn't
8991 be removed, the removal will continue with the other ones (compare
8992 with `_CreateDisks()`).
8994 @type lu: L{LogicalUnit}
8995 @param lu: the logical unit on whose behalf we execute
8996 @type instance: L{objects.Instance}
8997 @param instance: the instance whose disks we should remove
8998 @type target_node: string
8999 @param target_node: used to override the node on which to remove the disks
9001 @return: the success of the removal
9004 logging.info("Removing block devices for instance %s", instance.name)
9007 ports_to_release = set()
9008 for (idx, device) in enumerate(instance.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, dev).fail_msg
11165 self.lu.LogWarning("Failed to shutdown drbd for disk/%d on old"
11166 "node: %s" % (idx, msg),
11167 hint=("Please cleanup this device manually as"
11168 " soon as possible"))
11170 self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
11171 result = self.rpc.call_drbd_disconnect_net([pnode], self.node_secondary_ip,
11172 self.instance.disks)[pnode]
11174 msg = result.fail_msg
11176 # detaches didn't succeed (unlikely)
11177 self.cfg.ReleaseDRBDMinors(self.instance.name)
11178 raise errors.OpExecError("Can't detach the disks from the network on"
11179 " old node: %s" % (msg,))
11181 # if we managed to detach at least one, we update all the disks of
11182 # the instance to point to the new secondary
11183 self.lu.LogInfo("Updating instance configuration")
11184 for dev, _, new_logical_id in iv_names.itervalues():
11185 dev.logical_id = new_logical_id
11186 self.cfg.SetDiskID(dev, self.instance.primary_node)
11188 self.cfg.Update(self.instance, feedback_fn)
11190 # Release all node locks (the configuration has been updated)
11191 _ReleaseLocks(self.lu, locking.LEVEL_NODE)
11193 # and now perform the drbd attach
11194 self.lu.LogInfo("Attaching primary drbds to new secondary"
11195 " (standalone => connected)")
11196 result = self.rpc.call_drbd_attach_net([self.instance.primary_node,
11198 self.node_secondary_ip,
11199 (self.instance.disks, self.instance),
11200 self.instance.name,
11202 for to_node, to_result in result.items():
11203 msg = to_result.fail_msg
11205 self.lu.LogWarning("Can't attach drbd disks on node %s: %s",
11207 hint=("please do a gnt-instance info to see the"
11208 " status of disks"))
11210 cstep = itertools.count(5)
11212 if self.early_release:
11213 self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
11214 self._RemoveOldStorage(self.target_node, iv_names)
11215 # TODO: Check if releasing locks early still makes sense
11216 _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES)
11218 # Release all resource locks except those used by the instance
11219 _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES,
11220 keep=self.node_secondary_ip.keys())
11222 # TODO: Can the instance lock be downgraded here? Take the optional disk
11223 # shutdown in the caller into consideration.
11226 # This can fail as the old devices are degraded and _WaitForSync
11227 # does a combined result over all disks, so we don't check its return value
11228 self.lu.LogStep(cstep.next(), steps_total, "Sync devices")
11229 _WaitForSync(self.lu, self.instance)
11231 # Check all devices manually
11232 self._CheckDevices(self.instance.primary_node, iv_names)
11234 # Step: remove old storage
11235 if not self.early_release:
11236 self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
11237 self._RemoveOldStorage(self.target_node, iv_names)
11240 class LURepairNodeStorage(NoHooksLU):
11241 """Repairs the volume group on a node.
11246 def CheckArguments(self):
11247 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
11249 storage_type = self.op.storage_type
11251 if (constants.SO_FIX_CONSISTENCY not in
11252 constants.VALID_STORAGE_OPERATIONS.get(storage_type, [])):
11253 raise errors.OpPrereqError("Storage units of type '%s' can not be"
11254 " repaired" % storage_type,
11255 errors.ECODE_INVAL)
11257 def ExpandNames(self):
11258 self.needed_locks = {
11259 locking.LEVEL_NODE: [self.op.node_name],
11262 def _CheckFaultyDisks(self, instance, node_name):
11263 """Ensure faulty disks abort the opcode or at least warn."""
11265 if _FindFaultyInstanceDisks(self.cfg, self.rpc, instance,
11267 raise errors.OpPrereqError("Instance '%s' has faulty disks on"
11268 " node '%s'" % (instance.name, node_name),
11269 errors.ECODE_STATE)
11270 except errors.OpPrereqError, err:
11271 if self.op.ignore_consistency:
11272 self.proc.LogWarning(str(err.args[0]))
11276 def CheckPrereq(self):
11277 """Check prerequisites.
11280 # Check whether any instance on this node has faulty disks
11281 for inst in _GetNodeInstances(self.cfg, self.op.node_name):
11282 if inst.admin_state != constants.ADMINST_UP:
11284 check_nodes = set(inst.all_nodes)
11285 check_nodes.discard(self.op.node_name)
11286 for inst_node_name in check_nodes:
11287 self._CheckFaultyDisks(inst, inst_node_name)
11289 def Exec(self, feedback_fn):
11290 feedback_fn("Repairing storage unit '%s' on %s ..." %
11291 (self.op.name, self.op.node_name))
11293 st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
11294 result = self.rpc.call_storage_execute(self.op.node_name,
11295 self.op.storage_type, st_args,
11297 constants.SO_FIX_CONSISTENCY)
11298 result.Raise("Failed to repair storage unit '%s' on %s" %
11299 (self.op.name, self.op.node_name))
11302 class LUNodeEvacuate(NoHooksLU):
11303 """Evacuates instances off a list of nodes.
11308 _MODE2IALLOCATOR = {
11309 constants.NODE_EVAC_PRI: constants.IALLOCATOR_NEVAC_PRI,
11310 constants.NODE_EVAC_SEC: constants.IALLOCATOR_NEVAC_SEC,
11311 constants.NODE_EVAC_ALL: constants.IALLOCATOR_NEVAC_ALL,
11313 assert frozenset(_MODE2IALLOCATOR.keys()) == constants.NODE_EVAC_MODES
11314 assert (frozenset(_MODE2IALLOCATOR.values()) ==
11315 constants.IALLOCATOR_NEVAC_MODES)
11317 def CheckArguments(self):
11318 _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
11320 def ExpandNames(self):
11321 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
11323 if self.op.remote_node is not None:
11324 self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
11325 assert self.op.remote_node
11327 if self.op.remote_node == self.op.node_name:
11328 raise errors.OpPrereqError("Can not use evacuated node as a new"
11329 " secondary node", errors.ECODE_INVAL)
11331 if self.op.mode != constants.NODE_EVAC_SEC:
11332 raise errors.OpPrereqError("Without the use of an iallocator only"
11333 " secondary instances can be evacuated",
11334 errors.ECODE_INVAL)
11337 self.share_locks = _ShareAll()
11338 self.needed_locks = {
11339 locking.LEVEL_INSTANCE: [],
11340 locking.LEVEL_NODEGROUP: [],
11341 locking.LEVEL_NODE: [],
11344 # Determine nodes (via group) optimistically, needs verification once locks
11345 # have been acquired
11346 self.lock_nodes = self._DetermineNodes()
11348 def _DetermineNodes(self):
11349 """Gets the list of nodes to operate on.
11352 if self.op.remote_node is None:
11353 # Iallocator will choose any node(s) in the same group
11354 group_nodes = self.cfg.GetNodeGroupMembersByNodes([self.op.node_name])
11356 group_nodes = frozenset([self.op.remote_node])
11358 # Determine nodes to be locked
11359 return set([self.op.node_name]) | group_nodes
11361 def _DetermineInstances(self):
11362 """Builds list of instances to operate on.
11365 assert self.op.mode in constants.NODE_EVAC_MODES
11367 if self.op.mode == constants.NODE_EVAC_PRI:
11368 # Primary instances only
11369 inst_fn = _GetNodePrimaryInstances
11370 assert self.op.remote_node is None, \
11371 "Evacuating primary instances requires iallocator"
11372 elif self.op.mode == constants.NODE_EVAC_SEC:
11373 # Secondary instances only
11374 inst_fn = _GetNodeSecondaryInstances
11377 assert self.op.mode == constants.NODE_EVAC_ALL
11378 inst_fn = _GetNodeInstances
11379 # TODO: In 2.6, change the iallocator interface to take an evacuation mode
11381 raise errors.OpPrereqError("Due to an issue with the iallocator"
11382 " interface it is not possible to evacuate"
11383 " all instances at once; specify explicitly"
11384 " whether to evacuate primary or secondary"
11386 errors.ECODE_INVAL)
11388 return inst_fn(self.cfg, self.op.node_name)
11390 def DeclareLocks(self, level):
11391 if level == locking.LEVEL_INSTANCE:
11392 # Lock instances optimistically, needs verification once node and group
11393 # locks have been acquired
11394 self.needed_locks[locking.LEVEL_INSTANCE] = \
11395 set(i.name for i in self._DetermineInstances())
11397 elif level == locking.LEVEL_NODEGROUP:
11398 # Lock node groups for all potential target nodes optimistically, needs
11399 # verification once nodes have been acquired
11400 self.needed_locks[locking.LEVEL_NODEGROUP] = \
11401 self.cfg.GetNodeGroupsFromNodes(self.lock_nodes)
11403 elif level == locking.LEVEL_NODE:
11404 self.needed_locks[locking.LEVEL_NODE] = self.lock_nodes
11406 def CheckPrereq(self):
11408 owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
11409 owned_nodes = self.owned_locks(locking.LEVEL_NODE)
11410 owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
11412 need_nodes = self._DetermineNodes()
11414 if not owned_nodes.issuperset(need_nodes):
11415 raise errors.OpPrereqError("Nodes in same group as '%s' changed since"
11416 " locks were acquired, current nodes are"
11417 " are '%s', used to be '%s'; retry the"
11419 (self.op.node_name,
11420 utils.CommaJoin(need_nodes),
11421 utils.CommaJoin(owned_nodes)),
11422 errors.ECODE_STATE)
11424 wanted_groups = self.cfg.GetNodeGroupsFromNodes(owned_nodes)
11425 if owned_groups != wanted_groups:
11426 raise errors.OpExecError("Node groups changed since locks were acquired,"
11427 " current groups are '%s', used to be '%s';"
11428 " retry the operation" %
11429 (utils.CommaJoin(wanted_groups),
11430 utils.CommaJoin(owned_groups)))
11432 # Determine affected instances
11433 self.instances = self._DetermineInstances()
11434 self.instance_names = [i.name for i in self.instances]
11436 if set(self.instance_names) != owned_instances:
11437 raise errors.OpExecError("Instances on node '%s' changed since locks"
11438 " were acquired, current instances are '%s',"
11439 " used to be '%s'; retry the operation" %
11440 (self.op.node_name,
11441 utils.CommaJoin(self.instance_names),
11442 utils.CommaJoin(owned_instances)))
11444 if self.instance_names:
11445 self.LogInfo("Evacuating instances from node '%s': %s",
11447 utils.CommaJoin(utils.NiceSort(self.instance_names)))
11449 self.LogInfo("No instances to evacuate from node '%s'",
11452 if self.op.remote_node is not None:
11453 for i in self.instances:
11454 if i.primary_node == self.op.remote_node:
11455 raise errors.OpPrereqError("Node %s is the primary node of"
11456 " instance %s, cannot use it as"
11458 (self.op.remote_node, i.name),
11459 errors.ECODE_INVAL)
11461 def Exec(self, feedback_fn):
11462 assert (self.op.iallocator is not None) ^ (self.op.remote_node is not None)
11464 if not self.instance_names:
11465 # No instances to evacuate
11468 elif self.op.iallocator is not None:
11469 # TODO: Implement relocation to other group
11470 ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_NODE_EVAC,
11471 evac_mode=self._MODE2IALLOCATOR[self.op.mode],
11472 instances=list(self.instance_names))
11474 ial.Run(self.op.iallocator)
11476 if not ial.success:
11477 raise errors.OpPrereqError("Can't compute node evacuation using"
11478 " iallocator '%s': %s" %
11479 (self.op.iallocator, ial.info),
11480 errors.ECODE_NORES)
11482 jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, True)
11484 elif self.op.remote_node is not None:
11485 assert self.op.mode == constants.NODE_EVAC_SEC
11487 [opcodes.OpInstanceReplaceDisks(instance_name=instance_name,
11488 remote_node=self.op.remote_node,
11490 mode=constants.REPLACE_DISK_CHG,
11491 early_release=self.op.early_release)]
11492 for instance_name in self.instance_names
11496 raise errors.ProgrammerError("No iallocator or remote node")
11498 return ResultWithJobs(jobs)
11501 def _SetOpEarlyRelease(early_release, op):
11502 """Sets C{early_release} flag on opcodes if available.
11506 op.early_release = early_release
11507 except AttributeError:
11508 assert not isinstance(op, opcodes.OpInstanceReplaceDisks)
11513 def _NodeEvacDest(use_nodes, group, nodes):
11514 """Returns group or nodes depending on caller's choice.
11518 return utils.CommaJoin(nodes)
11523 def _LoadNodeEvacResult(lu, alloc_result, early_release, use_nodes):
11524 """Unpacks the result of change-group and node-evacuate iallocator requests.
11526 Iallocator modes L{constants.IALLOCATOR_MODE_NODE_EVAC} and
11527 L{constants.IALLOCATOR_MODE_CHG_GROUP}.
11529 @type lu: L{LogicalUnit}
11530 @param lu: Logical unit instance
11531 @type alloc_result: tuple/list
11532 @param alloc_result: Result from iallocator
11533 @type early_release: bool
11534 @param early_release: Whether to release locks early if possible
11535 @type use_nodes: bool
11536 @param use_nodes: Whether to display node names instead of groups
11539 (moved, failed, jobs) = alloc_result
11542 failreason = utils.CommaJoin("%s (%s)" % (name, reason)
11543 for (name, reason) in failed)
11544 lu.LogWarning("Unable to evacuate instances %s", failreason)
11545 raise errors.OpExecError("Unable to evacuate instances %s" % failreason)
11548 lu.LogInfo("Instances to be moved: %s",
11549 utils.CommaJoin("%s (to %s)" %
11550 (name, _NodeEvacDest(use_nodes, group, nodes))
11551 for (name, group, nodes) in moved))
11553 return [map(compat.partial(_SetOpEarlyRelease, early_release),
11554 map(opcodes.OpCode.LoadOpCode, ops))
11558 class LUInstanceGrowDisk(LogicalUnit):
11559 """Grow a disk of an instance.
11562 HPATH = "disk-grow"
11563 HTYPE = constants.HTYPE_INSTANCE
11566 def ExpandNames(self):
11567 self._ExpandAndLockInstance()
11568 self.needed_locks[locking.LEVEL_NODE] = []
11569 self.needed_locks[locking.LEVEL_NODE_RES] = []
11570 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
11571 self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
11573 def DeclareLocks(self, level):
11574 if level == locking.LEVEL_NODE:
11575 self._LockInstancesNodes()
11576 elif level == locking.LEVEL_NODE_RES:
11578 self.needed_locks[locking.LEVEL_NODE_RES] = \
11579 self.needed_locks[locking.LEVEL_NODE][:]
11581 def BuildHooksEnv(self):
11582 """Build hooks env.
11584 This runs on the master, the primary and all the secondaries.
11588 "DISK": self.op.disk,
11589 "AMOUNT": self.op.amount,
11590 "ABSOLUTE": self.op.absolute,
11592 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
11595 def BuildHooksNodes(self):
11596 """Build hooks nodes.
11599 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
11602 def CheckPrereq(self):
11603 """Check prerequisites.
11605 This checks that the instance is in the cluster.
11608 instance = self.cfg.GetInstanceInfo(self.op.instance_name)
11609 assert instance is not None, \
11610 "Cannot retrieve locked instance %s" % self.op.instance_name
11611 nodenames = list(instance.all_nodes)
11612 for node in nodenames:
11613 _CheckNodeOnline(self, node)
11615 self.instance = instance
11617 if instance.disk_template not in constants.DTS_GROWABLE:
11618 raise errors.OpPrereqError("Instance's disk layout does not support"
11619 " growing", errors.ECODE_INVAL)
11621 self.disk = instance.FindDisk(self.op.disk)
11623 if self.op.absolute:
11624 self.target = self.op.amount
11625 self.delta = self.target - self.disk.size
11627 raise errors.OpPrereqError("Requested size (%s) is smaller than "
11628 "current disk size (%s)" %
11629 (utils.FormatUnit(self.target, "h"),
11630 utils.FormatUnit(self.disk.size, "h")),
11631 errors.ECODE_STATE)
11633 self.delta = self.op.amount
11634 self.target = self.disk.size + self.delta
11636 raise errors.OpPrereqError("Requested increment (%s) is negative" %
11637 utils.FormatUnit(self.delta, "h"),
11638 errors.ECODE_INVAL)
11640 if instance.disk_template not in (constants.DT_FILE,
11641 constants.DT_SHARED_FILE,
11643 # TODO: check the free disk space for file, when that feature will be
11645 _CheckNodesFreeDiskPerVG(self, nodenames,
11646 self.disk.ComputeGrowth(self.delta))
11648 def Exec(self, feedback_fn):
11649 """Execute disk grow.
