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)
1615 result.Raise("Failed to get disk status from node %s" % node_name,
1616 prereq=prereq, ecode=errors.ECODE_ENVIRON)
1618 for idx, bdev_status in enumerate(result.payload):
1619 if bdev_status and bdev_status.ldisk_status == constants.LDS_FAULTY:
1625 def _CheckIAllocatorOrNode(lu, iallocator_slot, node_slot):
1626 """Check the sanity of iallocator and node arguments and use the
1627 cluster-wide iallocator if appropriate.
1629 Check that at most one of (iallocator, node) is specified. If none is
1630 specified, then the LU's opcode's iallocator slot is filled with the
1631 cluster-wide default iallocator.
1633 @type iallocator_slot: string
1634 @param iallocator_slot: the name of the opcode iallocator slot
1635 @type node_slot: string
1636 @param node_slot: the name of the opcode target node slot
1639 node = getattr(lu.op, node_slot, None)
1640 iallocator = getattr(lu.op, iallocator_slot, None)
1642 if node is not None and iallocator is not None:
1643 raise errors.OpPrereqError("Do not specify both, iallocator and node",
1645 elif node is None and iallocator is None:
1646 default_iallocator = lu.cfg.GetDefaultIAllocator()
1647 if default_iallocator:
1648 setattr(lu.op, iallocator_slot, default_iallocator)
1650 raise errors.OpPrereqError("No iallocator or node given and no"
1651 " cluster-wide default iallocator found;"
1652 " please specify either an iallocator or a"
1653 " node, or set a cluster-wide default"
1657 def _GetDefaultIAllocator(cfg, iallocator):
1658 """Decides on which iallocator to use.
1660 @type cfg: L{config.ConfigWriter}
1661 @param cfg: Cluster configuration object
1662 @type iallocator: string or None
1663 @param iallocator: Iallocator specified in opcode
1665 @return: Iallocator name
1669 # Use default iallocator
1670 iallocator = cfg.GetDefaultIAllocator()
1673 raise errors.OpPrereqError("No iallocator was specified, neither in the"
1674 " opcode nor as a cluster-wide default",
1680 class LUClusterPostInit(LogicalUnit):
1681 """Logical unit for running hooks after cluster initialization.
1684 HPATH = "cluster-init"
1685 HTYPE = constants.HTYPE_CLUSTER
1687 def BuildHooksEnv(self):
1692 "OP_TARGET": self.cfg.GetClusterName(),
1695 def BuildHooksNodes(self):
1696 """Build hooks nodes.
1699 return ([], [self.cfg.GetMasterNode()])
1701 def Exec(self, feedback_fn):
1708 class LUClusterDestroy(LogicalUnit):
1709 """Logical unit for destroying the cluster.
1712 HPATH = "cluster-destroy"
1713 HTYPE = constants.HTYPE_CLUSTER
1715 def BuildHooksEnv(self):
1720 "OP_TARGET": self.cfg.GetClusterName(),
1723 def BuildHooksNodes(self):
1724 """Build hooks nodes.
1729 def CheckPrereq(self):
1730 """Check prerequisites.
1732 This checks whether the cluster is empty.
1734 Any errors are signaled by raising errors.OpPrereqError.
1737 master = self.cfg.GetMasterNode()
1739 nodelist = self.cfg.GetNodeList()
1740 if len(nodelist) != 1 or nodelist[0] != master:
1741 raise errors.OpPrereqError("There are still %d node(s) in"
1742 " this cluster." % (len(nodelist) - 1),
1744 instancelist = self.cfg.GetInstanceList()
1746 raise errors.OpPrereqError("There are still %d instance(s) in"
1747 " this cluster." % len(instancelist),
1750 def Exec(self, feedback_fn):
1751 """Destroys the cluster.
1754 master_params = self.cfg.GetMasterNetworkParameters()
1756 # Run post hooks on master node before it's removed
1757 _RunPostHook(self, master_params.name)
1759 ems = self.cfg.GetUseExternalMipScript()
1760 result = self.rpc.call_node_deactivate_master_ip(master_params.name,
1763 self.LogWarning("Error disabling the master IP address: %s",
1766 return master_params.name
1769 def _VerifyCertificate(filename):
1770 """Verifies a certificate for L{LUClusterVerifyConfig}.
1772 @type filename: string
1773 @param filename: Path to PEM file
1777 cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
1778 utils.ReadFile(filename))
1779 except Exception, err: # pylint: disable=W0703
1780 return (LUClusterVerifyConfig.ETYPE_ERROR,
1781 "Failed to load X509 certificate %s: %s" % (filename, err))
1784 utils.VerifyX509Certificate(cert, constants.SSL_CERT_EXPIRATION_WARN,
1785 constants.SSL_CERT_EXPIRATION_ERROR)
1788 fnamemsg = "While verifying %s: %s" % (filename, msg)
1793 return (None, fnamemsg)
1794 elif errcode == utils.CERT_WARNING:
1795 return (LUClusterVerifyConfig.ETYPE_WARNING, fnamemsg)
1796 elif errcode == utils.CERT_ERROR:
1797 return (LUClusterVerifyConfig.ETYPE_ERROR, fnamemsg)
1799 raise errors.ProgrammerError("Unhandled certificate error code %r" % errcode)
1802 def _GetAllHypervisorParameters(cluster, instances):
1803 """Compute the set of all hypervisor parameters.
1805 @type cluster: L{objects.Cluster}
1806 @param cluster: the cluster object
1807 @param instances: list of L{objects.Instance}
1808 @param instances: additional instances from which to obtain parameters
1809 @rtype: list of (origin, hypervisor, parameters)
1810 @return: a list with all parameters found, indicating the hypervisor they
1811 apply to, and the origin (can be "cluster", "os X", or "instance Y")
1816 for hv_name in cluster.enabled_hypervisors:
1817 hvp_data.append(("cluster", hv_name, cluster.GetHVDefaults(hv_name)))
1819 for os_name, os_hvp in cluster.os_hvp.items():
1820 for hv_name, hv_params in os_hvp.items():
1822 full_params = cluster.GetHVDefaults(hv_name, os_name=os_name)
1823 hvp_data.append(("os %s" % os_name, hv_name, full_params))
1825 # TODO: collapse identical parameter values in a single one
1826 for instance in instances:
1827 if instance.hvparams:
1828 hvp_data.append(("instance %s" % instance.name, instance.hypervisor,
1829 cluster.FillHV(instance)))
1834 class _VerifyErrors(object):
1835 """Mix-in for cluster/group verify LUs.
1837 It provides _Error and _ErrorIf, and updates the self.bad boolean. (Expects
1838 self.op and self._feedback_fn to be available.)
1842 ETYPE_FIELD = "code"
1843 ETYPE_ERROR = "ERROR"
1844 ETYPE_WARNING = "WARNING"
1846 def _Error(self, ecode, item, msg, *args, **kwargs):
1847 """Format an error message.
1849 Based on the opcode's error_codes parameter, either format a
1850 parseable error code, or a simpler error string.
1852 This must be called only from Exec and functions called from Exec.
1855 ltype = kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR)
1856 itype, etxt, _ = ecode
1857 # first complete the msg
1860 # then format the whole message
1861 if self.op.error_codes: # This is a mix-in. pylint: disable=E1101
1862 msg = "%s:%s:%s:%s:%s" % (ltype, etxt, itype, item, msg)
1868 msg = "%s: %s%s: %s" % (ltype, itype, item, msg)
1869 # and finally report it via the feedback_fn
1870 self._feedback_fn(" - %s" % msg) # Mix-in. pylint: disable=E1101
1872 def _ErrorIf(self, cond, ecode, *args, **kwargs):
1873 """Log an error message if the passed condition is True.
1877 or self.op.debug_simulate_errors) # pylint: disable=E1101
1879 # If the error code is in the list of ignored errors, demote the error to a
1881 (_, etxt, _) = ecode
1882 if etxt in self.op.ignore_errors: # pylint: disable=E1101
1883 kwargs[self.ETYPE_FIELD] = self.ETYPE_WARNING
1886 self._Error(ecode, *args, **kwargs)
1888 # do not mark the operation as failed for WARN cases only
1889 if kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR) == self.ETYPE_ERROR:
1890 self.bad = self.bad or cond
1893 class LUClusterVerify(NoHooksLU):
1894 """Submits all jobs necessary to verify the cluster.