11652 instance = self.instance
11655 assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
11656 assert (self.owned_locks(locking.LEVEL_NODE) ==
11657 self.owned_locks(locking.LEVEL_NODE_RES))
11659 disks_ok, _ = _AssembleInstanceDisks(self, self.instance, disks=[disk])
11661 raise errors.OpExecError("Cannot activate block device to grow")
11663 feedback_fn("Growing disk %s of instance '%s' by %s to %s" %
11664 (self.op.disk, instance.name,
11665 utils.FormatUnit(self.delta, "h"),
11666 utils.FormatUnit(self.target, "h")))
11668 # First run all grow ops in dry-run mode
11669 for node in instance.all_nodes:
11670 self.cfg.SetDiskID(disk, node)
11671 result = self.rpc.call_blockdev_grow(node, (disk, instance), self.delta,
11673 result.Raise("Grow request failed to node %s" % node)
11675 # We know that (as far as we can test) operations across different
11676 # nodes will succeed, time to run it for real
11677 for node in instance.all_nodes:
11678 self.cfg.SetDiskID(disk, node)
11679 result = self.rpc.call_blockdev_grow(node, (disk, instance), self.delta,
11681 result.Raise("Grow request failed to node %s" % node)
11683 # TODO: Rewrite code to work properly
11684 # DRBD goes into sync mode for a short amount of time after executing the
11685 # "resize" command. DRBD 8.x below version 8.0.13 contains a bug whereby
11686 # calling "resize" in sync mode fails. Sleeping for a short amount of
11687 # time is a work-around.
11690 disk.RecordGrow(self.delta)
11691 self.cfg.Update(instance, feedback_fn)
11693 # Changes have been recorded, release node lock
11694 _ReleaseLocks(self, locking.LEVEL_NODE)
11696 # Downgrade lock while waiting for sync
11697 self.glm.downgrade(locking.LEVEL_INSTANCE)
11699 if self.op.wait_for_sync:
11700 disk_abort = not _WaitForSync(self, instance, disks=[disk])
11702 self.proc.LogWarning("Disk sync-ing has not returned a good"
11703 " status; please check the instance")
11704 if instance.admin_state != constants.ADMINST_UP:
11705 _SafeShutdownInstanceDisks(self, instance, disks=[disk])
11706 elif instance.admin_state != constants.ADMINST_UP:
11707 self.proc.LogWarning("Not shutting down the disk even if the instance is"
11708 " not supposed to be running because no wait for"
11709 " sync mode was requested")
11711 assert self.owned_locks(locking.LEVEL_NODE_RES)
11712 assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
11715 class LUInstanceQueryData(NoHooksLU):
11716 """Query runtime instance data.
11721 def ExpandNames(self):
11722 self.needed_locks = {}
11724 # Use locking if requested or when non-static information is wanted
11725 if not (self.op.static or self.op.use_locking):
11726 self.LogWarning("Non-static data requested, locks need to be acquired")
11727 self.op.use_locking = True
11729 if self.op.instances or not self.op.use_locking:
11730 # Expand instance names right here
11731 self.wanted_names = _GetWantedInstances(self, self.op.instances)
11733 # Will use acquired locks
11734 self.wanted_names = None
11736 if self.op.use_locking:
11737 self.share_locks = _ShareAll()
11739 if self.wanted_names is None:
11740 self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
11742 self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
11744 self.needed_locks[locking.LEVEL_NODEGROUP] = []
11745 self.needed_locks[locking.LEVEL_NODE] = []
11746 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
11748 def DeclareLocks(self, level):
11749 if self.op.use_locking:
11750 if level == locking.LEVEL_NODEGROUP:
11751 owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
11753 # Lock all groups used by instances optimistically; this requires going
11754 # via the node before it's locked, requiring verification later on
11755 self.needed_locks[locking.LEVEL_NODEGROUP] = \
11756 frozenset(group_uuid
11757 for instance_name in owned_instances
11759 self.cfg.GetInstanceNodeGroups(instance_name))
11761 elif level == locking.LEVEL_NODE:
11762 self._LockInstancesNodes()
11764 def CheckPrereq(self):
11765 """Check prerequisites.
11767 This only checks the optional instance list against the existing names.
11770 owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
11771 owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
11772 owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
11774 if self.wanted_names is None:
11775 assert self.op.use_locking, "Locking was not used"
11776 self.wanted_names = owned_instances
11778 instances = dict(self.cfg.GetMultiInstanceInfo(self.wanted_names))
11780 if self.op.use_locking:
11781 _CheckInstancesNodeGroups(self.cfg, instances, owned_groups, owned_nodes,
11784 assert not (owned_instances or owned_groups or owned_nodes)
11786 self.wanted_instances = instances.values()
11788 def _ComputeBlockdevStatus(self, node, instance, dev):
11789 """Returns the status of a block device
11792 if self.op.static or not node:
11795 self.cfg.SetDiskID(dev, node)
11797 result = self.rpc.call_blockdev_find(node, dev)
11801 result.Raise("Can't compute disk status for %s" % instance.name)
11803 status = result.payload
11807 return (status.dev_path, status.major, status.minor,
11808 status.sync_percent, status.estimated_time,
11809 status.is_degraded, status.ldisk_status)
11811 def _ComputeDiskStatus(self, instance, snode, dev):
11812 """Compute block device status.
11815 (anno_dev,) = _AnnotateDiskParams(instance, [dev], self.cfg)
11817 return self._ComputeDiskStatusInner(instance, snode, anno_dev)
11819 def _ComputeDiskStatusInner(self, instance, snode, dev):
11820 """Compute block device status.
11822 @attention: The device has to be annotated already.
11825 if dev.dev_type in constants.LDS_DRBD:
11826 # we change the snode then (otherwise we use the one passed in)
11827 if dev.logical_id[0] == instance.primary_node:
11828 snode = dev.logical_id[1]
11830 snode = dev.logical_id[0]
11832 dev_pstatus = self._ComputeBlockdevStatus(instance.primary_node,
11834 dev_sstatus = self._ComputeBlockdevStatus(snode, instance, dev)
11837 dev_children = map(compat.partial(self._ComputeDiskStatusInner,
11844 "iv_name": dev.iv_name,
11845 "dev_type": dev.dev_type,
11846 "logical_id": dev.logical_id,
11847 "physical_id": dev.physical_id,
11848 "pstatus": dev_pstatus,
11849 "sstatus": dev_sstatus,
11850 "children": dev_children,
11855 def Exec(self, feedback_fn):
11856 """Gather and return data"""
11859 cluster = self.cfg.GetClusterInfo()
11861 node_names = itertools.chain(*(i.all_nodes for i in self.wanted_instances))
11862 nodes = dict(self.cfg.GetMultiNodeInfo(node_names))
11864 groups = dict(self.cfg.GetMultiNodeGroupInfo(node.group
11865 for node in nodes.values()))
11867 group2name_fn = lambda uuid: groups[uuid].name
11869 for instance in self.wanted_instances:
11870 pnode = nodes[instance.primary_node]
11872 if self.op.static or pnode.offline:
11873 remote_state = None
11875 self.LogWarning("Primary node %s is marked offline, returning static"
11876 " information only for instance %s" %
11877 (pnode.name, instance.name))
11879 remote_info = self.rpc.call_instance_info(instance.primary_node,
11881 instance.hypervisor)
11882 remote_info.Raise("Error checking node %s" % instance.primary_node)
11883 remote_info = remote_info.payload
11884 if remote_info and "state" in remote_info:
11885 remote_state = "up"
11887 if instance.admin_state == constants.ADMINST_UP:
11888 remote_state = "down"
11890 remote_state = instance.admin_state
11892 disks = map(compat.partial(self._ComputeDiskStatus, instance, None),
11895 snodes_group_uuids = [nodes[snode_name].group
11896 for snode_name in instance.secondary_nodes]
11898 result[instance.name] = {
11899 "name": instance.name,
11900 "config_state": instance.admin_state,
11901 "run_state": remote_state,
11902 "pnode": instance.primary_node,
11903 "pnode_group_uuid": pnode.group,
11904 "pnode_group_name": group2name_fn(pnode.group),
11905 "snodes": instance.secondary_nodes,
11906 "snodes_group_uuids": snodes_group_uuids,
11907 "snodes_group_names": map(group2name_fn, snodes_group_uuids),
11909 # this happens to be the same format used for hooks
11910 "nics": _NICListToTuple(self, instance.nics),
11911 "disk_template": instance.disk_template,
11913 "hypervisor": instance.hypervisor,
11914 "network_port": instance.network_port,
11915 "hv_instance": instance.hvparams,
11916 "hv_actual": cluster.FillHV(instance, skip_globals=True),
11917 "be_instance": instance.beparams,
11918 "be_actual": cluster.FillBE(instance),
11919 "os_instance": instance.osparams,
11920 "os_actual": cluster.SimpleFillOS(instance.os, instance.osparams),
11921 "serial_no": instance.serial_no,
11922 "mtime": instance.mtime,
11923 "ctime": instance.ctime,
11924 "uuid": instance.uuid,
11930 def PrepareContainerMods(mods, private_fn):
11931 """Prepares a list of container modifications by adding a private data field.
11933 @type mods: list of tuples; (operation, index, parameters)
11934 @param mods: List of modifications
11935 @type private_fn: callable or None
11936 @param private_fn: Callable for constructing a private data field for a
11941 if private_fn is None:
11946 return [(op, idx, params, fn()) for (op, idx, params) in mods]
11949 #: Type description for changes as returned by L{ApplyContainerMods}'s
11951 _TApplyContModsCbChanges = \
11952 ht.TMaybeListOf(ht.TAnd(ht.TIsLength(2), ht.TItems([
11953 ht.TNonEmptyString,
11958 def ApplyContainerMods(kind, container, chgdesc, mods,
11959 create_fn, modify_fn, remove_fn):
11960 """Applies descriptions in C{mods} to C{container}.
11963 @param kind: One-word item description
11964 @type container: list
11965 @param container: Container to modify
11966 @type chgdesc: None or list
11967 @param chgdesc: List of applied changes
11969 @param mods: Modifications as returned by L{PrepareContainerMods}
11970 @type create_fn: callable
11971 @param create_fn: Callback for creating a new item (L{constants.DDM_ADD});
11972 receives absolute item index, parameters and private data object as added
11973 by L{PrepareContainerMods}, returns tuple containing new item and changes
11975 @type modify_fn: callable
11976 @param modify_fn: Callback for modifying an existing item
11977 (L{constants.DDM_MODIFY}); receives absolute item index, item, parameters
11978 and private data object as added by L{PrepareContainerMods}, returns
11980 @type remove_fn: callable
11981 @param remove_fn: Callback on removing item; receives absolute item index,
11982 item and private data object as added by L{PrepareContainerMods}
11985 for (op, idx, params, private) in mods:
11988 absidx = len(container) - 1
11990 raise IndexError("Not accepting negative indices other than -1")
11991 elif idx > len(container):
11992 raise IndexError("Got %s index %s, but there are only %s" %
11993 (kind, idx, len(container)))
11999 if op == constants.DDM_ADD:
12000 # Calculate where item will be added
12002 addidx = len(container)
12006 if create_fn is None:
12009 (item, changes) = create_fn(addidx, params, private)
12012 container.append(item)
12015 assert idx <= len(container)
12016 # list.insert does so before the specified index
12017 container.insert(idx, item)
12019 # Retrieve existing item
12021 item = container[absidx]
12023 raise IndexError("Invalid %s index %s" % (kind, idx))
12025 if op == constants.DDM_REMOVE:
12028 if remove_fn is not None:
12029 remove_fn(absidx, item, private)
12031 changes = [("%s/%s" % (kind, absidx), "remove")]
12033 assert container[absidx] == item
12034 del container[absidx]
12035 elif op == constants.DDM_MODIFY:
12036 if modify_fn is not None:
12037 changes = modify_fn(absidx, item, params, private)
12039 raise errors.ProgrammerError("Unhandled operation '%s'" % op)
12041 assert _TApplyContModsCbChanges(changes)
12043 if not (chgdesc is None or changes is None):
12044 chgdesc.extend(changes)
12047 def _UpdateIvNames(base_index, disks):
12048 """Updates the C{iv_name} attribute of disks.
12050 @type disks: list of L{objects.Disk}
12053 for (idx, disk) in enumerate(disks):
12054 disk.iv_name = "disk/%s" % (base_index + idx, )
12057 class _InstNicModPrivate:
12058 """Data structure for network interface modifications.
12060 Used by L{LUInstanceSetParams}.
12063 def __init__(self):
12068 class LUInstanceSetParams(LogicalUnit):
12069 """Modifies an instances's parameters.
12072 HPATH = "instance-modify"
12073 HTYPE = constants.HTYPE_INSTANCE
12077 def _UpgradeDiskNicMods(kind, mods, verify_fn):
12078 assert ht.TList(mods)
12079 assert not mods or len(mods[0]) in (2, 3)
12081 if mods and len(mods[0]) == 2:
12085 for op, params in mods:
12086 if op in (constants.DDM_ADD, constants.DDM_REMOVE):
12087 result.append((op, -1, params))
12091 raise errors.OpPrereqError("Only one %s add or remove operation is"
12092 " supported at a time" % kind,
12093 errors.ECODE_INVAL)
12095 result.append((constants.DDM_MODIFY, op, params))
12097 assert verify_fn(result)
12104 def _CheckMods(kind, mods, key_types, item_fn):
12105 """Ensures requested disk/NIC modifications are valid.
12108 for (op, _, params) in mods:
12109 assert ht.TDict(params)
12111 utils.ForceDictType(params, key_types)
12113 if op == constants.DDM_REMOVE:
12115 raise errors.OpPrereqError("No settings should be passed when"
12116 " removing a %s" % kind,
12117 errors.ECODE_INVAL)
12118 elif op in (constants.DDM_ADD, constants.DDM_MODIFY):
12119 item_fn(op, params)
12121 raise errors.ProgrammerError("Unhandled operation '%s'" % op)
12124 def _VerifyDiskModification(op, params):
12125 """Verifies a disk modification.
12128 if op == constants.DDM_ADD:
12129 mode = params.setdefault(constants.IDISK_MODE, constants.DISK_RDWR)
12130 if mode not in constants.DISK_ACCESS_SET:
12131 raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode,
12132 errors.ECODE_INVAL)
12134 size = params.get(constants.IDISK_SIZE, None)
12136 raise errors.OpPrereqError("Required disk parameter '%s' missing" %
12137 constants.IDISK_SIZE, errors.ECODE_INVAL)
12141 except (TypeError, ValueError), err:
12142 raise errors.OpPrereqError("Invalid disk size parameter: %s" % err,
12143 errors.ECODE_INVAL)
12145 params[constants.IDISK_SIZE] = size
12147 elif op == constants.DDM_MODIFY and constants.IDISK_SIZE in params:
12148 raise errors.OpPrereqError("Disk size change not possible, use"
12149 " grow-disk", errors.ECODE_INVAL)
12152 def _VerifyNicModification(op, params):
12153 """Verifies a network interface modification.
12156 if op in (constants.DDM_ADD, constants.DDM_MODIFY):
12157 ip = params.get(constants.INIC_IP, None)
12160 elif ip.lower() == constants.VALUE_NONE:
12161 params[constants.INIC_IP] = None
12162 elif not netutils.IPAddress.IsValid(ip):
12163 raise errors.OpPrereqError("Invalid IP address '%s'" % ip,
12164 errors.ECODE_INVAL)
12166 bridge = params.get("bridge", None)
12167 link = params.get(constants.INIC_LINK, None)
12168 if bridge and link:
12169 raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
12170 " at the same time", errors.ECODE_INVAL)
12171 elif bridge and bridge.lower() == constants.VALUE_NONE:
12172 params["bridge"] = None
12173 elif link and link.lower() == constants.VALUE_NONE:
12174 params[constants.INIC_LINK] = None
12176 if op == constants.DDM_ADD:
12177 macaddr = params.get(constants.INIC_MAC, None)
12178 if macaddr is None:
12179 params[constants.INIC_MAC] = constants.VALUE_AUTO
12181 if constants.INIC_MAC in params:
12182 macaddr = params[constants.INIC_MAC]
12183 if macaddr not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
12184 macaddr = utils.NormalizeAndValidateMac(macaddr)
12186 if op == constants.DDM_MODIFY and macaddr == constants.VALUE_AUTO:
12187 raise errors.OpPrereqError("'auto' is not a valid MAC address when"
12188 " modifying an existing NIC",
12189 errors.ECODE_INVAL)
12191 def CheckArguments(self):
12192 if not (self.op.nics or self.op.disks or self.op.disk_template or
12193 self.op.hvparams or self.op.beparams or self.op.os_name or
12194 self.op.offline is not None or self.op.runtime_mem):
12195 raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
12197 if self.op.hvparams:
12198 _CheckGlobalHvParams(self.op.hvparams)
12201 self._UpgradeDiskNicMods("disk", self.op.disks,
12202 opcodes.OpInstanceSetParams.TestDiskModifications)
12204 self._UpgradeDiskNicMods("NIC", self.op.nics,
12205 opcodes.OpInstanceSetParams.TestNicModifications)
12207 # Check disk modifications
12208 self._CheckMods("disk", self.op.disks, constants.IDISK_PARAMS_TYPES,
12209 self._VerifyDiskModification)
12211 if self.op.disks and self.op.disk_template is not None:
12212 raise errors.OpPrereqError("Disk template conversion and other disk"
12213 " changes not supported at the same time",
12214 errors.ECODE_INVAL)
12216 if (self.op.disk_template and
12217 self.op.disk_template in constants.DTS_INT_MIRROR and
12218 self.op.remote_node is None):
12219 raise errors.OpPrereqError("Changing the disk template to a mirrored"
12220 " one requires specifying a secondary node",
12221 errors.ECODE_INVAL)
12223 # Check NIC modifications
12224 self._CheckMods("NIC", self.op.nics, constants.INIC_PARAMS_TYPES,
12225 self._VerifyNicModification)
12227 def ExpandNames(self):
12228 self._ExpandAndLockInstance()
12229 # Can't even acquire node locks in shared mode as upcoming changes in
12230 # Ganeti 2.6 will start to modify the node object on disk conversion
12231 self.needed_locks[locking.LEVEL_NODE] = []
12232 self.needed_locks[locking.LEVEL_NODE_RES] = []
12233 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
12235 def DeclareLocks(self, level):
12236 # TODO: Acquire group lock in shared mode (disk parameters)
12237 if level == locking.LEVEL_NODE:
12238 self._LockInstancesNodes()
12239 if self.op.disk_template and self.op.remote_node:
12240 self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
12241 self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
12242 elif level == locking.LEVEL_NODE_RES and self.op.disk_template:
12244 self.needed_locks[locking.LEVEL_NODE_RES] = \
12245 self.needed_locks[locking.LEVEL_NODE][:]
12247 def BuildHooksEnv(self):
12248 """Build hooks env.