1899 def ExpandNames(self):
1900 self.needed_locks = {}
1902 def Exec(self, feedback_fn):
1905 if self.op.group_name:
1906 groups = [self.op.group_name]
1907 depends_fn = lambda: None
1909 groups = self.cfg.GetNodeGroupList()
1911 # Verify global configuration
1913 opcodes.OpClusterVerifyConfig(ignore_errors=self.op.ignore_errors)
1916 # Always depend on global verification
1917 depends_fn = lambda: [(-len(jobs), [])]
1919 jobs.extend([opcodes.OpClusterVerifyGroup(group_name=group,
1920 ignore_errors=self.op.ignore_errors,
1921 depends=depends_fn())]
1922 for group in groups)
1924 # Fix up all parameters
1925 for op in itertools.chain(*jobs): # pylint: disable=W0142
1926 op.debug_simulate_errors = self.op.debug_simulate_errors
1927 op.verbose = self.op.verbose
1928 op.error_codes = self.op.error_codes
1930 op.skip_checks = self.op.skip_checks
1931 except AttributeError:
1932 assert not isinstance(op, opcodes.OpClusterVerifyGroup)
1934 return ResultWithJobs(jobs)
1937 class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
1938 """Verifies the cluster config.
1943 def _VerifyHVP(self, hvp_data):
1944 """Verifies locally the syntax of the hypervisor parameters.
1947 for item, hv_name, hv_params in hvp_data:
1948 msg = ("hypervisor %s parameters syntax check (source %s): %%s" %
1951 hv_class = hypervisor.GetHypervisor(hv_name)
1952 utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
1953 hv_class.CheckParameterSyntax(hv_params)
1954 except errors.GenericError, err:
1955 self._ErrorIf(True, constants.CV_ECLUSTERCFG, None, msg % str(err))
1957 def ExpandNames(self):
1958 self.needed_locks = dict.fromkeys(locking.LEVELS, locking.ALL_SET)
1959 self.share_locks = _ShareAll()
1961 def CheckPrereq(self):
1962 """Check prerequisites.
1965 # Retrieve all information
1966 self.all_group_info = self.cfg.GetAllNodeGroupsInfo()
1967 self.all_node_info = self.cfg.GetAllNodesInfo()
1968 self.all_inst_info = self.cfg.GetAllInstancesInfo()
1970 def Exec(self, feedback_fn):
1971 """Verify integrity of cluster, performing various test on nodes.
1975 self._feedback_fn = feedback_fn
1977 feedback_fn("* Verifying cluster config")
1979 for msg in self.cfg.VerifyConfig():
1980 self._ErrorIf(True, constants.CV_ECLUSTERCFG, None, msg)
1982 feedback_fn("* Verifying cluster certificate files")
1984 for cert_filename in constants.ALL_CERT_FILES:
1985 (errcode, msg) = _VerifyCertificate(cert_filename)
1986 self._ErrorIf(errcode, constants.CV_ECLUSTERCERT, None, msg, code=errcode)
1988 feedback_fn("* Verifying hypervisor parameters")
1990 self._VerifyHVP(_GetAllHypervisorParameters(self.cfg.GetClusterInfo(),
1991 self.all_inst_info.values()))
1993 feedback_fn("* Verifying all nodes belong to an existing group")
1995 # We do this verification here because, should this bogus circumstance
1996 # occur, it would never be caught by VerifyGroup, which only acts on
1997 # nodes/instances reachable from existing node groups.
1999 dangling_nodes = set(node.name for node in self.all_node_info.values()
2000 if node.group not in self.all_group_info)
2002 dangling_instances = {}
2003 no_node_instances = []
2005 for inst in self.all_inst_info.values():
2006 if inst.primary_node in dangling_nodes:
2007 dangling_instances.setdefault(inst.primary_node, []).append(inst.name)
2008 elif inst.primary_node not in self.all_node_info:
2009 no_node_instances.append(inst.name)
2014 utils.CommaJoin(dangling_instances.get(node.name,
2016 for node in dangling_nodes]
2018 self._ErrorIf(bool(dangling_nodes), constants.CV_ECLUSTERDANGLINGNODES,
2020 "the following nodes (and their instances) belong to a non"
2021 " existing group: %s", utils.CommaJoin(pretty_dangling))
2023 self._ErrorIf(bool(no_node_instances), constants.CV_ECLUSTERDANGLINGINST,
2025 "the following instances have a non-existing primary-node:"
2026 " %s", utils.CommaJoin(no_node_instances))
2031 class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
2032 """Verifies the status of a node group.
2035 HPATH = "cluster-verify"
2036 HTYPE = constants.HTYPE_CLUSTER
2039 _HOOKS_INDENT_RE = re.compile("^", re.M)
2041 class NodeImage(object):
2042 """A class representing the logical and physical status of a node.
2045 @ivar name: the node name to which this object refers
2046 @ivar volumes: a structure as returned from
2047 L{ganeti.backend.GetVolumeList} (runtime)
2048 @ivar instances: a list of running instances (runtime)
2049 @ivar pinst: list of configured primary instances (config)
2050 @ivar sinst: list of configured secondary instances (config)
2051 @ivar sbp: dictionary of {primary-node: list of instances} for all
2052 instances for which this node is secondary (config)
2053 @ivar mfree: free memory, as reported by hypervisor (runtime)
2054 @ivar dfree: free disk, as reported by the node (runtime)
2055 @ivar offline: the offline status (config)
2056 @type rpc_fail: boolean
2057 @ivar rpc_fail: whether the RPC verify call was successfull (overall,
2058 not whether the individual keys were correct) (runtime)
2059 @type lvm_fail: boolean
2060 @ivar lvm_fail: whether the RPC call didn't return valid LVM data
2061 @type hyp_fail: boolean
2062 @ivar hyp_fail: whether the RPC call didn't return the instance list
2063 @type ghost: boolean
2064 @ivar ghost: whether this is a known node or not (config)
2065 @type os_fail: boolean
2066 @ivar os_fail: whether the RPC call didn't return valid OS data
2068 @ivar oslist: list of OSes as diagnosed by DiagnoseOS
2069 @type vm_capable: boolean
2070 @ivar vm_capable: whether the node can host instances
2073 def __init__(self, offline=False, name=None, vm_capable=True):
2082 self.offline = offline
2083 self.vm_capable = vm_capable
2084 self.rpc_fail = False
2085 self.lvm_fail = False
2086 self.hyp_fail = False
2088 self.os_fail = False
2091 def ExpandNames(self):
2092 # This raises errors.OpPrereqError on its own:
2093 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
2095 # Get instances in node group; this is unsafe and needs verification later
2097 self.cfg.GetNodeGroupInstances(self.group_uuid, primary_only=True)
2099 self.needed_locks = {
2100 locking.LEVEL_INSTANCE: inst_names,
2101 locking.LEVEL_NODEGROUP: [self.group_uuid],
2102 locking.LEVEL_NODE: [],
2105 self.share_locks = _ShareAll()
2107 def DeclareLocks(self, level):
2108 if level == locking.LEVEL_NODE:
2109 # Get members of node group; this is unsafe and needs verification later
2110 nodes = set(self.cfg.GetNodeGroup(self.group_uuid).members)
2112 all_inst_info = self.cfg.GetAllInstancesInfo()
2114 # In Exec(), we warn about mirrored instances that have primary and
2115 # secondary living in separate node groups. To fully verify that
2116 # volumes for these instances are healthy, we will need to do an
2117 # extra call to their secondaries. We ensure here those nodes will
2119 for inst in self.owned_locks(locking.LEVEL_INSTANCE):
2120 # Important: access only the instances whose lock is owned
2121 if all_inst_info[inst].disk_template in constants.DTS_INT_MIRROR:
2122 nodes.update(all_inst_info[inst].secondary_nodes)
2124 self.needed_locks[locking.LEVEL_NODE] = nodes
2126 def CheckPrereq(self):
2127 assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
2128 self.group_info = self.cfg.GetNodeGroup(self.group_uuid)
2130 group_nodes = set(self.group_info.members)
2132 self.cfg.GetNodeGroupInstances(self.group_uuid, primary_only=True)
2135 group_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
2137 unlocked_instances = \
2138 group_instances.difference(self.owned_locks(locking.LEVEL_INSTANCE))
2141 raise errors.OpPrereqError("Missing lock for nodes: %s" %
2142 utils.CommaJoin(unlocked_nodes),
2145 if unlocked_instances:
2146 raise errors.OpPrereqError("Missing lock for instances: %s" %
2147 utils.CommaJoin(unlocked_instances),
2150 self.all_node_info = self.cfg.GetAllNodesInfo()
2151 self.all_inst_info = self.cfg.GetAllInstancesInfo()
2153 self.my_node_names = utils.NiceSort(group_nodes)
2154 self.my_inst_names = utils.NiceSort(group_instances)
2156 self.my_node_info = dict((name, self.all_node_info[name])
2157 for name in self.my_node_names)
2159 self.my_inst_info = dict((name, self.all_inst_info[name])
2160 for name in self.my_inst_names)
2162 # We detect here the nodes that will need the extra RPC calls for verifying
2163 # split LV volumes; they should be locked.