12250 This runs on the master, primary and secondaries.
12254 if constants.BE_MINMEM in self.be_new:
12255 args["minmem"] = self.be_new[constants.BE_MINMEM]
12256 if constants.BE_MAXMEM in self.be_new:
12257 args["maxmem"] = self.be_new[constants.BE_MAXMEM]
12258 if constants.BE_VCPUS in self.be_new:
12259 args["vcpus"] = self.be_new[constants.BE_VCPUS]
12260 # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
12261 # information at all.
12263 if self._new_nics is not None:
12266 for nic in self._new_nics:
12267 nicparams = self.cluster.SimpleFillNIC(nic.nicparams)
12268 mode = nicparams[constants.NIC_MODE]
12269 link = nicparams[constants.NIC_LINK]
12270 nics.append((nic.ip, nic.mac, mode, link))
12272 args["nics"] = nics
12274 env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
12275 if self.op.disk_template:
12276 env["NEW_DISK_TEMPLATE"] = self.op.disk_template
12277 if self.op.runtime_mem:
12278 env["RUNTIME_MEMORY"] = self.op.runtime_mem
12282 def BuildHooksNodes(self):
12283 """Build hooks nodes.
12286 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
12289 def _PrepareNicModification(self, params, private, old_ip, old_params,
12291 update_params_dict = dict([(key, params[key])
12292 for key in constants.NICS_PARAMETERS
12295 if "bridge" in params:
12296 update_params_dict[constants.NIC_LINK] = params["bridge"]
12298 new_params = _GetUpdatedParams(old_params, update_params_dict)
12299 utils.ForceDictType(new_params, constants.NICS_PARAMETER_TYPES)
12301 new_filled_params = cluster.SimpleFillNIC(new_params)
12302 objects.NIC.CheckParameterSyntax(new_filled_params)
12304 new_mode = new_filled_params[constants.NIC_MODE]
12305 if new_mode == constants.NIC_MODE_BRIDGED:
12306 bridge = new_filled_params[constants.NIC_LINK]
12307 msg = self.rpc.call_bridges_exist(pnode, [bridge]).fail_msg
12309 msg = "Error checking bridges on node '%s': %s" % (pnode, msg)
12311 self.warn.append(msg)
12313 raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
12315 elif new_mode == constants.NIC_MODE_ROUTED:
12316 ip = params.get(constants.INIC_IP, old_ip)
12318 raise errors.OpPrereqError("Cannot set the NIC IP address to None"
12319 " on a routed NIC", errors.ECODE_INVAL)
12321 if constants.INIC_MAC in params:
12322 mac = params[constants.INIC_MAC]
12324 raise errors.OpPrereqError("Cannot unset the NIC MAC address",
12325 errors.ECODE_INVAL)
12326 elif mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
12327 # otherwise generate the MAC address
12328 params[constants.INIC_MAC] = \
12329 self.cfg.GenerateMAC(self.proc.GetECId())
12331 # or validate/reserve the current one
12333 self.cfg.ReserveMAC(mac, self.proc.GetECId())
12334 except errors.ReservationError:
12335 raise errors.OpPrereqError("MAC address '%s' already in use"
12336 " in cluster" % mac,
12337 errors.ECODE_NOTUNIQUE)
12339 private.params = new_params
12340 private.filled = new_filled_params
12342 def CheckPrereq(self):
12343 """Check prerequisites.
12345 This only checks the instance list against the existing names.
12348 # checking the new params on the primary/secondary nodes
12350 instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
12351 cluster = self.cluster = self.cfg.GetClusterInfo()
12352 assert self.instance is not None, \
12353 "Cannot retrieve locked instance %s" % self.op.instance_name
12354 pnode = instance.primary_node
12355 nodelist = list(instance.all_nodes)
12356 pnode_info = self.cfg.GetNodeInfo(pnode)
12357 self.diskparams = self.cfg.GetInstanceDiskParams(instance)
12359 # Prepare disk/NIC modifications
12360 self.diskmod = PrepareContainerMods(self.op.disks, None)
12361 self.nicmod = PrepareContainerMods(self.op.nics, _InstNicModPrivate)
12364 if self.op.os_name and not self.op.force:
12365 _CheckNodeHasOS(self, instance.primary_node, self.op.os_name,
12366 self.op.force_variant)
12367 instance_os = self.op.os_name
12369 instance_os = instance.os
12371 assert not (self.op.disk_template and self.op.disks), \
12372 "Can't modify disk template and apply disk changes at the same time"
12374 if self.op.disk_template:
12375 if instance.disk_template == self.op.disk_template:
12376 raise errors.OpPrereqError("Instance already has disk template %s" %
12377 instance.disk_template, errors.ECODE_INVAL)
12379 if (instance.disk_template,
12380 self.op.disk_template) not in self._DISK_CONVERSIONS:
12381 raise errors.OpPrereqError("Unsupported disk template conversion from"
12382 " %s to %s" % (instance.disk_template,
12383 self.op.disk_template),
12384 errors.ECODE_INVAL)
12385 _CheckInstanceState(self, instance, INSTANCE_DOWN,
12386 msg="cannot change disk template")
12387 if self.op.disk_template in constants.DTS_INT_MIRROR:
12388 if self.op.remote_node == pnode:
12389 raise errors.OpPrereqError("Given new secondary node %s is the same"
12390 " as the primary node of the instance" %
12391 self.op.remote_node, errors.ECODE_STATE)
12392 _CheckNodeOnline(self, self.op.remote_node)
12393 _CheckNodeNotDrained(self, self.op.remote_node)
12394 # FIXME: here we assume that the old instance type is DT_PLAIN
12395 assert instance.disk_template == constants.DT_PLAIN
12396 disks = [{constants.IDISK_SIZE: d.size,
12397 constants.IDISK_VG: d.logical_id[0]}
12398 for d in instance.disks]
12399 required = _ComputeDiskSizePerVG(self.op.disk_template, disks)
12400 _CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required)
12402 snode_info = self.cfg.GetNodeInfo(self.op.remote_node)
12403 snode_group = self.cfg.GetNodeGroup(snode_info.group)
12404 ipolicy = _CalculateGroupIPolicy(cluster, snode_group)
12405 _CheckTargetNodeIPolicy(self, ipolicy, instance, snode_info,
12406 ignore=self.op.ignore_ipolicy)
12407 if pnode_info.group != snode_info.group:
12408 self.LogWarning("The primary and secondary nodes are in two"
12409 " different node groups; the disk parameters"
12410 " from the first disk's node group will be"
12413 # hvparams processing
12414 if self.op.hvparams:
12415 hv_type = instance.hypervisor
12416 i_hvdict = _GetUpdatedParams(instance.hvparams, self.op.hvparams)
12417 utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
12418 hv_new = cluster.SimpleFillHV(hv_type, instance.os, i_hvdict)
12421 hypervisor.GetHypervisor(hv_type).CheckParameterSyntax(hv_new)
12422 _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
12423 self.hv_proposed = self.hv_new = hv_new # the new actual values
12424 self.hv_inst = i_hvdict # the new dict (without defaults)
12426 self.hv_proposed = cluster.SimpleFillHV(instance.hypervisor, instance.os,
12428 self.hv_new = self.hv_inst = {}
12430 # beparams processing
12431 if self.op.beparams:
12432 i_bedict = _GetUpdatedParams(instance.beparams, self.op.beparams,
12434 objects.UpgradeBeParams(i_bedict)
12435 utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
12436 be_new = cluster.SimpleFillBE(i_bedict)
12437 self.be_proposed = self.be_new = be_new # the new actual values
12438 self.be_inst = i_bedict # the new dict (without defaults)
12440 self.be_new = self.be_inst = {}
12441 self.be_proposed = cluster.SimpleFillBE(instance.beparams)
12442 be_old = cluster.FillBE(instance)
12444 # CPU param validation -- checking every time a parameter is
12445 # changed to cover all cases where either CPU mask or vcpus have
12447 if (constants.BE_VCPUS in self.be_proposed and
12448 constants.HV_CPU_MASK in self.hv_proposed):
12450 utils.ParseMultiCpuMask(self.hv_proposed[constants.HV_CPU_MASK])
12451 # Verify mask is consistent with number of vCPUs. Can skip this
12452 # test if only 1 entry in the CPU mask, which means same mask
12453 # is applied to all vCPUs.
12454 if (len(cpu_list) > 1 and
12455 len(cpu_list) != self.be_proposed[constants.BE_VCPUS]):
12456 raise errors.OpPrereqError("Number of vCPUs [%d] does not match the"
12458 (self.be_proposed[constants.BE_VCPUS],
12459 self.hv_proposed[constants.HV_CPU_MASK]),
12460 errors.ECODE_INVAL)
12462 # Only perform this test if a new CPU mask is given
12463 if constants.HV_CPU_MASK in self.hv_new:
12464 # Calculate the largest CPU number requested
12465 max_requested_cpu = max(map(max, cpu_list))
12466 # Check that all of the instance's nodes have enough physical CPUs to
12467 # satisfy the requested CPU mask
12468 _CheckNodesPhysicalCPUs(self, instance.all_nodes,
12469 max_requested_cpu + 1, instance.hypervisor)
12471 # osparams processing
12472 if self.op.osparams:
12473 i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
12474 _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
12475 self.os_inst = i_osdict # the new dict (without defaults)
12481 #TODO(dynmem): do the appropriate check involving MINMEM
12482 if (constants.BE_MAXMEM in self.op.beparams and not self.op.force and
12483 be_new[constants.BE_MAXMEM] > be_old[constants.BE_MAXMEM]):
12484 mem_check_list = [pnode]
12485 if be_new[constants.BE_AUTO_BALANCE]:
12486 # either we changed auto_balance to yes or it was from before
12487 mem_check_list.extend(instance.secondary_nodes)
12488 instance_info = self.rpc.call_instance_info(pnode, instance.name,
12489 instance.hypervisor)
12490 nodeinfo = self.rpc.call_node_info(mem_check_list, None,
12491 [instance.hypervisor])
12492 pninfo = nodeinfo[pnode]
12493 msg = pninfo.fail_msg
12495 # Assume the primary node is unreachable and go ahead
12496 self.warn.append("Can't get info from primary node %s: %s" %
12499 (_, _, (pnhvinfo, )) = pninfo.payload
12500 if not isinstance(pnhvinfo.get("memory_free", None), int):
12501 self.warn.append("Node data from primary node %s doesn't contain"
12502 " free memory information" % pnode)
12503 elif instance_info.fail_msg:
12504 self.warn.append("Can't get instance runtime information: %s" %
12505 instance_info.fail_msg)
12507 if instance_info.payload:
12508 current_mem = int(instance_info.payload["memory"])
12510 # Assume instance not running
12511 # (there is a slight race condition here, but it's not very
12512 # probable, and we have no other way to check)
12513 # TODO: Describe race condition
12515 #TODO(dynmem): do the appropriate check involving MINMEM
12516 miss_mem = (be_new[constants.BE_MAXMEM] - current_mem -
12517 pnhvinfo["memory_free"])
12519 raise errors.OpPrereqError("This change will prevent the instance"
12520 " from starting, due to %d MB of memory"
12521 " missing on its primary node" %
12523 errors.ECODE_NORES)
12525 if be_new[constants.BE_AUTO_BALANCE]:
12526 for node, nres in nodeinfo.items():
12527 if node not in instance.secondary_nodes:
12529 nres.Raise("Can't get info from secondary node %s" % node,
12530 prereq=True, ecode=errors.ECODE_STATE)
12531 (_, _, (nhvinfo, )) = nres.payload
12532 if not isinstance(nhvinfo.get("memory_free", None), int):
12533 raise errors.OpPrereqError("Secondary node %s didn't return free"
12534 " memory information" % node,
12535 errors.ECODE_STATE)
12536 #TODO(dynmem): do the appropriate check involving MINMEM
12537 elif be_new[constants.BE_MAXMEM] > nhvinfo["memory_free"]:
12538 raise errors.OpPrereqError("This change will prevent the instance"
12539 " from failover to its secondary node"
12540 " %s, due to not enough memory" % node,
12541 errors.ECODE_STATE)
12543 if self.op.runtime_mem:
12544 remote_info = self.rpc.call_instance_info(instance.primary_node,
12546 instance.hypervisor)
12547 remote_info.Raise("Error checking node %s" % instance.primary_node)
12548 if not remote_info.payload: # not running already
12549 raise errors.OpPrereqError("Instance %s is not running" % instance.name,
12550 errors.ECODE_STATE)
12552 current_memory = remote_info.payload["memory"]
12553 if (not self.op.force and
12554 (self.op.runtime_mem > self.be_proposed[constants.BE_MAXMEM] or
12555 self.op.runtime_mem < self.be_proposed[constants.BE_MINMEM])):
12556 raise errors.OpPrereqError("Instance %s must have memory between %d"
12557 " and %d MB of memory unless --force is"
12558 " given" % (instance.name,
12559 self.be_proposed[constants.BE_MINMEM],
12560 self.be_proposed[constants.BE_MAXMEM]),
12561 errors.ECODE_INVAL)
12563 if self.op.runtime_mem > current_memory:
12564 _CheckNodeFreeMemory(self, instance.primary_node,
12565 "ballooning memory for instance %s" %
12567 self.op.memory - current_memory,
12568 instance.hypervisor)
12570 if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
12571 raise errors.OpPrereqError("Disk operations not supported for"
12572 " diskless instances",
12573 errors.ECODE_INVAL)
12575 def _PrepareNicCreate(_, params, private):
12576 self._PrepareNicModification(params, private, None, {}, cluster, pnode)
12577 return (None, None)
12579 def _PrepareNicMod(_, nic, params, private):
12580 self._PrepareNicModification(params, private, nic.ip,
12581 nic.nicparams, cluster, pnode)
12584 # Verify NIC changes (operating on copy)
12585 nics = instance.nics[:]
12586 ApplyContainerMods("NIC", nics, None, self.nicmod,
12587 _PrepareNicCreate, _PrepareNicMod, None)
12588 if len(nics) > constants.MAX_NICS:
12589 raise errors.OpPrereqError("Instance has too many network interfaces"
12590 " (%d), cannot add more" % constants.MAX_NICS,
12591 errors.ECODE_STATE)
12593 # Verify disk changes (operating on a copy)
12594 disks = instance.disks[:]
12595 ApplyContainerMods("disk", disks, None, self.diskmod, None, None, None)
12596 if len(disks) > constants.MAX_DISKS:
12597 raise errors.OpPrereqError("Instance has too many disks (%d), cannot add"
12598 " more" % constants.MAX_DISKS,
12599 errors.ECODE_STATE)
12601 if self.op.offline is not None:
12602 if self.op.offline:
12603 msg = "can't change to offline"
12605 msg = "can't change to online"
12606 _CheckInstanceState(self, instance, CAN_CHANGE_INSTANCE_OFFLINE, msg=msg)
12608 # Pre-compute NIC changes (necessary to use result in hooks)
12609 self._nic_chgdesc = []
12611 # Operate on copies as this is still in prereq
12612 nics = [nic.Copy() for nic in instance.nics]
12613 ApplyContainerMods("NIC", nics, self._nic_chgdesc, self.nicmod,
12614 self._CreateNewNic, self._ApplyNicMods, None)
12615 self._new_nics = nics
12617 self._new_nics = None
12619 def _ConvertPlainToDrbd(self, feedback_fn):
12620 """Converts an instance from plain to drbd.
12623 feedback_fn("Converting template to drbd")
12624 instance = self.instance
12625 pnode = instance.primary_node
12626 snode = self.op.remote_node
12628 assert instance.disk_template == constants.DT_PLAIN
12630 # create a fake disk info for _GenerateDiskTemplate
12631 disk_info = [{constants.IDISK_SIZE: d.size, constants.IDISK_MODE: d.mode,
12632 constants.IDISK_VG: d.logical_id[0]}
12633 for d in instance.disks]
12634 new_disks = _GenerateDiskTemplate(self, self.op.disk_template,
12635 instance.name, pnode, [snode],
12636 disk_info, None, None, 0, feedback_fn,
12638 anno_disks = rpc.AnnotateDiskParams(constants.DT_DRBD8, new_disks,
12640 info = _GetInstanceInfoText(instance)
12641 feedback_fn("Creating additional volumes...")
12642 # first, create the missing data and meta devices
12643 for disk in anno_disks:
12644 # unfortunately this is... not too nice
12645 _CreateSingleBlockDev(self, pnode, instance, disk.children[1],
12647 for child in disk.children:
12648 _CreateSingleBlockDev(self, snode, instance, child, info, True)
12649 # at this stage, all new LVs have been created, we can rename the
12651 feedback_fn("Renaming original volumes...")
12652 rename_list = [(o, n.children[0].logical_id)
12653 for (o, n) in zip(instance.disks, new_disks)]
12654 result = self.rpc.call_blockdev_rename(pnode, rename_list)
12655 result.Raise("Failed to rename original LVs")
12657 feedback_fn("Initializing DRBD devices...")
12658 # all child devices are in place, we can now create the DRBD devices
12659 for disk in anno_disks:
12660 for node in [pnode, snode]:
12661 f_create = node == pnode
12662 _CreateSingleBlockDev(self, node, instance, disk, info, f_create)
12664 # at this point, the instance has been modified
12665 instance.disk_template = constants.DT_DRBD8
12666 instance.disks = new_disks
12667 self.cfg.Update(instance, feedback_fn)
12669 # Release node locks while waiting for sync
12670 _ReleaseLocks(self, locking.LEVEL_NODE)
12672 # disks are created, waiting for sync
12673 disk_abort = not _WaitForSync(self, instance,
12674 oneshot=not self.op.wait_for_sync)
12676 raise errors.OpExecError("There are some degraded disks for"
12677 " this instance, please cleanup manually")
12679 # Node resource locks will be released by caller
12681 def _ConvertDrbdToPlain(self, feedback_fn):
12682 """Converts an instance from drbd to plain.
12685 instance = self.instance
12687 assert len(instance.secondary_nodes) == 1
12688 assert instance.disk_template == constants.DT_DRBD8
12690 pnode = instance.primary_node
12691 snode = instance.secondary_nodes[0]
12692 feedback_fn("Converting template to plain")
12694 old_disks = instance.disks
12695 new_disks = [d.children[0] for d in old_disks]
12697 # copy over size and mode
12698 for parent, child in zip(old_disks, new_disks):
12699 child.size = parent.size
12700 child.mode = parent.mode
12702 # this is a DRBD disk, return its port to the pool
12703 # NOTE: this must be done right before the call to cfg.Update!
12704 for disk in old_disks:
12705 tcp_port = disk.logical_id[2]
12706 self.cfg.AddTcpUdpPort(tcp_port)
12708 # update instance structure
12709 instance.disks = new_disks
12710 instance.disk_template = constants.DT_PLAIN
12711 self.cfg.Update(instance, feedback_fn)
12713 # Release locks in case removing disks takes a while
12714 _ReleaseLocks(self, locking.LEVEL_NODE)
12716 feedback_fn("Removing volumes on the secondary node...")
12717 for disk in old_disks:
12718 self.cfg.SetDiskID(disk, snode)
12719 msg = self.rpc.call_blockdev_remove(snode, disk).fail_msg
12721 self.LogWarning("Could not remove block device %s on node %s,"
12722 " continuing anyway: %s", disk.iv_name, snode, msg)
12724 feedback_fn("Removing unneeded volumes on the primary node...")
12725 for idx, disk in enumerate(old_disks):
12726 meta = disk.children[1]
12727 self.cfg.SetDiskID(meta, pnode)
12728 msg = self.rpc.call_blockdev_remove(pnode, meta).fail_msg
12730 self.LogWarning("Could not remove metadata for disk %d on node %s,"
12731 " continuing anyway: %s", idx, pnode, msg)
12733 def _CreateNewDisk(self, idx, params, _):
12734 """Creates a new disk.
12737 instance = self.instance
12740 if instance.disk_template in constants.DTS_FILEBASED:
12741 (file_driver, file_path) = instance.disks[0].logical_id
12742 file_path = os.path.dirname(file_path)
12744 file_driver = file_path = None
12747 _GenerateDiskTemplate(self, instance.disk_template, instance.name,
12748 instance.primary_node, instance.secondary_nodes,
12749 [params], file_path, file_driver, idx,
12750 self.Log, self.diskparams)[0]
12752 info = _GetInstanceInfoText(instance)
12754 logging.info("Creating volume %s for instance %s",
12755 disk.iv_name, instance.name)
12756 # Note: this needs to be kept in sync with _CreateDisks
12758 for node in instance.all_nodes:
12759 f_create = (node == instance.primary_node)
12761 _CreateBlockDev(self, node, instance, disk, f_create, info, f_create)
12762 except errors.OpExecError, err:
12763 self.LogWarning("Failed to create volume %s (%s) on node '%s': %s",
12764 disk.iv_name, disk, node, err)
12767 ("disk/%d" % idx, "add:size=%s,mode=%s" % (disk.size, disk.mode)),
12771 def _ModifyDisk(idx, disk, params, _):
12772 """Modifies a disk.
12775 disk.mode = params[constants.IDISK_MODE]
12778 ("disk.mode/%d" % idx, disk.mode),
12781 def _RemoveDisk(self, idx, root, _):
12785 for node, disk in root.ComputeNodeTree(self.instance.primary_node):
12786 self.cfg.SetDiskID(disk, node)
12787 msg = self.rpc.call_blockdev_remove(node, disk).fail_msg
12789 self.LogWarning("Could not remove disk/%d on node '%s': %s,"
12790 " continuing anyway", idx, node, msg)
12792 # if this is a DRBD disk, return its port to the pool
12793 if root.dev_type in constants.LDS_DRBD:
12794 self.cfg.AddTcpUdpPort(root.logical_id[2])
12797 def _CreateNewNic(idx, params, private):
12798 """Creates data structure for a new network interface.
12801 mac = params[constants.INIC_MAC]
12802 ip = params.get(constants.INIC_IP, None)
12803 nicparams = private.params
12805 return (objects.NIC(mac=mac, ip=ip, nicparams=nicparams), [
12807 "add:mac=%s,ip=%s,mode=%s,link=%s" %
12808 (mac, ip, private.filled[constants.NIC_MODE],
12809 private.filled[constants.NIC_LINK])),
12813 def _ApplyNicMods(idx, nic, params, private):
12814 """Modifies a network interface.
12819 for key in [constants.INIC_MAC, constants.INIC_IP]:
12821 changes.append(("nic.%s/%d" % (key, idx), params[key]))
12822 setattr(nic, key, params[key])
12825 nic.nicparams = private.params
12827 for (key, val) in params.items():
12828 changes.append(("nic.%s/%d" % (key, idx), val))
12832 def Exec(self, feedback_fn):
12833 """Modifies an instance.
12835 All parameters take effect only at the next restart of the instance.
12838 # Process here the warnings from CheckPrereq, as we don't have a
12839 # feedback_fn there.
12840 # TODO: Replace with self.LogWarning
12841 for warn in self.warn:
12842 feedback_fn("WARNING: %s" % warn)
12844 assert ((self.op.disk_template is None) ^
12845 bool(self.owned_locks(locking.LEVEL_NODE_RES))), \
12846 "Not owning any node resource locks"
12849 instance = self.instance
12852 if self.op.runtime_mem:
12853 rpcres = self.rpc.call_instance_balloon_memory(instance.primary_node,
12855 self.op.runtime_mem)
12856 rpcres.Raise("Cannot modify instance runtime memory")
12857 result.append(("runtime_memory", self.op.runtime_mem))
12859 # Apply disk changes
12860 ApplyContainerMods("disk", instance.disks, result, self.diskmod,
12861 self._CreateNewDisk, self._ModifyDisk, self._RemoveDisk)
12862 _UpdateIvNames(0, instance.disks)
12864 if self.op.disk_template:
12866 check_nodes = set(instance.all_nodes)
12867 if self.op.remote_node:
12868 check_nodes.add(self.op.remote_node)
12869 for level in [locking.LEVEL_NODE, locking.LEVEL_NODE_RES]:
12870 owned = self.owned_locks(level)
12871 assert not (check_nodes - owned), \
12872 ("Not owning the correct locks, owning %r, expected at least %r" %
12873 (owned, check_nodes))
12875 r_shut = _ShutdownInstanceDisks(self, instance)
12877 raise errors.OpExecError("Cannot shutdown instance disks, unable to"
12878 " proceed with disk template conversion")
12879 mode = (instance.disk_template, self.op.disk_template)
12881 self._DISK_CONVERSIONS[mode](self, feedback_fn)
12883 self.cfg.ReleaseDRBDMinors(instance.name)
12885 result.append(("disk_template", self.op.disk_template))
12887 assert instance.disk_template == self.op.disk_template, \
12888 ("Expected disk template '%s', found '%s'" %
12889 (self.op.disk_template, instance.disk_template))
12891 # Release node and resource locks if there are any (they might already have
12892 # been released during disk conversion)
12893 _ReleaseLocks(self, locking.LEVEL_NODE)
12894 _ReleaseLocks(self, locking.LEVEL_NODE_RES)
12896 # Apply NIC changes
12897 if self._new_nics is not None:
12898 instance.nics = self._new_nics
12899 result.extend(self._nic_chgdesc)
12902 if self.op.hvparams:
12903 instance.hvparams = self.hv_inst
12904 for key, val in self.op.hvparams.iteritems():
12905 result.append(("hv/%s" % key, val))
12908 if self.op.beparams:
12909 instance.beparams = self.be_inst
12910 for key, val in self.op.beparams.iteritems():
12911 result.append(("be/%s" % key, val))
12914 if self.op.os_name:
12915 instance.os = self.op.os_name
12918 if self.op.osparams:
12919 instance.osparams = self.os_inst
12920 for key, val in self.op.osparams.iteritems():
12921 result.append(("os/%s" % key, val))
12923 if self.op.offline is None:
12926 elif self.op.offline:
12927 # Mark instance as offline
12928 self.cfg.MarkInstanceOffline(instance.name)
12929 result.append(("admin_state", constants.ADMINST_OFFLINE))
12931 # Mark instance as online, but stopped
12932 self.cfg.MarkInstanceDown(instance.name)
12933 result.append(("admin_state", constants.ADMINST_DOWN))
12935 self.cfg.Update(instance, feedback_fn)
12937 assert not (self.owned_locks(locking.LEVEL_NODE_RES) or
12938 self.owned_locks(locking.LEVEL_NODE)), \
12939 "All node locks should have been released by now"
12943 _DISK_CONVERSIONS = {
12944 (constants.DT_PLAIN, constants.DT_DRBD8): _ConvertPlainToDrbd,
12945 (constants.DT_DRBD8, constants.DT_PLAIN): _ConvertDrbdToPlain,
12949 class LUInstanceChangeGroup(LogicalUnit):
12950 HPATH = "instance-change-group"
12951 HTYPE = constants.HTYPE_INSTANCE
12954 def ExpandNames(self):
12955 self.share_locks = _ShareAll()
12956 self.needed_locks = {
12957 locking.LEVEL_NODEGROUP: [],
12958 locking.LEVEL_NODE: [],
12961 self._ExpandAndLockInstance()
12963 if self.op.target_groups:
12964 self.req_target_uuids = map(self.cfg.LookupNodeGroup,
12965 self.op.target_groups)
12967 self.req_target_uuids = None
12969 self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
12971 def DeclareLocks(self, level):
12972 if level == locking.LEVEL_NODEGROUP:
12973 assert not self.needed_locks[locking.LEVEL_NODEGROUP]
12975 if self.req_target_uuids:
12976 lock_groups = set(self.req_target_uuids)
12978 # Lock all groups used by instance optimistically; this requires going
12979 # via the node before it's locked, requiring verification later on
12980 instance_groups = self.cfg.GetInstanceNodeGroups(self.op.instance_name)
12981 lock_groups.update(instance_groups)
12983 # No target groups, need to lock all of them
12984 lock_groups = locking.ALL_SET
12986 self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
12988 elif level == locking.LEVEL_NODE:
12989 if self.req_target_uuids:
12990 # Lock all nodes used by instances
12991 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
12992 self._LockInstancesNodes()
12994 # Lock all nodes in all potential target groups
12995 lock_groups = (frozenset(self.owned_locks(locking.LEVEL_NODEGROUP)) -
12996 self.cfg.GetInstanceNodeGroups(self.op.instance_name))
12997 member_nodes = [node_name
12998 for group in lock_groups
12999 for node_name in self.cfg.GetNodeGroup(group).members]
13000 self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
13002 # Lock all nodes as all groups are potential targets
13003 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
13005 def CheckPrereq(self):
13006 owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
13007 owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
13008 owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
13010 assert (self.req_target_uuids is None or
13011 owned_groups.issuperset(self.req_target_uuids))
13012 assert owned_instances == set([self.op.instance_name])
13014 # Get instance information
13015 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
13017 # Check if node groups for locked instance are still correct
13018 assert owned_nodes.issuperset(self.instance.all_nodes), \
13019 ("Instance %s's nodes changed while we kept the lock" %
13020 self.op.instance_name)
13022 inst_groups = _CheckInstanceNodeGroups(self.cfg, self.op.instance_name,
13025 if self.req_target_uuids:
13026 # User requested specific target groups
13027 self.target_uuids = frozenset(self.req_target_uuids)
13029 # All groups except those used by the instance are potential targets
13030 self.target_uuids = owned_groups - inst_groups
13032 conflicting_groups = self.target_uuids & inst_groups
13033 if conflicting_groups:
13034 raise errors.OpPrereqError("Can't use group(s) '%s' as targets, they are"
13035 " used by the instance '%s'" %
13036 (utils.CommaJoin(conflicting_groups),
13037 self.op.instance_name),
13038 errors.ECODE_INVAL)
13040 if not self.target_uuids:
13041 raise errors.OpPrereqError("There are no possible target groups",
13042 errors.ECODE_INVAL)
13044 def BuildHooksEnv(self):
13045 """Build hooks env.
13048 assert self.target_uuids
13051 "TARGET_GROUPS": " ".join(self.target_uuids),
13054 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
13058 def BuildHooksNodes(self):
13059 """Build hooks nodes.
13062 mn = self.cfg.GetMasterNode()
13063 return ([mn], [mn])
13065 def Exec(self, feedback_fn):
13066 instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
13068 assert instances == [self.op.instance_name], "Instance not locked"
13070 ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_CHG_GROUP,
13071 instances=instances, target_groups=list(self.target_uuids))
13073 ial.Run(self.op.iallocator)
13075 if not ial.success:
13076 raise errors.OpPrereqError("Can't compute solution for changing group of"
13077 " instance '%s' using iallocator '%s': %s" %
13078 (self.op.instance_name, self.op.iallocator,
13080 errors.ECODE_NORES)
13082 jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, False)
13084 self.LogInfo("Iallocator returned %s job(s) for changing group of"
13085 " instance '%s'", len(jobs), self.op.instance_name)
13087 return ResultWithJobs(jobs)
13090 class LUBackupQuery(NoHooksLU):
13091 """Query the exports list
13096 def CheckArguments(self):
13097 self.expq = _ExportQuery(qlang.MakeSimpleFilter("node", self.op.nodes),
13098 ["node", "export"], self.op.use_locking)
13100 def ExpandNames(self):
13101 self.expq.ExpandNames(self)
13103 def DeclareLocks(self, level):
13104 self.expq.DeclareLocks(self, level)
13106 def Exec(self, feedback_fn):
13109 for (node, expname) in self.expq.OldStyleQuery(self):
13110 if expname is None:
13111 result[node] = False
13113 result.setdefault(node, []).append(expname)
13118 class _ExportQuery(_QueryBase):
13119 FIELDS = query.EXPORT_FIELDS
13121 #: The node name is not a unique key for this query
13122 SORT_FIELD = "node"
13124 def ExpandNames(self, lu):
13125 lu.needed_locks = {}
13127 # The following variables interact with _QueryBase._GetNames
13129 self.wanted = _GetWantedNodes(lu, self.names)
13131 self.wanted = locking.ALL_SET
13133 self.do_locking = self.use_locking
13135 if self.do_locking:
13136 lu.share_locks = _ShareAll()
13137 lu.needed_locks = {
13138 locking.LEVEL_NODE: self.wanted,
13141 def DeclareLocks(self, lu, level):
13144 def _GetQueryData(self, lu):
13145 """Computes the list of nodes and their attributes.