2164 extra_lv_nodes = set()
2166 for inst in self.my_inst_info.values():
2167 if inst.disk_template in constants.DTS_INT_MIRROR:
2168 for nname in inst.all_nodes:
2169 if self.all_node_info[nname].group != self.group_uuid:
2170 extra_lv_nodes.add(nname)
2172 unlocked_lv_nodes = \
2173 extra_lv_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
2175 if unlocked_lv_nodes:
2176 raise errors.OpPrereqError("Missing node locks for LV check: %s" %
2177 utils.CommaJoin(unlocked_lv_nodes),
2179 self.extra_lv_nodes = list(extra_lv_nodes)
2181 def _VerifyNode(self, ninfo, nresult):
2182 """Perform some basic validation on data returned from a node.
2184 - check the result data structure is well formed and has all the
2186 - check ganeti version
2188 @type ninfo: L{objects.Node}
2189 @param ninfo: the node to check
2190 @param nresult: the results from the node
2192 @return: whether overall this call was successful (and we can expect
2193 reasonable values in the respose)
2197 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2199 # main result, nresult should be a non-empty dict
2200 test = not nresult or not isinstance(nresult, dict)
2201 _ErrorIf(test, constants.CV_ENODERPC, node,
2202 "unable to verify node: no data returned")
2206 # compares ganeti version
2207 local_version = constants.PROTOCOL_VERSION
2208 remote_version = nresult.get("version", None)
2209 test = not (remote_version and
2210 isinstance(remote_version, (list, tuple)) and
2211 len(remote_version) == 2)
2212 _ErrorIf(test, constants.CV_ENODERPC, node,
2213 "connection to node returned invalid data")
2217 test = local_version != remote_version[0]
2218 _ErrorIf(test, constants.CV_ENODEVERSION, node,
2219 "incompatible protocol versions: master %s,"
2220 " node %s", local_version, remote_version[0])
2224 # node seems compatible, we can actually try to look into its results
2226 # full package version
2227 self._ErrorIf(constants.RELEASE_VERSION != remote_version[1],
2228 constants.CV_ENODEVERSION, node,
2229 "software version mismatch: master %s, node %s",
2230 constants.RELEASE_VERSION, remote_version[1],
2231 code=self.ETYPE_WARNING)
2233 hyp_result = nresult.get(constants.NV_HYPERVISOR, None)
2234 if ninfo.vm_capable and isinstance(hyp_result, dict):
2235 for hv_name, hv_result in hyp_result.iteritems():
2236 test = hv_result is not None
2237 _ErrorIf(test, constants.CV_ENODEHV, node,
2238 "hypervisor %s verify failure: '%s'", hv_name, hv_result)
2240 hvp_result = nresult.get(constants.NV_HVPARAMS, None)
2241 if ninfo.vm_capable and isinstance(hvp_result, list):
2242 for item, hv_name, hv_result in hvp_result:
2243 _ErrorIf(True, constants.CV_ENODEHV, node,
2244 "hypervisor %s parameter verify failure (source %s): %s",
2245 hv_name, item, hv_result)
2247 test = nresult.get(constants.NV_NODESETUP,
2248 ["Missing NODESETUP results"])
2249 _ErrorIf(test, constants.CV_ENODESETUP, node, "node setup error: %s",
2254 def _VerifyNodeTime(self, ninfo, nresult,
2255 nvinfo_starttime, nvinfo_endtime):
2256 """Check the node time.
2258 @type ninfo: L{objects.Node}
2259 @param ninfo: the node to check
2260 @param nresult: the remote results for the node
2261 @param nvinfo_starttime: the start time of the RPC call
2262 @param nvinfo_endtime: the end time of the RPC call
2266 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2268 ntime = nresult.get(constants.NV_TIME, None)
2270 ntime_merged = utils.MergeTime(ntime)
2271 except (ValueError, TypeError):
2272 _ErrorIf(True, constants.CV_ENODETIME, node, "Node returned invalid time")
2275 if ntime_merged < (nvinfo_starttime - constants.NODE_MAX_CLOCK_SKEW):
2276 ntime_diff = "%.01fs" % abs(nvinfo_starttime - ntime_merged)
2277 elif ntime_merged > (nvinfo_endtime + constants.NODE_MAX_CLOCK_SKEW):
2278 ntime_diff = "%.01fs" % abs(ntime_merged - nvinfo_endtime)
2282 _ErrorIf(ntime_diff is not None, constants.CV_ENODETIME, node,
2283 "Node time diverges by at least %s from master node time",
2286 def _VerifyNodeLVM(self, ninfo, nresult, vg_name):
2287 """Check the node LVM results.
2289 @type ninfo: L{objects.Node}
2290 @param ninfo: the node to check
2291 @param nresult: the remote results for the node
2292 @param vg_name: the configured VG name
2299 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2301 # checks vg existence and size > 20G
2302 vglist = nresult.get(constants.NV_VGLIST, None)
2304 _ErrorIf(test, constants.CV_ENODELVM, node, "unable to check volume groups")
2306 vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
2307 constants.MIN_VG_SIZE)
2308 _ErrorIf(vgstatus, constants.CV_ENODELVM, node, vgstatus)
2311 pvlist = nresult.get(constants.NV_PVLIST, None)
2312 test = pvlist is None
2313 _ErrorIf(test, constants.CV_ENODELVM, node, "Can't get PV list from node")
2315 # check that ':' is not present in PV names, since it's a
2316 # special character for lvcreate (denotes the range of PEs to
2318 for _, pvname, owner_vg in pvlist:
2319 test = ":" in pvname
2320 _ErrorIf(test, constants.CV_ENODELVM, node,
2321 "Invalid character ':' in PV '%s' of VG '%s'",
2324 def _VerifyNodeBridges(self, ninfo, nresult, bridges):
2325 """Check the node bridges.
2327 @type ninfo: L{objects.Node}
2328 @param ninfo: the node to check
2329 @param nresult: the remote results for the node
2330 @param bridges: the expected list of bridges
2337 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2339 missing = nresult.get(constants.NV_BRIDGES, None)
2340 test = not isinstance(missing, list)
2341 _ErrorIf(test, constants.CV_ENODENET, node,
2342 "did not return valid bridge information")
2344 _ErrorIf(bool(missing), constants.CV_ENODENET, node,
2345 "missing bridges: %s" % utils.CommaJoin(sorted(missing)))
2347 def _VerifyNodeUserScripts(self, ninfo, nresult):
2348 """Check the results of user scripts presence and executability on the node
2350 @type ninfo: L{objects.Node}
2351 @param ninfo: the node to check
2352 @param nresult: the remote results for the node
2357 test = not constants.NV_USERSCRIPTS in nresult
2358 self._ErrorIf(test, constants.CV_ENODEUSERSCRIPTS, node,
2359 "did not return user scripts information")
2361 broken_scripts = nresult.get(constants.NV_USERSCRIPTS, None)
2363 self._ErrorIf(broken_scripts, constants.CV_ENODEUSERSCRIPTS, node,
2364 "user scripts not present or not executable: %s" %
2365 utils.CommaJoin(sorted(broken_scripts)))
2367 def _VerifyNodeNetwork(self, ninfo, nresult):
2368 """Check the node network connectivity results.