13148 # Locking is not used
13150 assert not (compat.any(lu.glm.is_owned(level)
13151 for level in locking.LEVELS
13152 if level != locking.LEVEL_CLUSTER) or
13153 self.do_locking or self.use_locking)
13155 nodes = self._GetNames(lu, lu.cfg.GetNodeList(), locking.LEVEL_NODE)
13159 for (node, nres) in lu.rpc.call_export_list(nodes).items():
13161 result.append((node, None))
13163 result.extend((node, expname) for expname in nres.payload)
13168 class LUBackupPrepare(NoHooksLU):
13169 """Prepares an instance for an export and returns useful information.
13174 def ExpandNames(self):
13175 self._ExpandAndLockInstance()
13177 def CheckPrereq(self):
13178 """Check prerequisites.
13181 instance_name = self.op.instance_name
13183 self.instance = self.cfg.GetInstanceInfo(instance_name)
13184 assert self.instance is not None, \
13185 "Cannot retrieve locked instance %s" % self.op.instance_name
13186 _CheckNodeOnline(self, self.instance.primary_node)
13188 self._cds = _GetClusterDomainSecret()
13190 def Exec(self, feedback_fn):
13191 """Prepares an instance for an export.
13194 instance = self.instance
13196 if self.op.mode == constants.EXPORT_MODE_REMOTE:
13197 salt = utils.GenerateSecret(8)
13199 feedback_fn("Generating X509 certificate on %s" % instance.primary_node)
13200 result = self.rpc.call_x509_cert_create(instance.primary_node,
13201 constants.RIE_CERT_VALIDITY)
13202 result.Raise("Can't create X509 key and certificate on %s" % result.node)
13204 (name, cert_pem) = result.payload
13206 cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
13210 "handshake": masterd.instance.ComputeRemoteExportHandshake(self._cds),
13211 "x509_key_name": (name, utils.Sha1Hmac(self._cds, name, salt=salt),
13213 "x509_ca": utils.SignX509Certificate(cert, self._cds, salt),
13219 class LUBackupExport(LogicalUnit):
13220 """Export an instance to an image in the cluster.
13223 HPATH = "instance-export"
13224 HTYPE = constants.HTYPE_INSTANCE
13227 def CheckArguments(self):
13228 """Check the arguments.
13231 self.x509_key_name = self.op.x509_key_name
13232 self.dest_x509_ca_pem = self.op.destination_x509_ca
13234 if self.op.mode == constants.EXPORT_MODE_REMOTE:
13235 if not self.x509_key_name:
13236 raise errors.OpPrereqError("Missing X509 key name for encryption",
13237 errors.ECODE_INVAL)
13239 if not self.dest_x509_ca_pem:
13240 raise errors.OpPrereqError("Missing destination X509 CA",
13241 errors.ECODE_INVAL)
13243 def ExpandNames(self):
13244 self._ExpandAndLockInstance()
13246 # Lock all nodes for local exports
13247 if self.op.mode == constants.EXPORT_MODE_LOCAL:
13248 # FIXME: lock only instance primary and destination node
13250 # Sad but true, for now we have do lock all nodes, as we don't know where
13251 # the previous export might be, and in this LU we search for it and
13252 # remove it from its current node. In the future we could fix this by:
13253 # - making a tasklet to search (share-lock all), then create the
13254 # new one, then one to remove, after
13255 # - removing the removal operation altogether
13256 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
13258 def DeclareLocks(self, level):
13259 """Last minute lock declaration."""
13260 # All nodes are locked anyway, so nothing to do here.
13262 def BuildHooksEnv(self):
13263 """Build hooks env.
13265 This will run on the master, primary node and target node.
13269 "EXPORT_MODE": self.op.mode,
13270 "EXPORT_NODE": self.op.target_node,
13271 "EXPORT_DO_SHUTDOWN": self.op.shutdown,
13272 "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
13273 # TODO: Generic function for boolean env variables
13274 "REMOVE_INSTANCE": str(bool(self.op.remove_instance)),
13277 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
13281 def BuildHooksNodes(self):
13282 """Build hooks nodes.
13285 nl = [self.cfg.GetMasterNode(), self.instance.primary_node]
13287 if self.op.mode == constants.EXPORT_MODE_LOCAL:
13288 nl.append(self.op.target_node)
13292 def CheckPrereq(self):
13293 """Check prerequisites.
13295 This checks that the instance and node names are valid.
13298 instance_name = self.op.instance_name
13300 self.instance = self.cfg.GetInstanceInfo(instance_name)
13301 assert self.instance is not None, \
13302 "Cannot retrieve locked instance %s" % self.op.instance_name
13303 _CheckNodeOnline(self, self.instance.primary_node)
13305 if (self.op.remove_instance and
13306 self.instance.admin_state == constants.ADMINST_UP and
13307 not self.op.shutdown):
13308 raise errors.OpPrereqError("Can not remove instance without shutting it"
13311 if self.op.mode == constants.EXPORT_MODE_LOCAL:
13312 self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
13313 self.dst_node = self.cfg.GetNodeInfo(self.op.target_node)
13314 assert self.dst_node is not None
13316 _CheckNodeOnline(self, self.dst_node.name)
13317 _CheckNodeNotDrained(self, self.dst_node.name)
13320 self.dest_disk_info = None
13321 self.dest_x509_ca = None
13323 elif self.op.mode == constants.EXPORT_MODE_REMOTE:
13324 self.dst_node = None
13326 if len(self.op.target_node) != len(self.instance.disks):
13327 raise errors.OpPrereqError(("Received destination information for %s"
13328 " disks, but instance %s has %s disks") %
13329 (len(self.op.target_node), instance_name,
13330 len(self.instance.disks)),
13331 errors.ECODE_INVAL)
13333 cds = _GetClusterDomainSecret()
13335 # Check X509 key name
13337 (key_name, hmac_digest, hmac_salt) = self.x509_key_name
13338 except (TypeError, ValueError), err:
13339 raise errors.OpPrereqError("Invalid data for X509 key name: %s" % err)
13341 if not utils.VerifySha1Hmac(cds, key_name, hmac_digest, salt=hmac_salt):
13342 raise errors.OpPrereqError("HMAC for X509 key name is wrong",
13343 errors.ECODE_INVAL)
13345 # Load and verify CA
13347 (cert, _) = utils.LoadSignedX509Certificate(self.dest_x509_ca_pem, cds)
13348 except OpenSSL.crypto.Error, err:
13349 raise errors.OpPrereqError("Unable to load destination X509 CA (%s)" %
13350 (err, ), errors.ECODE_INVAL)
13352 (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
13353 if errcode is not None:
13354 raise errors.OpPrereqError("Invalid destination X509 CA (%s)" %
13355 (msg, ), errors.ECODE_INVAL)
13357 self.dest_x509_ca = cert
13359 # Verify target information
13361 for idx, disk_data in enumerate(self.op.target_node):
13363 (host, port, magic) = \
13364 masterd.instance.CheckRemoteExportDiskInfo(cds, idx, disk_data)
13365 except errors.GenericError, err:
13366 raise errors.OpPrereqError("Target info for disk %s: %s" %
13367 (idx, err), errors.ECODE_INVAL)
13369 disk_info.append((host, port, magic))
13371 assert len(disk_info) == len(self.op.target_node)
13372 self.dest_disk_info = disk_info
13375 raise errors.ProgrammerError("Unhandled export mode %r" %
13378 # instance disk type verification
13379 # TODO: Implement export support for file-based disks
13380 for disk in self.instance.disks:
13381 if disk.dev_type == constants.LD_FILE:
13382 raise errors.OpPrereqError("Export not supported for instances with"
13383 " file-based disks", errors.ECODE_INVAL)
13385 def _CleanupExports(self, feedback_fn):
13386 """Removes exports of current instance from all other nodes.
13388 If an instance in a cluster with nodes A..D was exported to node C, its
13389 exports will be removed from the nodes A, B and D.
13392 assert self.op.mode != constants.EXPORT_MODE_REMOTE
13394 nodelist = self.cfg.GetNodeList()
13395 nodelist.remove(self.dst_node.name)
13397 # on one-node clusters nodelist will be empty after the removal
13398 # if we proceed the backup would be removed because OpBackupQuery
13399 # substitutes an empty list with the full cluster node list.
13400 iname = self.instance.name
13402 feedback_fn("Removing old exports for instance %s" % iname)
13403 exportlist = self.rpc.call_export_list(nodelist)
13404 for node in exportlist:
13405 if exportlist[node].fail_msg:
13407 if iname in exportlist[node].payload:
13408 msg = self.rpc.call_export_remove(node, iname).fail_msg
13410 self.LogWarning("Could not remove older export for instance %s"
13411 " on node %s: %s", iname, node, msg)
13413 def Exec(self, feedback_fn):
13414 """Export an instance to an image in the cluster.
13417 assert self.op.mode in constants.EXPORT_MODES
13419 instance = self.instance
13420 src_node = instance.primary_node
13422 if self.op.shutdown:
13423 # shutdown the instance, but not the disks
13424 feedback_fn("Shutting down instance %s" % instance.name)
13425 result = self.rpc.call_instance_shutdown(src_node, instance,
13426 self.op.shutdown_timeout)
13427 # TODO: Maybe ignore failures if ignore_remove_failures is set
13428 result.Raise("Could not shutdown instance %s on"
13429 " node %s" % (instance.name, src_node))
13431 # set the disks ID correctly since call_instance_start needs the
13432 # correct drbd minor to create the symlinks
13433 for disk in instance.disks:
13434 self.cfg.SetDiskID(disk, src_node)
13436 activate_disks = (instance.admin_state != constants.ADMINST_UP)
13439 # Activate the instance disks if we'exporting a stopped instance
13440 feedback_fn("Activating disks for %s" % instance.name)
13441 _StartInstanceDisks(self, instance, None)
13444 helper = masterd.instance.ExportInstanceHelper(self, feedback_fn,
13447 helper.CreateSnapshots()
13449 if (self.op.shutdown and
13450 instance.admin_state == constants.ADMINST_UP and
13451 not self.op.remove_instance):
13452 assert not activate_disks
13453 feedback_fn("Starting instance %s" % instance.name)
13454 result = self.rpc.call_instance_start(src_node,
13455 (instance, None, None), False)
13456 msg = result.fail_msg
13458 feedback_fn("Failed to start instance: %s" % msg)
13459 _ShutdownInstanceDisks(self, instance)
13460 raise errors.OpExecError("Could not start instance: %s" % msg)
13462 if self.op.mode == constants.EXPORT_MODE_LOCAL:
13463 (fin_resu, dresults) = helper.LocalExport(self.dst_node)
13464 elif self.op.mode == constants.EXPORT_MODE_REMOTE:
13465 connect_timeout = constants.RIE_CONNECT_TIMEOUT
13466 timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
13468 (key_name, _, _) = self.x509_key_name
13471 OpenSSL.crypto.dump_certificate(OpenSSL.crypto.FILETYPE_PEM,
13474 (fin_resu, dresults) = helper.RemoteExport(self.dest_disk_info,
13475 key_name, dest_ca_pem,
13480 # Check for backwards compatibility
13481 assert len(dresults) == len(instance.disks)
13482 assert compat.all(isinstance(i, bool) for i in dresults), \
13483 "Not all results are boolean: %r" % dresults
13487 feedback_fn("Deactivating disks for %s" % instance.name)
13488 _ShutdownInstanceDisks(self, instance)
13490 if not (compat.all(dresults) and fin_resu):
13493 failures.append("export finalization")
13494 if not compat.all(dresults):
13495 fdsk = utils.CommaJoin(idx for (idx, dsk) in enumerate(dresults)
13497 failures.append("disk export: disk(s) %s" % fdsk)
13499 raise errors.OpExecError("Export failed, errors in %s" %
13500 utils.CommaJoin(failures))
13502 # At this point, the export was successful, we can cleanup/finish
13504 # Remove instance if requested
13505 if self.op.remove_instance:
13506 feedback_fn("Removing instance %s" % instance.name)
13507 _RemoveInstance(self, feedback_fn, instance,
13508 self.op.ignore_remove_failures)
13510 if self.op.mode == constants.EXPORT_MODE_LOCAL:
13511 self._CleanupExports(feedback_fn)
13513 return fin_resu, dresults
13516 class LUBackupRemove(NoHooksLU):
13517 """Remove exports related to the named instance.
13522 def ExpandNames(self):
13523 self.needed_locks = {}
13524 # We need all nodes to be locked in order for RemoveExport to work, but we
13525 # don't need to lock the instance itself, as nothing will happen to it (and
13526 # we can remove exports also for a removed instance)
13527 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
13529 def Exec(self, feedback_fn):
13530 """Remove any export.
13533 instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
13534 # If the instance was not found we'll try with the name that was passed in.
13535 # This will only work if it was an FQDN, though.
13537 if not instance_name:
13539 instance_name = self.op.instance_name
13541 locked_nodes = self.owned_locks(locking.LEVEL_NODE)
13542 exportlist = self.rpc.call_export_list(locked_nodes)
13544 for node in exportlist:
13545 msg = exportlist[node].fail_msg
13547 self.LogWarning("Failed to query node %s (continuing): %s", node, msg)
13549 if instance_name in exportlist[node].payload:
13551 result = self.rpc.call_export_remove(node, instance_name)
13552 msg = result.fail_msg
13554 logging.error("Could not remove export for instance %s"
13555 " on node %s: %s", instance_name, node, msg)
13557 if fqdn_warn and not found:
13558 feedback_fn("Export not found. If trying to remove an export belonging"
13559 " to a deleted instance please use its Fully Qualified"
13563 class LUGroupAdd(LogicalUnit):
13564 """Logical unit for creating node groups.
13567 HPATH = "group-add"
13568 HTYPE = constants.HTYPE_GROUP
13571 def ExpandNames(self):
13572 # We need the new group's UUID here so that we can create and acquire the
13573 # corresponding lock. Later, in Exec(), we'll indicate to cfg.AddNodeGroup
13574 # that it should not check whether the UUID exists in the configuration.
13575 self.group_uuid = self.cfg.GenerateUniqueID(self.proc.GetECId())
13576 self.needed_locks = {}
13577 self.add_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
13579 def CheckPrereq(self):
13580 """Check prerequisites.
13582 This checks that the given group name is not an existing node group
13587 existing_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
13588 except errors.OpPrereqError:
13591 raise errors.OpPrereqError("Desired group name '%s' already exists as a"
13592 " node group (UUID: %s)" %
13593 (self.op.group_name, existing_uuid),
13594 errors.ECODE_EXISTS)
13596 if self.op.ndparams:
13597 utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
13599 if self.op.hv_state:
13600 self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state, None)
13602 self.new_hv_state = None
13604 if self.op.disk_state:
13605 self.new_disk_state = _MergeAndVerifyDiskState(self.op.disk_state, None)
13607 self.new_disk_state = None
13609 if self.op.diskparams:
13610 for templ in constants.DISK_TEMPLATES:
13611 if templ in self.op.diskparams:
13612 utils.ForceDictType(self.op.diskparams[templ],
13613 constants.DISK_DT_TYPES)
13614 self.new_diskparams = self.op.diskparams
13616 utils.VerifyDictOptions(self.new_diskparams, constants.DISK_DT_DEFAULTS)
13617 except errors.OpPrereqError, err:
13618 raise errors.OpPrereqError("While verify diskparams options: %s" % err,
13619 errors.ECODE_INVAL)
13621 self.new_diskparams = {}
13623 if self.op.ipolicy:
13624 cluster = self.cfg.GetClusterInfo()
13625 full_ipolicy = cluster.SimpleFillIPolicy(self.op.ipolicy)
13627 objects.InstancePolicy.CheckParameterSyntax(full_ipolicy)
13628 except errors.ConfigurationError, err:
13629 raise errors.OpPrereqError("Invalid instance policy: %s" % err,
13630 errors.ECODE_INVAL)
13632 def BuildHooksEnv(self):
13633 """Build hooks env.
13637 "GROUP_NAME": self.op.group_name,
13640 def BuildHooksNodes(self):
13641 """Build hooks nodes.
13644 mn = self.cfg.GetMasterNode()
13645 return ([mn], [mn])
13647 def Exec(self, feedback_fn):
13648 """Add the node group to the cluster.
13651 group_obj = objects.NodeGroup(name=self.op.group_name, members=[],
13652 uuid=self.group_uuid,
13653 alloc_policy=self.op.alloc_policy,
13654 ndparams=self.op.ndparams,
13655 diskparams=self.new_diskparams,
13656 ipolicy=self.op.ipolicy,
13657 hv_state_static=self.new_hv_state,
13658 disk_state_static=self.new_disk_state)
13660 self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False)
13661 del self.remove_locks[locking.LEVEL_NODEGROUP]
13664 class LUGroupAssignNodes(NoHooksLU):
13665 """Logical unit for assigning nodes to groups.
13670 def ExpandNames(self):
13671 # These raise errors.OpPrereqError on their own:
13672 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
13673 self.op.nodes = _GetWantedNodes(self, self.op.nodes)
13675 # We want to lock all the affected nodes and groups. We have readily
13676 # available the list of nodes, and the *destination* group. To gather the
13677 # list of "source" groups, we need to fetch node information later on.
13678 self.needed_locks = {
13679 locking.LEVEL_NODEGROUP: set([self.group_uuid]),
13680 locking.LEVEL_NODE: self.op.nodes,
13683 def DeclareLocks(self, level):
13684 if level == locking.LEVEL_NODEGROUP:
13685 assert len(self.needed_locks[locking.LEVEL_NODEGROUP]) == 1
13687 # Try to get all affected nodes' groups without having the group or node
13688 # lock yet. Needs verification later in the code flow.
13689 groups = self.cfg.GetNodeGroupsFromNodes(self.op.nodes)
13691 self.needed_locks[locking.LEVEL_NODEGROUP].update(groups)
13693 def CheckPrereq(self):
13694 """Check prerequisites.
13697 assert self.needed_locks[locking.LEVEL_NODEGROUP]
13698 assert (frozenset(self.owned_locks(locking.LEVEL_NODE)) ==
13699 frozenset(self.op.nodes))
13701 expected_locks = (set([self.group_uuid]) |
13702 self.cfg.GetNodeGroupsFromNodes(self.op.nodes))
13703 actual_locks = self.owned_locks(locking.LEVEL_NODEGROUP)
13704 if actual_locks != expected_locks:
13705 raise errors.OpExecError("Nodes changed groups since locks were acquired,"
13706 " current groups are '%s', used to be '%s'" %
13707 (utils.CommaJoin(expected_locks),
13708 utils.CommaJoin(actual_locks)))
13710 self.node_data = self.cfg.GetAllNodesInfo()
13711 self.group = self.cfg.GetNodeGroup(self.group_uuid)
13712 instance_data = self.cfg.GetAllInstancesInfo()
13714 if self.group is None:
13715 raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
13716 (self.op.group_name, self.group_uuid))
13718 (new_splits, previous_splits) = \
13719 self.CheckAssignmentForSplitInstances([(node, self.group_uuid)
13720 for node in self.op.nodes],
13721 self.node_data, instance_data)
13724 fmt_new_splits = utils.CommaJoin(utils.NiceSort(new_splits))
13726 if not self.op.force:
13727 raise errors.OpExecError("The following instances get split by this"
13728 " change and --force was not given: %s" %
13731 self.LogWarning("This operation will split the following instances: %s",
13734 if previous_splits:
13735 self.LogWarning("In addition, these already-split instances continue"
13736 " to be split across groups: %s",
13737 utils.CommaJoin(utils.NiceSort(previous_splits)))
13739 def Exec(self, feedback_fn):
13740 """Assign nodes to a new group.
13743 mods = [(node_name, self.group_uuid) for node_name in self.op.nodes]
13745 self.cfg.AssignGroupNodes(mods)
13748 def CheckAssignmentForSplitInstances(changes, node_data, instance_data):
13749 """Check for split instances after a node assignment.
13751 This method considers a series of node assignments as an atomic operation,
13752 and returns information about split instances after applying the set of
13755 In particular, it returns information about newly split instances, and
13756 instances that were already split, and remain so after the change.
13758 Only instances whose disk template is listed in constants.DTS_INT_MIRROR are
13761 @type changes: list of (node_name, new_group_uuid) pairs.
13762 @param changes: list of node assignments to consider.
13763 @param node_data: a dict with data for all nodes
13764 @param instance_data: a dict with all instances to consider
13765 @rtype: a two-tuple
13766 @return: a list of instances that were previously okay and result split as a
13767 consequence of this change, and a list of instances that were previously
13768 split and this change does not fix.
13771 changed_nodes = dict((node, group) for node, group in changes
13772 if node_data[node].group != group)
13774 all_split_instances = set()
13775 previously_split_instances = set()
13777 def InstanceNodes(instance):
13778 return [instance.primary_node] + list(instance.secondary_nodes)
13780 for inst in instance_data.values():
13781 if inst.disk_template not in constants.DTS_INT_MIRROR:
13784 instance_nodes = InstanceNodes(inst)
13786 if len(set(node_data[node].group for node in instance_nodes)) > 1:
13787 previously_split_instances.add(inst.name)
13789 if len(set(changed_nodes.get(node, node_data[node].group)
13790 for node in instance_nodes)) > 1:
13791 all_split_instances.add(inst.name)
13793 return (list(all_split_instances - previously_split_instances),
13794 list(previously_split_instances & all_split_instances))
13797 class _GroupQuery(_QueryBase):
13798 FIELDS = query.GROUP_FIELDS
13800 def ExpandNames(self, lu):
13801 lu.needed_locks = {}
13803 self._all_groups = lu.cfg.GetAllNodeGroupsInfo()
13804 self._cluster = lu.cfg.GetClusterInfo()
13805 name_to_uuid = dict((g.name, g.uuid) for g in self._all_groups.values())
13808 self.wanted = [name_to_uuid[name]
13809 for name in utils.NiceSort(name_to_uuid.keys())]
13811 # Accept names to be either names or UUIDs.
13814 all_uuid = frozenset(self._all_groups.keys())
13816 for name in self.names:
13817 if name in all_uuid:
13818 self.wanted.append(name)
13819 elif name in name_to_uuid:
13820 self.wanted.append(name_to_uuid[name])
13822 missing.append(name)
13825 raise errors.OpPrereqError("Some groups do not exist: %s" %
13826 utils.CommaJoin(missing),
13827 errors.ECODE_NOENT)
13829 def DeclareLocks(self, lu, level):
13832 def _GetQueryData(self, lu):
13833 """Computes the list of node groups and their attributes.
13836 do_nodes = query.GQ_NODE in self.requested_data
13837 do_instances = query.GQ_INST in self.requested_data
13839 group_to_nodes = None
13840 group_to_instances = None
13842 # For GQ_NODE, we need to map group->[nodes], and group->[instances] for
13843 # GQ_INST. The former is attainable with just GetAllNodesInfo(), but for the
13844 # latter GetAllInstancesInfo() is not enough, for we have to go through
13845 # instance->node. Hence, we will need to process nodes even if we only need
13846 # instance information.
13847 if do_nodes or do_instances:
13848 all_nodes = lu.cfg.GetAllNodesInfo()
13849 group_to_nodes = dict((uuid, []) for uuid in self.wanted)
13852 for node in all_nodes.values():
13853 if node.group in group_to_nodes:
13854 group_to_nodes[node.group].append(node.name)
13855 node_to_group[node.name] = node.group
13858 all_instances = lu.cfg.GetAllInstancesInfo()
13859 group_to_instances = dict((uuid, []) for uuid in self.wanted)
13861 for instance in all_instances.values():
13862 node = instance.primary_node
13863 if node in node_to_group:
13864 group_to_instances[node_to_group[node]].append(instance.name)
13867 # Do not pass on node information if it was not requested.
13868 group_to_nodes = None
13870 return query.GroupQueryData(self._cluster,
13871 [self._all_groups[uuid]
13872 for uuid in self.wanted],
13873 group_to_nodes, group_to_instances,
13874 query.GQ_DISKPARAMS in self.requested_data)
13877 class LUGroupQuery(NoHooksLU):
13878 """Logical unit for querying node groups.
13883 def CheckArguments(self):
13884 self.gq = _GroupQuery(qlang.MakeSimpleFilter("name", self.op.names),
13885 self.op.output_fields, False)
13887 def ExpandNames(self):
13888 self.gq.ExpandNames(self)
13890 def DeclareLocks(self, level):
13891 self.gq.DeclareLocks(self, level)
13893 def Exec(self, feedback_fn):
13894 return self.gq.OldStyleQuery(self)
13897 class LUGroupSetParams(LogicalUnit):
13898 """Modifies the parameters of a node group.
13901 HPATH = "group-modify"
13902 HTYPE = constants.HTYPE_GROUP
13905 def CheckArguments(self):
13908 self.op.diskparams,
13909 self.op.alloc_policy,
13911 self.op.disk_state,
13915 if all_changes.count(None) == len(all_changes):
13916 raise errors.OpPrereqError("Please pass at least one modification",
13917 errors.ECODE_INVAL)
13919 def ExpandNames(self):
13920 # This raises errors.OpPrereqError on its own:
13921 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
13923 self.needed_locks = {
13924 locking.LEVEL_INSTANCE: [],
13925 locking.LEVEL_NODEGROUP: [self.group_uuid],
13928 self.share_locks[locking.LEVEL_INSTANCE] = 1
13930 def DeclareLocks(self, level):
13931 if level == locking.LEVEL_INSTANCE:
13932 assert not self.needed_locks[locking.LEVEL_INSTANCE]
13934 # Lock instances optimistically, needs verification once group lock has
13936 self.needed_locks[locking.LEVEL_INSTANCE] = \
13937 self.cfg.GetNodeGroupInstances(self.group_uuid)
13940 def _UpdateAndVerifyDiskParams(old, new):
13941 """Updates and verifies disk parameters.
13944 new_params = _GetUpdatedParams(old, new)
13945 utils.ForceDictType(new_params, constants.DISK_DT_TYPES)
13948 def CheckPrereq(self):
13949 """Check prerequisites.
13952 owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
13954 # Check if locked instances are still correct
13955 _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
13957 self.group = self.cfg.GetNodeGroup(self.group_uuid)
13958 cluster = self.cfg.GetClusterInfo()
13960 if self.group is None:
13961 raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
13962 (self.op.group_name, self.group_uuid))
13964 if self.op.ndparams:
13965 new_ndparams = _GetUpdatedParams(self.group.ndparams, self.op.ndparams)
13966 utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
13967 self.new_ndparams = new_ndparams
13969 if self.op.diskparams:
13970 diskparams = self.group.diskparams
13971 uavdp = self._UpdateAndVerifyDiskParams
13972 # For each disktemplate subdict update and verify the values
13973 new_diskparams = dict((dt,
13974 uavdp(diskparams.get(dt, {}),
13975 self.op.diskparams[dt]))
13976 for dt in constants.DISK_TEMPLATES
13977 if dt in self.op.diskparams)
13978 # As we've all subdicts of diskparams ready, lets merge the actual
13979 # dict with all updated subdicts
13980 self.new_diskparams = objects.FillDict(diskparams, new_diskparams)
13982 utils.VerifyDictOptions(self.new_diskparams, constants.DISK_DT_DEFAULTS)
13983 except errors.OpPrereqError, err:
13984 raise errors.OpPrereqError("While verify diskparams options: %s" % err,
13985 errors.ECODE_INVAL)
13987 if self.op.hv_state:
13988 self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state,
13989 self.group.hv_state_static)
13991 if self.op.disk_state:
13992 self.new_disk_state = \
13993 _MergeAndVerifyDiskState(self.op.disk_state,
13994 self.group.disk_state_static)
13996 if self.op.ipolicy:
13997 self.new_ipolicy = _GetUpdatedIPolicy(self.group.ipolicy,
14001 new_ipolicy = cluster.SimpleFillIPolicy(self.new_ipolicy)
14002 inst_filter = lambda inst: inst.name in owned_instances
14003 instances = self.cfg.GetInstancesInfoByFilter(inst_filter).values()
14005 _ComputeNewInstanceViolations(_CalculateGroupIPolicy(cluster,
14007 new_ipolicy, instances)
14010 self.LogWarning("After the ipolicy change the following instances"
14011 " violate them: %s",
14012 utils.CommaJoin(violations))
14014 def BuildHooksEnv(self):
14015 """Build hooks env.
14019 "GROUP_NAME": self.op.group_name,
14020 "NEW_ALLOC_POLICY": self.op.alloc_policy,
14023 def BuildHooksNodes(self):
14024 """Build hooks nodes.
14027 mn = self.cfg.GetMasterNode()
14028 return ([mn], [mn])
14030 def Exec(self, feedback_fn):
14031 """Modifies the node group.
14036 if self.op.ndparams:
14037 self.group.ndparams = self.new_ndparams
14038 result.append(("ndparams", str(self.group.ndparams)))
14040 if self.op.diskparams:
14041 self.group.diskparams = self.new_diskparams
14042 result.append(("diskparams", str(self.group.diskparams)))
14044 if self.op.alloc_policy:
14045 self.group.alloc_policy = self.op.alloc_policy
14047 if self.op.hv_state:
14048 self.group.hv_state_static = self.new_hv_state
14050 if self.op.disk_state:
14051 self.group.disk_state_static = self.new_disk_state
14053 if self.op.ipolicy:
14054 self.group.ipolicy = self.new_ipolicy
14056 self.cfg.Update(self.group, feedback_fn)
14060 class LUGroupRemove(LogicalUnit):
14061 HPATH = "group-remove"
14062 HTYPE = constants.HTYPE_GROUP
14065 def ExpandNames(self):
14066 # This will raises errors.OpPrereqError on its own:
14067 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
14068 self.needed_locks = {
14069 locking.LEVEL_NODEGROUP: [self.group_uuid],
14072 def CheckPrereq(self):
14073 """Check prerequisites.
14075 This checks that the given group name exists as a node group, that is
14076 empty (i.e., contains no nodes), and that is not the last group of the
14080 # Verify that the group is empty.
14081 group_nodes = [node.name
14082 for node in self.cfg.GetAllNodesInfo().values()
14083 if node.group == self.group_uuid]
14086 raise errors.OpPrereqError("Group '%s' not empty, has the following"
14088 (self.op.group_name,
14089 utils.CommaJoin(utils.NiceSort(group_nodes))),
14090 errors.ECODE_STATE)
14092 # Verify the cluster would not be left group-less.
14093 if len(self.cfg.GetNodeGroupList()) == 1:
14094 raise errors.OpPrereqError("Group '%s' is the only group,"
14095 " cannot be removed" %
14096 self.op.group_name,
14097 errors.ECODE_STATE)
14099 def BuildHooksEnv(self):
14100 """Build hooks env.
14104 "GROUP_NAME": self.op.group_name,
14107 def BuildHooksNodes(self):
14108 """Build hooks nodes.
14111 mn = self.cfg.GetMasterNode()
14112 return ([mn], [mn])
14114 def Exec(self, feedback_fn):
14115 """Remove the node group.
14119 self.cfg.RemoveNodeGroup(self.group_uuid)
14120 except errors.ConfigurationError:
14121 raise errors.OpExecError("Group '%s' with UUID %s disappeared" %
14122 (self.op.group_name, self.group_uuid))
14124 self.remove_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
14127 class LUGroupRename(LogicalUnit):
14128 HPATH = "group-rename"
14129 HTYPE = constants.HTYPE_GROUP
14132 def ExpandNames(self):
14133 # This raises errors.OpPrereqError on its own:
14134 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
14136 self.needed_locks = {
14137 locking.LEVEL_NODEGROUP: [self.group_uuid],
14140 def CheckPrereq(self):
14141 """Check prerequisites.
14143 Ensures requested new name is not yet used.
14147 new_name_uuid = self.cfg.LookupNodeGroup(self.op.new_name)
14148 except errors.OpPrereqError:
14151 raise errors.OpPrereqError("Desired new name '%s' clashes with existing"
14152 " node group (UUID: %s)" %
14153 (self.op.new_name, new_name_uuid),
14154 errors.ECODE_EXISTS)
14156 def BuildHooksEnv(self):
14157 """Build hooks env.
14161 "OLD_NAME": self.op.group_name,
14162 "NEW_NAME": self.op.new_name,
14165 def BuildHooksNodes(self):
14166 """Build hooks nodes.
14169 mn = self.cfg.GetMasterNode()
14171 all_nodes = self.cfg.GetAllNodesInfo()
14172 all_nodes.pop(mn, None)
14175 run_nodes.extend(node.name for node in all_nodes.values()
14176 if node.group == self.group_uuid)
14178 return (run_nodes, run_nodes)
14180 def Exec(self, feedback_fn):
14181 """Rename the node group.