2370 @type ninfo: L{objects.Node}
2371 @param ninfo: the node to check
2372 @param nresult: the remote results for the node
2376 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2378 test = constants.NV_NODELIST not in nresult
2379 _ErrorIf(test, constants.CV_ENODESSH, node,
2380 "node hasn't returned node ssh connectivity data")
2382 if nresult[constants.NV_NODELIST]:
2383 for a_node, a_msg in nresult[constants.NV_NODELIST].items():
2384 _ErrorIf(True, constants.CV_ENODESSH, node,
2385 "ssh communication with node '%s': %s", a_node, a_msg)
2387 test = constants.NV_NODENETTEST not in nresult
2388 _ErrorIf(test, constants.CV_ENODENET, node,
2389 "node hasn't returned node tcp connectivity data")
2391 if nresult[constants.NV_NODENETTEST]:
2392 nlist = utils.NiceSort(nresult[constants.NV_NODENETTEST].keys())
2394 _ErrorIf(True, constants.CV_ENODENET, node,
2395 "tcp communication with node '%s': %s",
2396 anode, nresult[constants.NV_NODENETTEST][anode])
2398 test = constants.NV_MASTERIP not in nresult
2399 _ErrorIf(test, constants.CV_ENODENET, node,
2400 "node hasn't returned node master IP reachability data")
2402 if not nresult[constants.NV_MASTERIP]:
2403 if node == self.master_node:
2404 msg = "the master node cannot reach the master IP (not configured?)"
2406 msg = "cannot reach the master IP"
2407 _ErrorIf(True, constants.CV_ENODENET, node, msg)
2409 def _VerifyInstance(self, instance, instanceconfig, node_image,
2411 """Verify an instance.
2413 This function checks to see if the required block devices are
2414 available on the instance's node.
2417 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2418 node_current = instanceconfig.primary_node
2420 node_vol_should = {}
2421 instanceconfig.MapLVsByNode(node_vol_should)
2423 ipolicy = _CalculateGroupIPolicy(self.cfg.GetClusterInfo(), self.group_info)
2424 err = _ComputeIPolicyInstanceViolation(ipolicy, instanceconfig)
2425 _ErrorIf(err, constants.CV_EINSTANCEPOLICY, instance, utils.CommaJoin(err))
2427 for node in node_vol_should:
2428 n_img = node_image[node]
2429 if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
2430 # ignore missing volumes on offline or broken nodes
2432 for volume in node_vol_should[node]:
2433 test = volume not in n_img.volumes
2434 _ErrorIf(test, constants.CV_EINSTANCEMISSINGDISK, instance,
2435 "volume %s missing on node %s", volume, node)
2437 if instanceconfig.admin_state == constants.ADMINST_UP:
2438 pri_img = node_image[node_current]
2439 test = instance not in pri_img.instances and not pri_img.offline
2440 _ErrorIf(test, constants.CV_EINSTANCEDOWN, instance,
2441 "instance not running on its primary node %s",
2444 diskdata = [(nname, success, status, idx)
2445 for (nname, disks) in diskstatus.items()
2446 for idx, (success, status) in enumerate(disks)]
2448 for nname, success, bdev_status, idx in diskdata:
2449 # the 'ghost node' construction in Exec() ensures that we have a
2451 snode = node_image[nname]
2452 bad_snode = snode.ghost or snode.offline
2453 _ErrorIf(instanceconfig.admin_state == constants.ADMINST_UP and
2454 not success and not bad_snode,
2455 constants.CV_EINSTANCEFAULTYDISK, instance,
2456 "couldn't retrieve status for disk/%s on %s: %s",
2457 idx, nname, bdev_status)
2458 _ErrorIf((instanceconfig.admin_state == constants.ADMINST_UP and
2459 success and bdev_status.ldisk_status == constants.LDS_FAULTY),
2460 constants.CV_EINSTANCEFAULTYDISK, instance,
2461 "disk/%s on %s is faulty", idx, nname)
2463 def _VerifyOrphanVolumes(self, node_vol_should, node_image, reserved):
2464 """Verify if there are any unknown volumes in the cluster.
2466 The .os, .swap and backup volumes are ignored. All other volumes are
2467 reported as unknown.
2469 @type reserved: L{ganeti.utils.FieldSet}
2470 @param reserved: a FieldSet of reserved volume names
2473 for node, n_img in node_image.items():
2474 if (n_img.offline or n_img.rpc_fail or n_img.lvm_fail or
2475 self.all_node_info[node].group != self.group_uuid):
2476 # skip non-healthy nodes
2478 for volume in n_img.volumes:
2479 test = ((node not in node_vol_should or
2480 volume not in node_vol_should[node]) and
2481 not reserved.Matches(volume))
2482 self._ErrorIf(test, constants.CV_ENODEORPHANLV, node,
2483 "volume %s is unknown", volume)
2485 def _VerifyNPlusOneMemory(self, node_image, instance_cfg):
2486 """Verify N+1 Memory Resilience.
2488 Check that if one single node dies we can still start all the
2489 instances it was primary for.
2492 cluster_info = self.cfg.GetClusterInfo()
2493 for node, n_img in node_image.items():
2494 # This code checks that every node which is now listed as
2495 # secondary has enough memory to host all instances it is
2496 # supposed to should a single other node in the cluster fail.
2497 # FIXME: not ready for failover to an arbitrary node
2498 # FIXME: does not support file-backed instances
2499 # WARNING: we currently take into account down instances as well
2500 # as up ones, considering that even if they're down someone
2501 # might want to start them even in the event of a node failure.
2502 if n_img.offline or self.all_node_info[node].group != self.group_uuid:
2503 # we're skipping nodes marked offline and nodes in other groups from
2504 # the N+1 warning, since most likely we don't have good memory
2505 # infromation from them; we already list instances living on such
2506 # nodes, and that's enough warning
2508 #TODO(dynmem): also consider ballooning out other instances
2509 for prinode, instances in n_img.sbp.items():
2511 for instance in instances:
2512 bep = cluster_info.FillBE(instance_cfg[instance])
2513 if bep[constants.BE_AUTO_BALANCE]:
2514 needed_mem += bep[constants.BE_MINMEM]
2515 test = n_img.mfree < needed_mem
2516 self._ErrorIf(test, constants.CV_ENODEN1, node,
2517 "not enough memory to accomodate instance failovers"
2518 " should node %s fail (%dMiB needed, %dMiB available)",
2519 prinode, needed_mem, n_img.mfree)
2522 def _VerifyFiles(cls, errorif, nodeinfo, master_node, all_nvinfo,
2523 (files_all, files_opt, files_mc, files_vm)):
2524 """Verifies file checksums collected from all nodes.