14184 group = self.cfg.GetNodeGroup(self.group_uuid)
14187 raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
14188 (self.op.group_name, self.group_uuid))
14190 group.name = self.op.new_name
14191 self.cfg.Update(group, feedback_fn)
14193 return self.op.new_name
14196 class LUGroupEvacuate(LogicalUnit):
14197 HPATH = "group-evacuate"
14198 HTYPE = constants.HTYPE_GROUP
14201 def ExpandNames(self):
14202 # This raises errors.OpPrereqError on its own:
14203 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
14205 if self.op.target_groups:
14206 self.req_target_uuids = map(self.cfg.LookupNodeGroup,
14207 self.op.target_groups)
14209 self.req_target_uuids = []
14211 if self.group_uuid in self.req_target_uuids:
14212 raise errors.OpPrereqError("Group to be evacuated (%s) can not be used"
14213 " as a target group (targets are %s)" %
14215 utils.CommaJoin(self.req_target_uuids)),
14216 errors.ECODE_INVAL)
14218 self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
14220 self.share_locks = _ShareAll()
14221 self.needed_locks = {
14222 locking.LEVEL_INSTANCE: [],
14223 locking.LEVEL_NODEGROUP: [],
14224 locking.LEVEL_NODE: [],
14227 def DeclareLocks(self, level):
14228 if level == locking.LEVEL_INSTANCE:
14229 assert not self.needed_locks[locking.LEVEL_INSTANCE]
14231 # Lock instances optimistically, needs verification once node and group
14232 # locks have been acquired
14233 self.needed_locks[locking.LEVEL_INSTANCE] = \
14234 self.cfg.GetNodeGroupInstances(self.group_uuid)
14236 elif level == locking.LEVEL_NODEGROUP:
14237 assert not self.needed_locks[locking.LEVEL_NODEGROUP]
14239 if self.req_target_uuids:
14240 lock_groups = set([self.group_uuid] + self.req_target_uuids)
14242 # Lock all groups used by instances optimistically; this requires going
14243 # via the node before it's locked, requiring verification later on
14244 lock_groups.update(group_uuid
14245 for instance_name in
14246 self.owned_locks(locking.LEVEL_INSTANCE)
14248 self.cfg.GetInstanceNodeGroups(instance_name))
14250 # No target groups, need to lock all of them
14251 lock_groups = locking.ALL_SET
14253 self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
14255 elif level == locking.LEVEL_NODE:
14256 # This will only lock the nodes in the group to be evacuated which
14257 # contain actual instances
14258 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
14259 self._LockInstancesNodes()
14261 # Lock all nodes in group to be evacuated and target groups
14262 owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
14263 assert self.group_uuid in owned_groups
14264 member_nodes = [node_name
14265 for group in owned_groups
14266 for node_name in self.cfg.GetNodeGroup(group).members]
14267 self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
14269 def CheckPrereq(self):
14270 owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
14271 owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
14272 owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
14274 assert owned_groups.issuperset(self.req_target_uuids)
14275 assert self.group_uuid in owned_groups
14277 # Check if locked instances are still correct
14278 _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
14280 # Get instance information
14281 self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
14283 # Check if node groups for locked instances are still correct
14284 _CheckInstancesNodeGroups(self.cfg, self.instances,
14285 owned_groups, owned_nodes, self.group_uuid)
14287 if self.req_target_uuids:
14288 # User requested specific target groups
14289 self.target_uuids = self.req_target_uuids
14291 # All groups except the one to be evacuated are potential targets
14292 self.target_uuids = [group_uuid for group_uuid in owned_groups
14293 if group_uuid != self.group_uuid]
14295 if not self.target_uuids:
14296 raise errors.OpPrereqError("There are no possible target groups",
14297 errors.ECODE_INVAL)
14299 def BuildHooksEnv(self):
14300 """Build hooks env.
14304 "GROUP_NAME": self.op.group_name,
14305 "TARGET_GROUPS": " ".join(self.target_uuids),
14308 def BuildHooksNodes(self):
14309 """Build hooks nodes.
14312 mn = self.cfg.GetMasterNode()
14314 assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
14316 run_nodes = [mn] + self.cfg.GetNodeGroup(self.group_uuid).members
14318 return (run_nodes, run_nodes)
14320 def Exec(self, feedback_fn):
14321 instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
14323 assert self.group_uuid not in self.target_uuids
14325 ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_CHG_GROUP,
14326 instances=instances, target_groups=self.target_uuids)
14328 ial.Run(self.op.iallocator)
14330 if not ial.success:
14331 raise errors.OpPrereqError("Can't compute group evacuation using"
14332 " iallocator '%s': %s" %
14333 (self.op.iallocator, ial.info),
14334 errors.ECODE_NORES)
14336 jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, False)
14338 self.LogInfo("Iallocator returned %s job(s) for evacuating node group %s",
14339 len(jobs), self.op.group_name)
14341 return ResultWithJobs(jobs)
14344 class TagsLU(NoHooksLU): # pylint: disable=W0223
14345 """Generic tags LU.
14347 This is an abstract class which is the parent of all the other tags LUs.
14350 def ExpandNames(self):
14351 self.group_uuid = None
14352 self.needed_locks = {}
14354 if self.op.kind == constants.TAG_NODE:
14355 self.op.name = _ExpandNodeName(self.cfg, self.op.name)
14356 lock_level = locking.LEVEL_NODE
14357 lock_name = self.op.name
14358 elif self.op.kind == constants.TAG_INSTANCE:
14359 self.op.name = _ExpandInstanceName(self.cfg, self.op.name)
14360 lock_level = locking.LEVEL_INSTANCE
14361 lock_name = self.op.name
14362 elif self.op.kind == constants.TAG_NODEGROUP:
14363 self.group_uuid = self.cfg.LookupNodeGroup(self.op.name)
14364 lock_level = locking.LEVEL_NODEGROUP
14365 lock_name = self.group_uuid
14370 if lock_level and getattr(self.op, "use_locking", True):
14371 self.needed_locks[lock_level] = lock_name
14373 # FIXME: Acquire BGL for cluster tag operations (as of this writing it's
14374 # not possible to acquire the BGL based on opcode parameters)
14376 def CheckPrereq(self):
14377 """Check prerequisites.
14380 if self.op.kind == constants.TAG_CLUSTER:
14381 self.target = self.cfg.GetClusterInfo()
14382 elif self.op.kind == constants.TAG_NODE:
14383 self.target = self.cfg.GetNodeInfo(self.op.name)
14384 elif self.op.kind == constants.TAG_INSTANCE:
14385 self.target = self.cfg.GetInstanceInfo(self.op.name)
14386 elif self.op.kind == constants.TAG_NODEGROUP:
14387 self.target = self.cfg.GetNodeGroup(self.group_uuid)
14389 raise errors.OpPrereqError("Wrong tag type requested (%s)" %
14390 str(self.op.kind), errors.ECODE_INVAL)
14393 class LUTagsGet(TagsLU):
14394 """Returns the tags of a given object.
14399 def ExpandNames(self):
14400 TagsLU.ExpandNames(self)
14402 # Share locks as this is only a read operation
14403 self.share_locks = _ShareAll()
14405 def Exec(self, feedback_fn):
14406 """Returns the tag list.
14409 return list(self.target.GetTags())
14412 class LUTagsSearch(NoHooksLU):
14413 """Searches the tags for a given pattern.
14418 def ExpandNames(self):
14419 self.needed_locks = {}
14421 def CheckPrereq(self):
14422 """Check prerequisites.
14424 This checks the pattern passed for validity by compiling it.
14428 self.re = re.compile(self.op.pattern)
14429 except re.error, err:
14430 raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
14431 (self.op.pattern, err), errors.ECODE_INVAL)
14433 def Exec(self, feedback_fn):
14434 """Returns the tag list.
14438 tgts = [("/cluster", cfg.GetClusterInfo())]
14439 ilist = cfg.GetAllInstancesInfo().values()
14440 tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
14441 nlist = cfg.GetAllNodesInfo().values()
14442 tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
14443 tgts.extend(("/nodegroup/%s" % n.name, n)
14444 for n in cfg.GetAllNodeGroupsInfo().values())
14446 for path, target in tgts:
14447 for tag in target.GetTags():
14448 if self.re.search(tag):
14449 results.append((path, tag))
14453 class LUTagsSet(TagsLU):
14454 """Sets a tag on a given object.
14459 def CheckPrereq(self):
14460 """Check prerequisites.
14462 This checks the type and length of the tag name and value.
14465 TagsLU.CheckPrereq(self)
14466 for tag in self.op.tags:
14467 objects.TaggableObject.ValidateTag(tag)
14469 def Exec(self, feedback_fn):
14474 for tag in self.op.tags:
14475 self.target.AddTag(tag)
14476 except errors.TagError, err:
14477 raise errors.OpExecError("Error while setting tag: %s" % str(err))
14478 self.cfg.Update(self.target, feedback_fn)
14481 class LUTagsDel(TagsLU):
14482 """Delete a list of tags from a given object.
14487 def CheckPrereq(self):
14488 """Check prerequisites.
14490 This checks that we have the given tag.
14493 TagsLU.CheckPrereq(self)
14494 for tag in self.op.tags:
14495 objects.TaggableObject.ValidateTag(tag)
14496 del_tags = frozenset(self.op.tags)
14497 cur_tags = self.target.GetTags()
14499 diff_tags = del_tags - cur_tags
14501 diff_names = ("'%s'" % i for i in sorted(diff_tags))
14502 raise errors.OpPrereqError("Tag(s) %s not found" %
14503 (utils.CommaJoin(diff_names), ),
14504 errors.ECODE_NOENT)
14506 def Exec(self, feedback_fn):
14507 """Remove the tag from the object.
14510 for tag in self.op.tags:
14511 self.target.RemoveTag(tag)
14512 self.cfg.Update(self.target, feedback_fn)
14515 class LUTestDelay(NoHooksLU):
14516 """Sleep for a specified amount of time.
14518 This LU sleeps on the master and/or nodes for a specified amount of
14524 def ExpandNames(self):
14525 """Expand names and set required locks.
14527 This expands the node list, if any.
14530 self.needed_locks = {}
14531 if self.op.on_nodes:
14532 # _GetWantedNodes can be used here, but is not always appropriate to use
14533 # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
14534 # more information.
14535 self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
14536 self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
14538 def _TestDelay(self):
14539 """Do the actual sleep.
14542 if self.op.on_master:
14543 if not utils.TestDelay(self.op.duration):
14544 raise errors.OpExecError("Error during master delay test")
14545 if self.op.on_nodes:
14546 result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
14547 for node, node_result in result.items():
14548 node_result.Raise("Failure during rpc call to node %s" % node)
14550 def Exec(self, feedback_fn):
14551 """Execute the test delay opcode, with the wanted repetitions.
14554 if self.op.repeat == 0:
14557 top_value = self.op.repeat - 1
14558 for i in range(self.op.repeat):
14559 self.LogInfo("Test delay iteration %d/%d" % (i, top_value))
14563 class LUTestJqueue(NoHooksLU):
14564 """Utility LU to test some aspects of the job queue.
14569 # Must be lower than default timeout for WaitForJobChange to see whether it
14570 # notices changed jobs
14571 _CLIENT_CONNECT_TIMEOUT = 20.0
14572 _CLIENT_CONFIRM_TIMEOUT = 60.0
14575 def _NotifyUsingSocket(cls, cb, errcls):
14576 """Opens a Unix socket and waits for another program to connect.
14579 @param cb: Callback to send socket name to client
14580 @type errcls: class
14581 @param errcls: Exception class to use for errors
14584 # Using a temporary directory as there's no easy way to create temporary
14585 # sockets without writing a custom loop around tempfile.mktemp and
14587 tmpdir = tempfile.mkdtemp()
14589 tmpsock = utils.PathJoin(tmpdir, "sock")
14591 logging.debug("Creating temporary socket at %s", tmpsock)
14592 sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
14597 # Send details to client
14600 # Wait for client to connect before continuing
14601 sock.settimeout(cls._CLIENT_CONNECT_TIMEOUT)
14603 (conn, _) = sock.accept()
14604 except socket.error, err:
14605 raise errcls("Client didn't connect in time (%s)" % err)
14609 # Remove as soon as client is connected
14610 shutil.rmtree(tmpdir)
14612 # Wait for client to close
14615 # pylint: disable=E1101
14616 # Instance of '_socketobject' has no ... member
14617 conn.settimeout(cls._CLIENT_CONFIRM_TIMEOUT)
14619 except socket.error, err:
14620 raise errcls("Client failed to confirm notification (%s)" % err)
14624 def _SendNotification(self, test, arg, sockname):
14625 """Sends a notification to the client.
14628 @param test: Test name
14629 @param arg: Test argument (depends on test)
14630 @type sockname: string
14631 @param sockname: Socket path
14634 self.Log(constants.ELOG_JQUEUE_TEST, (sockname, test, arg))
14636 def _Notify(self, prereq, test, arg):
14637 """Notifies the client of a test.
14640 @param prereq: Whether this is a prereq-phase test
14642 @param test: Test name
14643 @param arg: Test argument (depends on test)
14647 errcls = errors.OpPrereqError
14649 errcls = errors.OpExecError
14651 return self._NotifyUsingSocket(compat.partial(self._SendNotification,
14655 def CheckArguments(self):
14656 self.checkargs_calls = getattr(self, "checkargs_calls", 0) + 1
14657 self.expandnames_calls = 0
14659 def ExpandNames(self):
14660 checkargs_calls = getattr(self, "checkargs_calls", 0)
14661 if checkargs_calls < 1:
14662 raise errors.ProgrammerError("CheckArguments was not called")
14664 self.expandnames_calls += 1
14666 if self.op.notify_waitlock:
14667 self._Notify(True, constants.JQT_EXPANDNAMES, None)
14669 self.LogInfo("Expanding names")
14671 # Get lock on master node (just to get a lock, not for a particular reason)
14672 self.needed_locks = {
14673 locking.LEVEL_NODE: self.cfg.GetMasterNode(),
14676 def Exec(self, feedback_fn):
14677 if self.expandnames_calls < 1:
14678 raise errors.ProgrammerError("ExpandNames was not called")
14680 if self.op.notify_exec:
14681 self._Notify(False, constants.JQT_EXEC, None)
14683 self.LogInfo("Executing")
14685 if self.op.log_messages:
14686 self._Notify(False, constants.JQT_STARTMSG, len(self.op.log_messages))
14687 for idx, msg in enumerate(self.op.log_messages):
14688 self.LogInfo("Sending log message %s", idx + 1)
14689 feedback_fn(constants.JQT_MSGPREFIX + msg)
14690 # Report how many test messages have been sent
14691 self._Notify(False, constants.JQT_LOGMSG, idx + 1)
14694 raise errors.OpExecError("Opcode failure was requested")
14699 class IAllocator(object):
14700 """IAllocator framework.
14702 An IAllocator instance has three sets of attributes:
14703 - cfg that is needed to query the cluster
14704 - input data (all members of the _KEYS class attribute are required)
14705 - four buffer attributes (in|out_data|text), that represent the
14706 input (to the external script) in text and data structure format,
14707 and the output from it, again in two formats
14708 - the result variables from the script (success, info, nodes) for
14712 # pylint: disable=R0902
14713 # lots of instance attributes
14715 def __init__(self, cfg, rpc_runner, mode, **kwargs):
14717 self.rpc = rpc_runner
14718 # init buffer variables
14719 self.in_text = self.out_text = self.in_data = self.out_data = None
14720 # init all input fields so that pylint is happy
14722 self.memory = self.disks = self.disk_template = self.spindle_use = None
14723 self.os = self.tags = self.nics = self.vcpus = None
14724 self.hypervisor = None
14725 self.relocate_from = None
14727 self.instances = None
14728 self.evac_mode = None
14729 self.target_groups = []
14731 self.required_nodes = None
14732 # init result fields
14733 self.success = self.info = self.result = None
14736 (fn, keydata, self._result_check) = self._MODE_DATA[self.mode]
14738 raise errors.ProgrammerError("Unknown mode '%s' passed to the"
14739 " IAllocator" % self.mode)
14741 keyset = [n for (n, _) in keydata]
14744 if key not in keyset:
14745 raise errors.ProgrammerError("Invalid input parameter '%s' to"
14746 " IAllocator" % key)
14747 setattr(self, key, kwargs[key])
14750 if key not in kwargs:
14751 raise errors.ProgrammerError("Missing input parameter '%s' to"
14752 " IAllocator" % key)
14753 self._BuildInputData(compat.partial(fn, self), keydata)
14755 def _ComputeClusterData(self):
14756 """Compute the generic allocator input data.
14758 This is the data that is independent of the actual operation.
14762 cluster_info = cfg.GetClusterInfo()
14765 "version": constants.IALLOCATOR_VERSION,
14766 "cluster_name": cfg.GetClusterName(),
14767 "cluster_tags": list(cluster_info.GetTags()),
14768 "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
14769 "ipolicy": cluster_info.ipolicy,
14771 ninfo = cfg.GetAllNodesInfo()
14772 iinfo = cfg.GetAllInstancesInfo().values()
14773 i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
14776 node_list = [n.name for n in ninfo.values() if n.vm_capable]
14778 if self.mode == constants.IALLOCATOR_MODE_ALLOC:
14779 hypervisor_name = self.hypervisor
14780 elif self.mode == constants.IALLOCATOR_MODE_RELOC:
14781 hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
14783 hypervisor_name = cluster_info.primary_hypervisor
14785 node_data = self.rpc.call_node_info(node_list, [cfg.GetVGName()],
14788 self.rpc.call_all_instances_info(node_list,
14789 cluster_info.enabled_hypervisors)
14791 data["nodegroups"] = self._ComputeNodeGroupData(cfg)
14793 config_ndata = self._ComputeBasicNodeData(cfg, ninfo)
14794 data["nodes"] = self._ComputeDynamicNodeData(ninfo, node_data, node_iinfo,
14795 i_list, config_ndata)
14796 assert len(data["nodes"]) == len(ninfo), \
14797 "Incomplete node data computed"
14799 data["instances"] = self._ComputeInstanceData(cluster_info, i_list)
14801 self.in_data = data
14804 def _ComputeNodeGroupData(cfg):
14805 """Compute node groups data.
14808 cluster = cfg.GetClusterInfo()
14809 ng = dict((guuid, {
14810 "name": gdata.name,
14811 "alloc_policy": gdata.alloc_policy,
14812 "ipolicy": _CalculateGroupIPolicy(cluster, gdata),
14814 for guuid, gdata in cfg.GetAllNodeGroupsInfo().items())
14819 def _ComputeBasicNodeData(cfg, node_cfg):
14820 """Compute global node data.
14823 @returns: a dict of name: (node dict, node config)
14826 # fill in static (config-based) values
14827 node_results = dict((ninfo.name, {
14828 "tags": list(ninfo.GetTags()),
14829 "primary_ip": ninfo.primary_ip,
14830 "secondary_ip": ninfo.secondary_ip,
14831 "offline": ninfo.offline,
14832 "drained": ninfo.drained,
14833 "master_candidate": ninfo.master_candidate,
14834 "group": ninfo.group,
14835 "master_capable": ninfo.master_capable,
14836 "vm_capable": ninfo.vm_capable,
14837 "ndparams": cfg.GetNdParams(ninfo),
14839 for ninfo in node_cfg.values())
14841 return node_results
14844 def _ComputeDynamicNodeData(node_cfg, node_data, node_iinfo, i_list,
14846 """Compute global node data.
14848 @param node_results: the basic node structures as filled from the config
14851 #TODO(dynmem): compute the right data on MAX and MIN memory
14852 # make a copy of the current dict
14853 node_results = dict(node_results)
14854 for nname, nresult in node_data.items():
14855 assert nname in node_results, "Missing basic data for node %s" % nname
14856 ninfo = node_cfg[nname]
14858 if not (ninfo.offline or ninfo.drained):
14859 nresult.Raise("Can't get data for node %s" % nname)
14860 node_iinfo[nname].Raise("Can't get node instance info from node %s" %
14862 remote_info = _MakeLegacyNodeInfo(nresult.payload)
14864 for attr in ["memory_total", "memory_free", "memory_dom0",
14865 "vg_size", "vg_free", "cpu_total"]:
14866 if attr not in remote_info:
14867 raise errors.OpExecError("Node '%s' didn't return attribute"
14868 " '%s'" % (nname, attr))
14869 if not isinstance(remote_info[attr], int):
14870 raise errors.OpExecError("Node '%s' returned invalid value"
14872 (nname, attr, remote_info[attr]))
14873 # compute memory used by primary instances
14874 i_p_mem = i_p_up_mem = 0
14875 for iinfo, beinfo in i_list:
14876 if iinfo.primary_node == nname:
14877 i_p_mem += beinfo[constants.BE_MAXMEM]
14878 if iinfo.name not in node_iinfo[nname].payload:
14881 i_used_mem = int(node_iinfo[nname].payload[iinfo.name]["memory"])
14882 i_mem_diff = beinfo[constants.BE_MAXMEM] - i_used_mem
14883 remote_info["memory_free"] -= max(0, i_mem_diff)
14885 if iinfo.admin_state == constants.ADMINST_UP:
14886 i_p_up_mem += beinfo[constants.BE_MAXMEM]
14888 # compute memory used by instances
14890 "total_memory": remote_info["memory_total"],
14891 "reserved_memory": remote_info["memory_dom0"],
14892 "free_memory": remote_info["memory_free"],
14893 "total_disk": remote_info["vg_size"],
14894 "free_disk": remote_info["vg_free"],
14895 "total_cpus": remote_info["cpu_total"],
14896 "i_pri_memory": i_p_mem,
14897 "i_pri_up_memory": i_p_up_mem,
14899 pnr_dyn.update(node_results[nname])
14900 node_results[nname] = pnr_dyn
14902 return node_results
14905 def _ComputeInstanceData(cluster_info, i_list):
14906 """Compute global instance data.
14910 for iinfo, beinfo in i_list:
14912 for nic in iinfo.nics:
14913 filled_params = cluster_info.SimpleFillNIC(nic.nicparams)
14917 "mode": filled_params[constants.NIC_MODE],
14918 "link": filled_params[constants.NIC_LINK],
14920 if filled_params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
14921 nic_dict["bridge"] = filled_params[constants.NIC_LINK]
14922 nic_data.append(nic_dict)
14924 "tags": list(iinfo.GetTags()),
14925 "admin_state": iinfo.admin_state,
14926 "vcpus": beinfo[constants.BE_VCPUS],
14927 "memory": beinfo[constants.BE_MAXMEM],
14928 "spindle_use": beinfo[constants.BE_SPINDLE_USE],
14930 "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
14932 "disks": [{constants.IDISK_SIZE: dsk.size,
14933 constants.IDISK_MODE: dsk.mode}
14934 for dsk in iinfo.disks],
14935 "disk_template": iinfo.disk_template,
14936 "hypervisor": iinfo.hypervisor,
14938 pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
14940 instance_data[iinfo.name] = pir
14942 return instance_data
14944 def _AddNewInstance(self):
14945 """Add new instance data to allocator structure.
14947 This in combination with _AllocatorGetClusterData will create the
14948 correct structure needed as input for the allocator.
14950 The checks for the completeness of the opcode must have already been
14954 disk_space = _ComputeDiskSize(self.disk_template, self.disks)
14956 if self.disk_template in constants.DTS_INT_MIRROR:
14957 self.required_nodes = 2
14959 self.required_nodes = 1
14963 "disk_template": self.disk_template,
14966 "vcpus": self.vcpus,
14967 "memory": self.memory,
14968 "spindle_use": self.spindle_use,
14969 "disks": self.disks,
14970 "disk_space_total": disk_space,
14972 "required_nodes": self.required_nodes,
14973 "hypervisor": self.hypervisor,
14978 def _AddRelocateInstance(self):
14979 """Add relocate instance data to allocator structure.
14981 This in combination with _IAllocatorGetClusterData will create the
14982 correct structure needed as input for the allocator.
14984 The checks for the completeness of the opcode must have already been
14988 instance = self.cfg.GetInstanceInfo(self.name)
14989 if instance is None:
14990 raise errors.ProgrammerError("Unknown instance '%s' passed to"
14991 " IAllocator" % self.name)
14993 if instance.disk_template not in constants.DTS_MIRRORED:
14994 raise errors.OpPrereqError("Can't relocate non-mirrored instances",
14995 errors.ECODE_INVAL)
14997 if instance.disk_template in constants.DTS_INT_MIRROR and \
14998 len(instance.secondary_nodes) != 1:
14999 raise errors.OpPrereqError("Instance has not exactly one secondary node",
15000 errors.ECODE_STATE)
15002 self.required_nodes = 1
15003 disk_sizes = [{constants.IDISK_SIZE: disk.size} for disk in instance.disks]
15004 disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
15008 "disk_space_total": disk_space,
15009 "required_nodes": self.required_nodes,
15010 "relocate_from": self.relocate_from,
15014 def _AddNodeEvacuate(self):
15015 """Get data for node-evacuate requests.
15019 "instances": self.instances,
15020 "evac_mode": self.evac_mode,
15023 def _AddChangeGroup(self):
15024 """Get data for node-evacuate requests.
15028 "instances": self.instances,
15029 "target_groups": self.target_groups,
15032 def _BuildInputData(self, fn, keydata):
15033 """Build input data structures.
15036 self._ComputeClusterData()
15039 request["type"] = self.mode
15040 for keyname, keytype in keydata:
15041 if keyname not in request:
15042 raise errors.ProgrammerError("Request parameter %s is missing" %
15044 val = request[keyname]
15045 if not keytype(val):
15046 raise errors.ProgrammerError("Request parameter %s doesn't pass"
15047 " validation, value %s, expected"
15048 " type %s" % (keyname, val, keytype))
15049 self.in_data["request"] = request
15051 self.in_text = serializer.Dump(self.in_data)
15053 _STRING_LIST = ht.TListOf(ht.TString)
15054 _JOB_LIST = ht.TListOf(ht.TListOf(ht.TStrictDict(True, False, {
15055 # pylint: disable=E1101
15056 # Class '...' has no 'OP_ID' member
15057 "OP_ID": ht.TElemOf([opcodes.OpInstanceFailover.OP_ID,
15058 opcodes.OpInstanceMigrate.OP_ID,
15059 opcodes.OpInstanceReplaceDisks.OP_ID])
15063 ht.TListOf(ht.TAnd(ht.TIsLength(3),
15064 ht.TItems([ht.TNonEmptyString,
15065 ht.TNonEmptyString,
15066 ht.TListOf(ht.TNonEmptyString),
15069 ht.TListOf(ht.TAnd(ht.TIsLength(2),
15070 ht.TItems([ht.TNonEmptyString,
15073 _NEVAC_RESULT = ht.TAnd(ht.TIsLength(3),
15074 ht.TItems([_NEVAC_MOVED, _NEVAC_FAILED, _JOB_LIST]))
15077 constants.IALLOCATOR_MODE_ALLOC:
15080 ("name", ht.TString),
15081 ("memory", ht.TInt),
15082 ("spindle_use", ht.TInt),
15083 ("disks", ht.TListOf(ht.TDict)),
15084 ("disk_template", ht.TString),
15085 ("os", ht.TString),
15086 ("tags", _STRING_LIST),
15087 ("nics", ht.TListOf(ht.TDict)),
15088 ("vcpus", ht.TInt),
15089 ("hypervisor", ht.TString),
15091 constants.IALLOCATOR_MODE_RELOC:
15092 (_AddRelocateInstance,
15093 [("name", ht.TString), ("relocate_from", _STRING_LIST)],
15095 constants.IALLOCATOR_MODE_NODE_EVAC:
15096 (_AddNodeEvacuate, [
15097 ("instances", _STRING_LIST),
15098 ("evac_mode", ht.TElemOf(constants.IALLOCATOR_NEVAC_MODES)),
15100 constants.IALLOCATOR_MODE_CHG_GROUP:
15101 (_AddChangeGroup, [
15102 ("instances", _STRING_LIST),
15103 ("target_groups", _STRING_LIST),
15107 def Run(self, name, validate=True, call_fn=None):
15108 """Run an instance allocator and return the results.
15111 if call_fn is None:
15112 call_fn = self.rpc.call_iallocator_runner
15114 result = call_fn(self.cfg.GetMasterNode(), name, self.in_text)
15115 result.Raise("Failure while running the iallocator script")
15117 self.out_text = result.payload
15119 self._ValidateResult()
15121 def _ValidateResult(self):
15122 """Process the allocator results.
15124 This will process and if successful save the result in
15125 self.out_data and the other parameters.
15129 rdict = serializer.Load(self.out_text)
15130 except Exception, err:
15131 raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
15133 if not isinstance(rdict, dict):
15134 raise errors.OpExecError("Can't parse iallocator results: not a dict")
15136 # TODO: remove backwards compatiblity in later versions
15137 if "nodes" in rdict and "result" not in rdict:
15138 rdict["result"] = rdict["nodes"]
15141 for key in "success", "info", "result":
15142 if key not in rdict:
15143 raise errors.OpExecError("Can't parse iallocator results:"
15144 " missing key '%s'" % key)
15145 setattr(self, key, rdict[key])
15147 if not self._result_check(self.result):
15148 raise errors.OpExecError("Iallocator returned invalid result,"
15149 " expected %s, got %s" %
15150 (self._result_check, self.result),
15151 errors.ECODE_INVAL)
15153 if self.mode == constants.IALLOCATOR_MODE_RELOC:
15154 assert self.relocate_from is not None
15155 assert self.required_nodes == 1
15157 node2group = dict((name, ndata["group"])
15158 for (name, ndata) in self.in_data["nodes"].items())
15160 fn = compat.partial(self._NodesToGroups, node2group,
15161 self.in_data["nodegroups"])
15163 instance = self.cfg.GetInstanceInfo(self.name)
15164 request_groups = fn(self.relocate_from + [instance.primary_node])
15165 result_groups = fn(rdict["result"] + [instance.primary_node])
15167 if self.success and not set(result_groups).issubset(request_groups):
15168 raise errors.OpExecError("Groups of nodes returned by iallocator (%s)"
15169 " differ from original groups (%s)" %
15170 (utils.CommaJoin(result_groups),
15171 utils.CommaJoin(request_groups)))
15173 elif self.mode == constants.IALLOCATOR_MODE_NODE_EVAC:
15174 assert self.evac_mode in constants.IALLOCATOR_NEVAC_MODES
15176 self.out_data = rdict
15179 def _NodesToGroups(node2group, groups, nodes):
15180 """Returns a list of unique group names for a list of nodes.
15182 @type node2group: dict
15183 @param node2group: Map from node name to group UUID
15185 @param groups: Group information
15187 @param nodes: Node names
15194 group_uuid = node2group[node]
15196 # Ignore unknown node
15200 group = groups[group_uuid]
15202 # Can't find group, let's use UUID
15203 group_name = group_uuid
15205 group_name = group["name"]
15207 result.add(group_name)
15209 return sorted(result)
15212 class LUTestAllocator(NoHooksLU):
15213 """Run allocator tests.
15215 This LU runs the allocator tests
15218 def CheckPrereq(self):
15219 """Check prerequisites.
15221 This checks the opcode parameters depending on the director and mode test.
15224 if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
15225 for attr in ["memory", "disks", "disk_template",
15226 "os", "tags", "nics", "vcpus"]:
15227 if not hasattr(self.op, attr):
15228 raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
15229 attr, errors.ECODE_INVAL)
15230 iname = self.cfg.ExpandInstanceName(self.op.name)
15231 if iname is not None:
15232 raise errors.OpPrereqError("Instance '%s' already in the cluster" %
15233 iname, errors.ECODE_EXISTS)
15234 if not isinstance(self.op.nics, list):
15235 raise errors.OpPrereqError("Invalid parameter 'nics'",
15236 errors.ECODE_INVAL)
15237 if not isinstance(self.op.disks, list):
15238 raise errors.OpPrereqError("Invalid parameter 'disks'",
15239 errors.ECODE_INVAL)
15240 for row in self.op.disks:
15241 if (not isinstance(row, dict) or
15242 constants.IDISK_SIZE not in row or
15243 not isinstance(row[constants.IDISK_SIZE], int) or
15244 constants.IDISK_MODE not in row or
15245 row[constants.IDISK_MODE] not in constants.DISK_ACCESS_SET):
15246 raise errors.OpPrereqError("Invalid contents of the 'disks'"
15247 " parameter", errors.ECODE_INVAL)
15248 if self.op.hypervisor is None:
15249 self.op.hypervisor = self.cfg.GetHypervisorType()
15250 elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
15251 fname = _ExpandInstanceName(self.cfg, self.op.name)
15252 self.op.name = fname
15253 self.relocate_from = \
15254 list(self.cfg.GetInstanceInfo(fname).secondary_nodes)
15255 elif self.op.mode in (constants.IALLOCATOR_MODE_CHG_GROUP,
15256 constants.IALLOCATOR_MODE_NODE_EVAC):
15257 if not self.op.instances:
15258 raise errors.OpPrereqError("Missing instances", errors.ECODE_INVAL)
15259 self.op.instances = _GetWantedInstances(self, self.op.instances)
15261 raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
15262 self.op.mode, errors.ECODE_INVAL)
15264 if self.op.direction == constants.IALLOCATOR_DIR_OUT:
15265 if self.op.allocator is None:
15266 raise errors.OpPrereqError("Missing allocator name",
15267 errors.ECODE_INVAL)
15268 elif self.op.direction != constants.IALLOCATOR_DIR_IN:
15269 raise errors.OpPrereqError("Wrong allocator test '%s'" %
15270 self.op.direction, errors.ECODE_INVAL)
15272 def Exec(self, feedback_fn):
15273 """Run the allocator test.
15276 if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
15277 ial = IAllocator(self.cfg, self.rpc,
15280 memory=self.op.memory,
15281 disks=self.op.disks,
15282 disk_template=self.op.disk_template,
15286 vcpus=self.op.vcpus,
15287 hypervisor=self.op.hypervisor,
15289 elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
15290 ial = IAllocator(self.cfg, self.rpc,
15293 relocate_from=list(self.relocate_from),
15295 elif self.op.mode == constants.IALLOCATOR_MODE_CHG_GROUP:
15296 ial = IAllocator(self.cfg, self.rpc,
15298 instances=self.op.instances,
15299 target_groups=self.op.target_groups)
15300 elif self.op.mode == constants.IALLOCATOR_MODE_NODE_EVAC:
15301 ial = IAllocator(self.cfg, self.rpc,
15303 instances=self.op.instances,
15304 evac_mode=self.op.evac_mode)
15306 raise errors.ProgrammerError("Uncatched mode %s in"
15307 " LUTestAllocator.Exec", self.op.mode)
15309 if self.op.direction == constants.IALLOCATOR_DIR_IN:
15310 result = ial.in_text
15312 ial.Run(self.op.allocator, validate=False)
15313 result = ial.out_text
15317 #: Query type implementations
15319 constants.QR_CLUSTER: _ClusterQuery,
15320 constants.QR_INSTANCE: _InstanceQuery,
15321 constants.QR_NODE: _NodeQuery,
15322 constants.QR_GROUP: _GroupQuery,
15323 constants.QR_OS: _OsQuery,
15324 constants.QR_EXPORT: _ExportQuery,
15327 assert set(_QUERY_IMPL.keys()) == constants.QR_VIA_OP
15330 def _GetQueryImplementation(name):
15331 """Returns the implemtnation for a query type.
15333 @param name: Query type, must be one of L{constants.QR_VIA_OP}
15337 return _QUERY_IMPL[name]
15339 raise errors.OpPrereqError("Unknown query resource '%s'" % name,
15340 errors.ECODE_INVAL)