2526 @param errorif: Callback for reporting errors
2527 @param nodeinfo: List of L{objects.Node} objects
2528 @param master_node: Name of master node
2529 @param all_nvinfo: RPC results
2532 # Define functions determining which nodes to consider for a file
2535 (files_mc, lambda node: (node.master_candidate or
2536 node.name == master_node)),
2537 (files_vm, lambda node: node.vm_capable),
2540 # Build mapping from filename to list of nodes which should have the file
2542 for (files, fn) in files2nodefn:
2544 filenodes = nodeinfo
2546 filenodes = filter(fn, nodeinfo)
2547 nodefiles.update((filename,
2548 frozenset(map(operator.attrgetter("name"), filenodes)))
2549 for filename in files)
2551 assert set(nodefiles) == (files_all | files_mc | files_vm)
2553 fileinfo = dict((filename, {}) for filename in nodefiles)
2554 ignore_nodes = set()
2556 for node in nodeinfo:
2558 ignore_nodes.add(node.name)
2561 nresult = all_nvinfo[node.name]
2563 if nresult.fail_msg or not nresult.payload:
2566 node_files = nresult.payload.get(constants.NV_FILELIST, None)
2568 test = not (node_files and isinstance(node_files, dict))
2569 errorif(test, constants.CV_ENODEFILECHECK, node.name,
2570 "Node did not return file checksum data")
2572 ignore_nodes.add(node.name)
2575 # Build per-checksum mapping from filename to nodes having it
2576 for (filename, checksum) in node_files.items():
2577 assert filename in nodefiles
2578 fileinfo[filename].setdefault(checksum, set()).add(node.name)
2580 for (filename, checksums) in fileinfo.items():
2581 assert compat.all(len(i) > 10 for i in checksums), "Invalid checksum"
2583 # Nodes having the file
2584 with_file = frozenset(node_name
2585 for nodes in fileinfo[filename].values()
2586 for node_name in nodes) - ignore_nodes
2588 expected_nodes = nodefiles[filename] - ignore_nodes
2590 # Nodes missing file
2591 missing_file = expected_nodes - with_file
2593 if filename in files_opt:
2595 errorif(missing_file and missing_file != expected_nodes,
2596 constants.CV_ECLUSTERFILECHECK, None,
2597 "File %s is optional, but it must exist on all or no"
2598 " nodes (not found on %s)",
2599 filename, utils.CommaJoin(utils.NiceSort(missing_file)))
2601 errorif(missing_file, constants.CV_ECLUSTERFILECHECK, None,
2602 "File %s is missing from node(s) %s", filename,
2603 utils.CommaJoin(utils.NiceSort(missing_file)))
2605 # Warn if a node has a file it shouldn't
2606 unexpected = with_file - expected_nodes
2608 constants.CV_ECLUSTERFILECHECK, None,
2609 "File %s should not exist on node(s) %s",
2610 filename, utils.CommaJoin(utils.NiceSort(unexpected)))
2612 # See if there are multiple versions of the file
2613 test = len(checksums) > 1
2615 variants = ["variant %s on %s" %
2616 (idx + 1, utils.CommaJoin(utils.NiceSort(nodes)))
2617 for (idx, (checksum, nodes)) in
2618 enumerate(sorted(checksums.items()))]
2622 errorif(test, constants.CV_ECLUSTERFILECHECK, None,
2623 "File %s found with %s different checksums (%s)",
2624 filename, len(checksums), "; ".join(variants))
2626 def _VerifyNodeDrbd(self, ninfo, nresult, instanceinfo, drbd_helper,
2628 """Verifies and the node DRBD status.
2630 @type ninfo: L{objects.Node}
2631 @param ninfo: the node to check
2632 @param nresult: the remote results for the node
2633 @param instanceinfo: the dict of instances
2634 @param drbd_helper: the configured DRBD usermode helper
2635 @param drbd_map: the DRBD map as returned by
2636 L{ganeti.config.ConfigWriter.ComputeDRBDMap}
2640 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2643 helper_result = nresult.get(constants.NV_DRBDHELPER, None)
2644 test = (helper_result == None)
2645 _ErrorIf(test, constants.CV_ENODEDRBDHELPER, node,
2646 "no drbd usermode helper returned")
2648 status, payload = helper_result
2650 _ErrorIf(test, constants.CV_ENODEDRBDHELPER, node,
2651 "drbd usermode helper check unsuccessful: %s", payload)
2652 test = status and (payload != drbd_helper)
2653 _ErrorIf(test, constants.CV_ENODEDRBDHELPER, node,
2654 "wrong drbd usermode helper: %s", payload)
2656 # compute the DRBD minors
2658 for minor, instance in drbd_map[node].items():
2659 test = instance not in instanceinfo
2660 _ErrorIf(test, constants.CV_ECLUSTERCFG, None,
2661 "ghost instance '%s' in temporary DRBD map", instance)
2662 # ghost instance should not be running, but otherwise we
2663 # don't give double warnings (both ghost instance and
2664 # unallocated minor in use)
2666 node_drbd[minor] = (instance, False)
2668 instance = instanceinfo[instance]
2669 node_drbd[minor] = (instance.name,
2670 instance.admin_state == constants.ADMINST_UP)
2672 # and now check them
2673 used_minors = nresult.get(constants.NV_DRBDLIST, [])
2674 test = not isinstance(used_minors, (tuple, list))
2675 _ErrorIf(test, constants.CV_ENODEDRBD, node,
2676 "cannot parse drbd status file: %s", str(used_minors))
2678 # we cannot check drbd status
2681 for minor, (iname, must_exist) in node_drbd.items():
2682 test = minor not in used_minors and must_exist
2683 _ErrorIf(test, constants.CV_ENODEDRBD, node,
2684 "drbd minor %d of instance %s is not active", minor, iname)
2685 for minor in used_minors:
2686 test = minor not in node_drbd
2687 _ErrorIf(test, constants.CV_ENODEDRBD, node,
2688 "unallocated drbd minor %d is in use", minor)
2690 def _UpdateNodeOS(self, ninfo, nresult, nimg):
2691 """Builds the node OS structures.
2693 @type ninfo: L{objects.Node}
2694 @param ninfo: the node to check
2695 @param nresult: the remote results for the node
2696 @param nimg: the node image object
2700 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2702 remote_os = nresult.get(constants.NV_OSLIST, None)
2703 test = (not isinstance(remote_os, list) or
2704 not compat.all(isinstance(v, list) and len(v) == 7
2705 for v in remote_os))
2707 _ErrorIf(test, constants.CV_ENODEOS, node,
2708 "node hasn't returned valid OS data")
2717 for (name, os_path, status, diagnose,
2718 variants, parameters, api_ver) in nresult[constants.NV_OSLIST]:
2720 if name not in os_dict:
2723 # parameters is a list of lists instead of list of tuples due to
2724 # JSON lacking a real tuple type, fix it:
2725 parameters = [tuple(v) for v in parameters]
2726 os_dict[name].append((os_path, status, diagnose,
2727 set(variants), set(parameters), set(api_ver)))
2729 nimg.oslist = os_dict
2731 def _VerifyNodeOS(self, ninfo, nimg, base):
2732 """Verifies the node OS list.
2734 @type ninfo: L{objects.Node}
2735 @param ninfo: the node to check
2736 @param nimg: the node image object
2737 @param base: the 'template' node we match against (e.g. from the master)
2741 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2743 assert not nimg.os_fail, "Entered _VerifyNodeOS with failed OS rpc?"
2745 beautify_params = lambda l: ["%s: %s" % (k, v) for (k, v) in l]
2746 for os_name, os_data in nimg.oslist.items():
2747 assert os_data, "Empty OS status for OS %s?!" % os_name
2748 f_path, f_status, f_diag, f_var, f_param, f_api = os_data[0]
2749 _ErrorIf(not f_status, constants.CV_ENODEOS, node,
2750 "Invalid OS %s (located at %s): %s", os_name, f_path, f_diag)
2751 _ErrorIf(len(os_data) > 1, constants.CV_ENODEOS, node,
2752 "OS '%s' has multiple entries (first one shadows the rest): %s",
2753 os_name, utils.CommaJoin([v[0] for v in os_data]))
2754 # comparisons with the 'base' image
2755 test = os_name not in base.oslist
2756 _ErrorIf(test, constants.CV_ENODEOS, node,
2757 "Extra OS %s not present on reference node (%s)",
2761 assert base.oslist[os_name], "Base node has empty OS status?"
2762 _, b_status, _, b_var, b_param, b_api = base.oslist[os_name][0]
2764 # base OS is invalid, skipping
2766 for kind, a, b in [("API version", f_api, b_api),
2767 ("variants list", f_var, b_var),
2768 ("parameters", beautify_params(f_param),
2769 beautify_params(b_param))]:
2770 _ErrorIf(a != b, constants.CV_ENODEOS, node,
2771 "OS %s for %s differs from reference node %s: [%s] vs. [%s]",
2772 kind, os_name, base.name,
2773 utils.CommaJoin(sorted(a)), utils.CommaJoin(sorted(b)))
2775 # check any missing OSes
2776 missing = set(base.oslist.keys()).difference(nimg.oslist.keys())
2777 _ErrorIf(missing, constants.CV_ENODEOS, node,
2778 "OSes present on reference node %s but missing on this node: %s",
2779 base.name, utils.CommaJoin(missing))
2781 def _VerifyOob(self, ninfo, nresult):
2782 """Verifies out of band functionality of a node.
2784 @type ninfo: L{objects.Node}
2785 @param ninfo: the node to check
2786 @param nresult: the remote results for the node
2790 # We just have to verify the paths on master and/or master candidates
2791 # as the oob helper is invoked on the master
2792 if ((ninfo.master_candidate or ninfo.master_capable) and
2793 constants.NV_OOB_PATHS in nresult):
2794 for path_result in nresult[constants.NV_OOB_PATHS]:
2795 self._ErrorIf(path_result, constants.CV_ENODEOOBPATH, node, path_result)
2797 def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
2798 """Verifies and updates the node volume data.
2800 This function will update a L{NodeImage}'s internal structures
2801 with data from the remote call.
2803 @type ninfo: L{objects.Node}
2804 @param ninfo: the node to check
2805 @param nresult: the remote results for the node
2806 @param nimg: the node image object
2807 @param vg_name: the configured VG name
2811 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2813 nimg.lvm_fail = True
2814 lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
2817 elif isinstance(lvdata, basestring):
2818 _ErrorIf(True, constants.CV_ENODELVM, node, "LVM problem on node: %s",
2819 utils.SafeEncode(lvdata))
2820 elif not isinstance(lvdata, dict):
2821 _ErrorIf(True, constants.CV_ENODELVM, node,
2822 "rpc call to node failed (lvlist)")
2824 nimg.volumes = lvdata
2825 nimg.lvm_fail = False
2827 def _UpdateNodeInstances(self, ninfo, nresult, nimg):
2828 """Verifies and updates the node instance list.
2830 If the listing was successful, then updates this node's instance
2831 list. Otherwise, it marks the RPC call as failed for the instance
2834 @type ninfo: L{objects.Node}
2835 @param ninfo: the node to check
2836 @param nresult: the remote results for the node
2837 @param nimg: the node image object
2840 idata = nresult.get(constants.NV_INSTANCELIST, None)
2841 test = not isinstance(idata, list)
2842 self._ErrorIf(test, constants.CV_ENODEHV, ninfo.name,
2843 "rpc call to node failed (instancelist): %s",
2844 utils.SafeEncode(str(idata)))
2846 nimg.hyp_fail = True
2848 nimg.instances = idata
2850 def _UpdateNodeInfo(self, ninfo, nresult, nimg, vg_name):
2851 """Verifies and computes a node information map
2853 @type ninfo: L{objects.Node}
2854 @param ninfo: the node to check
2855 @param nresult: the remote results for the node
2856 @param nimg: the node image object
2857 @param vg_name: the configured VG name
2861 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2863 # try to read free memory (from the hypervisor)
2864 hv_info = nresult.get(constants.NV_HVINFO, None)
2865 test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
2866 _ErrorIf(test, constants.CV_ENODEHV, node,
2867 "rpc call to node failed (hvinfo)")
2870 nimg.mfree = int(hv_info["memory_free"])
2871 except (ValueError, TypeError):
2872 _ErrorIf(True, constants.CV_ENODERPC, node,
2873 "node returned invalid nodeinfo, check hypervisor")
2875 # FIXME: devise a free space model for file based instances as well
2876 if vg_name is not None:
2877 test = (constants.NV_VGLIST not in nresult or
2878 vg_name not in nresult[constants.NV_VGLIST])
2879 _ErrorIf(test, constants.CV_ENODELVM, node,
2880 "node didn't return data for the volume group '%s'"
2881 " - it is either missing or broken", vg_name)
2884 nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
2885 except (ValueError, TypeError):
2886 _ErrorIf(True, constants.CV_ENODERPC, node,
2887 "node returned invalid LVM info, check LVM status")
2889 def _CollectDiskInfo(self, nodelist, node_image, instanceinfo):
2890 """Gets per-disk status information for all instances.
2892 @type nodelist: list of strings
2893 @param nodelist: Node names
2894 @type node_image: dict of (name, L{objects.Node})
2895 @param node_image: Node objects
2896 @type instanceinfo: dict of (name, L{objects.Instance})
2897 @param instanceinfo: Instance objects
2898 @rtype: {instance: {node: [(succes, payload)]}}
2899 @return: a dictionary of per-instance dictionaries with nodes as
2900 keys and disk information as values; the disk information is a
2901 list of tuples (success, payload)
2904 _ErrorIf = self._ErrorIf # pylint: disable=C0103
2907 node_disks_devonly = {}
2908 diskless_instances = set()
2909 diskless = constants.DT_DISKLESS
2911 for nname in nodelist:
2912 node_instances = list(itertools.chain(node_image[nname].pinst,
2913 node_image[nname].sinst))
2914 diskless_instances.update(inst for inst in node_instances
2915 if instanceinfo[inst].disk_template == diskless)
2916 disks = [(inst, disk)
2917 for inst in node_instances
2918 for disk in instanceinfo[inst].disks]
2921 # No need to collect data
2924 node_disks[nname] = disks
2926 # Creating copies as SetDiskID below will modify the objects and that can
2927 # lead to incorrect data returned from nodes
2928 devonly = [dev.Copy() for (_, dev) in disks]
2931 self.cfg.SetDiskID(dev, nname)
2933 node_disks_devonly[nname] = devonly
2935 assert len(node_disks) == len(node_disks_devonly)
2937 # Collect data from all nodes with disks
2938 result = self.rpc.call_blockdev_getmirrorstatus_multi(node_disks.keys(),
2941 assert len(result) == len(node_disks)
2945 for (nname, nres) in result.items():
2946 disks = node_disks[nname]
2949 # No data from this node
2950 data = len(disks) * [(False, "node offline")]
2953 _ErrorIf(msg, constants.CV_ENODERPC, nname,
2954 "while getting disk information: %s", msg)
2956 # No data from this node
2957 data = len(disks) * [(False, msg)]
2960 for idx, i in enumerate(nres.payload):
2961 if isinstance(i, (tuple, list)) and len(i) == 2:
2964 logging.warning("Invalid result from node %s, entry %d: %s",
2966 data.append((False, "Invalid result from the remote node"))
2968 for ((inst, _), status) in zip(disks, data):
2969 instdisk.setdefault(inst, {}).setdefault(nname, []).append(status)
2971 # Add empty entries for diskless instances.
2972 for inst in diskless_instances:
2973 assert inst not in instdisk
2976 assert compat.all(len(statuses) == len(instanceinfo[inst].disks) and
2977 len(nnames) <= len(instanceinfo[inst].all_nodes) and
2978 compat.all(isinstance(s, (tuple, list)) and
2979 len(s) == 2 for s in statuses)
2980 for inst, nnames in instdisk.items()
2981 for nname, statuses in nnames.items())
2982 assert set(instdisk) == set(instanceinfo), "instdisk consistency failure"
2987 def _SshNodeSelector(group_uuid, all_nodes):
2988 """Create endless iterators for all potential SSH check hosts.
2991 nodes = [node for node in all_nodes
2992 if (node.group != group_uuid and
2994 keyfunc = operator.attrgetter("group")
2996 return map(itertools.cycle,
2997 [sorted(map(operator.attrgetter("name"), names))
2998 for _, names in itertools.groupby(sorted(nodes, key=keyfunc),
3002 def _SelectSshCheckNodes(cls, group_nodes, group_uuid, all_nodes):
3003 """Choose which nodes should talk to which other nodes.
3005 We will make nodes contact all nodes in their group, and one node from
3008 @warning: This algorithm has a known issue if one node group is much
3009 smaller than others (e.g. just one node). In such a case all other
3010 nodes will talk to the single node.
3013 online_nodes = sorted(node.name for node in group_nodes if not node.offline)
3014 sel = cls._SshNodeSelector(group_uuid, all_nodes)
3016 return (online_nodes,
3017 dict((name, sorted([i.next() for i in sel]))
3018 for name in online_nodes))
3020 def BuildHooksEnv(self):
3023 Cluster-Verify hooks just ran in the post phase and their failure makes
3024 the output be logged in the verify output and the verification to fail.
3028 "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
3031 env.update(("NODE_TAGS_%s" % node.name, " ".join(node.GetTags()))
3032 for node in self.my_node_info.values())
3036 def BuildHooksNodes(self):
3037 """Build hooks nodes.
3040 return ([], self.my_node_names)
3042 def Exec(self, feedback_fn):
3043 """Verify integrity of the node group, performing various test on nodes.
3046 # This method has too many local variables. pylint: disable=R0914
3047 feedback_fn("* Verifying group '%s'" % self.group_info.name)
3049 if not self.my_node_names:
3051 feedback_fn("* Empty node group, skipping verification")
3055 _ErrorIf = self._ErrorIf # pylint: disable=C0103
3056 verbose = self.op.verbose
3057 self._feedback_fn = feedback_fn
3059 vg_name = self.cfg.GetVGName()
3060 drbd_helper = self.cfg.GetDRBDHelper()
3061 cluster = self.cfg.GetClusterInfo()
3062 groupinfo = self.cfg.GetAllNodeGroupsInfo()
3063 hypervisors = cluster.enabled_hypervisors
3064 node_data_list = [self.my_node_info[name] for name in self.my_node_names]
3066 i_non_redundant = [] # Non redundant instances
3067 i_non_a_balanced = [] # Non auto-balanced instances
3068 i_offline = 0 # Count of offline instances
3069 n_offline = 0 # Count of offline nodes
3070 n_drained = 0 # Count of nodes being drained
3071 node_vol_should = {}
3073 # FIXME: verify OS list
3076 filemap = _ComputeAncillaryFiles(cluster, False)
3078 # do local checksums
3079 master_node = self.master_node = self.cfg.GetMasterNode()
3080 master_ip = self.cfg.GetMasterIP()
3082 feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
3085 if self.cfg.GetUseExternalMipScript():
3086 user_scripts.append(constants.EXTERNAL_MASTER_SETUP_SCRIPT)
3088 node_verify_param = {
3089 constants.NV_FILELIST:
3090 utils.UniqueSequence(filename
3091 for files in filemap
3092 for filename in files),
3093 constants.NV_NODELIST:
3094 self._SelectSshCheckNodes(node_data_list, self.group_uuid,
3095 self.all_node_info.values()),
3096 constants.NV_HYPERVISOR: hypervisors,
3097 constants.NV_HVPARAMS:
3098 _GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
3099 constants.NV_NODENETTEST: [(node.name, node.primary_ip, node.secondary_ip)
3100 for node in node_data_list
3101 if not node.offline],
3102 constants.NV_INSTANCELIST: hypervisors,
3103 constants.NV_VERSION: None,
3104 constants.NV_HVINFO: self.cfg.GetHypervisorType(),
3105 constants.NV_NODESETUP: None,
3106 constants.NV_TIME: None,
3107 constants.NV_MASTERIP: (master_node, master_ip),
3108 constants.NV_OSLIST: None,
3109 constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
3110 constants.NV_USERSCRIPTS: user_scripts,
3113 if vg_name is not None:
3114 node_verify_param[constants.NV_VGLIST] = None
3115 node_verify_param[constants.NV_LVLIST] = vg_name
3116 node_verify_param[constants.NV_PVLIST] = [vg_name]
3117 node_verify_param[constants.NV_DRBDLIST] = None
3120 node_verify_param[constants.NV_DRBDHELPER] = drbd_helper
3123 # FIXME: this needs to be changed per node-group, not cluster-wide
3125 default_nicpp = cluster.nicparams[constants.PP_DEFAULT]
3126 if default_nicpp[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
3127 bridges.add(default_nicpp[constants.NIC_LINK])
3128 for instance in self.my_inst_info.values():
3129 for nic in instance.nics:
3130 full_nic = cluster.SimpleFillNIC(nic.nicparams)
3131 if full_nic[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
3132 bridges.add(full_nic[constants.NIC_LINK])
3135 node_verify_param[constants.NV_BRIDGES] = list(bridges)
3137 # Build our expected cluster state
3138 node_image = dict((node.name, self.NodeImage(offline=node.offline,
3140 vm_capable=node.vm_capable))
3141 for node in node_data_list)
3145 for node in self.all_node_info.values():
3146 path = _SupportsOob(self.cfg, node)
3147 if path and path not in oob_paths:
3148 oob_paths.append(path)
3151 node_verify_param[constants.NV_OOB_PATHS] = oob_paths
3153 for instance in self.my_inst_names:
3154 inst_config = self.my_inst_info[instance]
3155 if inst_config.admin_state == constants.ADMINST_OFFLINE:
3158 for nname in inst_config.all_nodes:
3159 if nname not in node_image:
3160 gnode = self.NodeImage(name=nname)
3161 gnode.ghost = (nname not in self.all_node_info)
3162 node_image[nname] = gnode
3164 inst_config.MapLVsByNode(node_vol_should)
3166 pnode = inst_config.primary_node
3167 node_image[pnode].pinst.append(instance)
3169 for snode in inst_config.secondary_nodes:
3170 nimg = node_image[snode]
3171 nimg.sinst.append(instance)
3172 if pnode not in nimg.sbp:
3173 nimg.sbp[pnode] = []
3174 nimg.sbp[pnode].append(instance)
3176 # At this point, we have the in-memory data structures complete,
3177 # except for the runtime information, which we'll gather next
3179 # Due to the way our RPC system works, exact response times cannot be
3180 # guaranteed (e.g. a broken node could run into a timeout). By keeping the
3181 # time before and after executing the request, we can at least have a time
3183 nvinfo_starttime = time.time()
3184 all_nvinfo = self.rpc.call_node_verify(self.my_node_names,
3186 self.cfg.GetClusterName())
3187 nvinfo_endtime = time.time()
3189 if self.extra_lv_nodes and vg_name is not None:
3191 self.rpc.call_node_verify(self.extra_lv_nodes,
3192 {constants.NV_LVLIST: vg_name},
3193 self.cfg.GetClusterName())
3195 extra_lv_nvinfo = {}
3197 all_drbd_map = self.cfg.ComputeDRBDMap()
3199 feedback_fn("* Gathering disk information (%s nodes)" %
3200 len(self.my_node_names))
3201 instdisk = self._CollectDiskInfo(self.my_node_names, node_image,
3204 feedback_fn("* Verifying configuration file consistency")
3206 # If not all nodes are being checked, we need to make sure the master node
3207 # and a non-checked vm_capable node are in the list.
3208 absent_nodes = set(self.all_node_info).difference(self.my_node_info)
3210 vf_nvinfo = all_nvinfo.copy()
3211 vf_node_info = list(self.my_node_info.values())
3212 additional_nodes = []
3213 if master_node not in self.my_node_info:
3214 additional_nodes.append(master_node)
3215 vf_node_info.append(self.all_node_info[master_node])
3216 # Add the first vm_capable node we find which is not included
3217 for node in absent_nodes:
3218 nodeinfo = self.all_node_info[node]
3219 if nodeinfo.vm_capable and not nodeinfo.offline:
3220 additional_nodes.append(node)
3221 vf_node_info.append(self.all_node_info[node])
3223 key = constants.NV_FILELIST
3224 vf_nvinfo.update(self.rpc.call_node_verify(additional_nodes,
3225 {key: node_verify_param[key]},
3226 self.cfg.GetClusterName()))
3228 vf_nvinfo = all_nvinfo
3229 vf_node_info = self.my_node_info.values()
3231 self._VerifyFiles(_ErrorIf, vf_node_info, master_node, vf_nvinfo, filemap)
3233 feedback_fn("* Verifying node status")
3237 for node_i in node_data_list:
3239 nimg = node_image[node]
3243 feedback_fn("* Skipping offline node %s" % (node,))
3247 if node == master_node:
3249 elif node_i.master_candidate:
3250 ntype = "master candidate"
3251 elif node_i.drained:
3257 feedback_fn("* Verifying node %s (%s)" % (node, ntype))
3259 msg = all_nvinfo[node].fail_msg
3260 _ErrorIf(msg, constants.CV_ENODERPC, node, "while contacting node: %s",
3263 nimg.rpc_fail = True
3266 nresult = all_nvinfo[node].payload
3268 nimg.call_ok = self._VerifyNode(node_i, nresult)
3269 self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
3270 self._VerifyNodeNetwork(node_i, nresult)
3271 self._VerifyNodeUserScripts(node_i, nresult)
3272 self._VerifyOob(node_i, nresult)
3275 self._VerifyNodeLVM(node_i, nresult, vg_name)
3276 self._VerifyNodeDrbd(node_i, nresult, self.all_inst_info, drbd_helper,
3279 self._UpdateNodeVolumes(node_i, nresult, nimg, vg_name)
3280 self._UpdateNodeInstances(node_i, nresult, nimg)
3281 self._UpdateNodeInfo(node_i, nresult, nimg, vg_name)
3282 self._UpdateNodeOS(node_i, nresult, nimg)
3284 if not nimg.os_fail:
3285 if refos_img is None:
3287 self._VerifyNodeOS(node_i, nimg, refos_img)
3288 self._VerifyNodeBridges(node_i, nresult, bridges)
3290 # Check whether all running instancies are primary for the node. (This
3291 # can no longer be done from _VerifyInstance below, since some of the
3292 # wrong instances could be from other node groups.)
3293 non_primary_inst = set(nimg.instances).difference(nimg.pinst)
3295 for inst in non_primary_inst:
3296 test = inst in self.all_inst_info
3297 _ErrorIf(test, constants.CV_EINSTANCEWRONGNODE, inst,
3298 "instance should not run on node %s", node_i.name)
3299 _ErrorIf(not test, constants.CV_ENODEORPHANINSTANCE, node_i.name,
3300 "node is running unknown instance %s", inst)
3302 for node, result in extra_lv_nvinfo.items():
3303 self._UpdateNodeVolumes(self.all_node_info[node], result.payload,
3304 node_image[node], vg_name)
3306 feedback_fn("* Verifying instance status")
3307 for instance in self.my_inst_names:
3309 feedback_fn("* Verifying instance %s" % instance)
3310 inst_config = self.my_inst_info[instance]
3311 self._VerifyInstance(instance, inst_config, node_image,
3313 inst_nodes_offline = []
3315 pnode = inst_config.primary_node
3316 pnode_img = node_image[pnode]
3317 _ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
3318 constants.CV_ENODERPC, pnode, "instance %s, connection to"
3319 " primary node failed", instance)
3321 _ErrorIf(inst_config.admin_state == constants.ADMINST_UP and
3323 constants.CV_EINSTANCEBADNODE, instance,
3324 "instance is marked as running and lives on offline node %s",
3325 inst_config.primary_node)
3327 # If the instance is non-redundant we cannot survive losing its primary
3328 # node, so we are not N+1 compliant. On the other hand we have no disk
3329 # templates with more than one secondary so that situation is not well
3331 # FIXME: does not support file-backed instances
3332 if not inst_config.secondary_nodes:
3333 i_non_redundant.append(instance)
3335 _ErrorIf(len(inst_config.secondary_nodes) > 1,
3336 constants.CV_EINSTANCELAYOUT,
3337 instance, "instance has multiple secondary nodes: %s",
3338 utils.CommaJoin(inst_config.secondary_nodes),
3339 code=self.ETYPE_WARNING)
3341 if inst_config.disk_template in constants.DTS_INT_MIRROR:
3342 pnode = inst_config.primary_node
3343 instance_nodes = utils.NiceSort(inst_config.all_nodes)
3344 instance_groups = {}
3346 for node in instance_nodes:
3347 instance_groups.setdefault(self.all_node_info[node].group,
3351 "%s (group %s)" % (utils.CommaJoin(nodes), groupinfo[group].name)
3352 # Sort so that we always list the primary node first.
3353 for group, nodes in sorted(instance_groups.items(),
3354 key=lambda (_, nodes): pnode in nodes,
3357 self._ErrorIf(len(instance_groups) > 1,
3358 constants.CV_EINSTANCESPLITGROUPS,
3359 instance, "instance has primary and secondary nodes in"
3360 " different groups: %s", utils.CommaJoin(pretty_list),
3361 code=self.ETYPE_WARNING)
3363 if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
3364 i_non_a_balanced.append(instance)
3366 for snode in inst_config.secondary_nodes:
3367 s_img = node_image[snode]
3368 _ErrorIf(s_img.rpc_fail and not s_img.offline, constants.CV_ENODERPC,
3369 snode, "instance %s, connection to secondary node failed",
3373 inst_nodes_offline.append(snode)
3375 # warn that the instance lives on offline nodes
3376 _ErrorIf(inst_nodes_offline, constants.CV_EINSTANCEBADNODE, instance,
3377 "instance has offline secondary node(s) %s",
3378 utils.CommaJoin(inst_nodes_offline))
3379 # ... or ghost/non-vm_capable nodes
3380 for node in inst_config.all_nodes:
3381 _ErrorIf(node_image[node].ghost, constants.CV_EINSTANCEBADNODE,
3382 instance, "instance lives on ghost node %s", node)
3383 _ErrorIf(not node_image[node].vm_capable, constants.CV_EINSTANCEBADNODE,
3384 instance, "instance lives on non-vm_capable node %s", node)
3386 feedback_fn("* Verifying orphan volumes")
3387 reserved = utils.FieldSet(*cluster.reserved_lvs)
3389 # We will get spurious "unknown volume" warnings if any node of this group
3390 # is secondary for an instance whose primary is in another group. To avoid
3391 # them, we find these instances and add their volumes to node_vol_should.
3392 for inst in self.all_inst_info.values():
3393 for secondary in inst.secondary_nodes:
3394 if (secondary in self.my_node_info
3395 and inst.name not in self.my_inst_info):
3396 inst.MapLVsByNode(node_vol_should)
3399 self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
3401 if constants.VERIFY_NPLUSONE_MEM not in self.op.skip_checks:
3402 feedback_fn("* Verifying N+1 Memory redundancy")
3403 self._VerifyNPlusOneMemory(node_image, self.my_inst_info)
3405 feedback_fn("* Other Notes")
3407 feedback_fn(" - NOTICE: %d non-redundant instance(s) found."
3408 % len(i_non_redundant))
3410 if i_non_a_balanced:
3411 feedback_fn(" - NOTICE: %d non-auto-balanced instance(s) found."
3412 % len(i_non_a_balanced))
3415 feedback_fn(" - NOTICE: %d offline instance(s) found." % i_offline)
3418 feedback_fn(" - NOTICE: %d offline node(s) found." % n_offline)
3421 feedback_fn(" - NOTICE: %d drained node(s) found." % n_drained)
3425 def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
3426 """Analyze the post-hooks' result
3428 This method analyses the hook result, handles it, and sends some
3429 nicely-formatted feedback back to the user.
3431 @param phase: one of L{constants.HOOKS_PHASE_POST} or
3432 L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
3433 @param hooks_results: the results of the multi-node hooks rpc call
3434 @param feedback_fn: function used send feedback back to the caller
3435 @param lu_result: previous Exec result
3436 @return: the new Exec result, based on the previous result
3440 # We only really run POST phase hooks, only for non-empty groups,
3441 # and are only interested in their results
3442 if not self.my_node_names:
3445 elif phase == constants.HOOKS_PHASE_POST:
3446 # Used to change hooks' output to proper indentation
3447 feedback_fn("* Hooks Results")
3448 assert hooks_results, "invalid result from hooks"
3450 for node_name in hooks_results:
3451 res = hooks_results[node_name]
3453 test = msg and not res.offline
3454 self._ErrorIf(test, constants.CV_ENODEHOOKS, node_name,
3455 "Communication failure in hooks execution: %s", msg)
3456 if res.offline or msg:
3457 # No need to investigate payload if node is offline or gave
3460 for script, hkr, output in res.payload:
3461 test = hkr == constants.HKR_FAIL
3462 self._ErrorIf(test, constants.CV_ENODEHOOKS, node_name,
3463 "Script %s failed, output:", script)
3465 output = self._HOOKS_INDENT_RE.sub(" ", output)
3466 feedback_fn("%s" % output)
3472 class LUClusterVerifyDisks(NoHooksLU):
3473 """Verifies the cluster disks status.
3478 def ExpandNames(self):
3479 self.share_locks = _ShareAll()
3480 self.needed_locks = {
3481 locking.LEVEL_NODEGROUP: locking.ALL_SET,
3484 def Exec(self, feedback_fn):
3485 group_names = self.owned_locks(locking.LEVEL_NODEGROUP)
3487 # Submit one instance of L{opcodes.OpGroupVerifyDisks} per node group
3488 return ResultWithJobs([[opcodes.OpGroupVerifyDisks(group_name=group)]
3489 for group in group_names])
3492 class LUGroupVerifyDisks(NoHooksLU):
3493 """Verifies the status of all disks in a node group.
3498 def ExpandNames(self):
3499 # Raises errors.OpPrereqError on its own if group can't be found
3500 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
3502 self.share_locks = _ShareAll()
3503 self.needed_locks = {
3504 locking.LEVEL_INSTANCE: [],
3505 locking.LEVEL_NODEGROUP: [],
3506 locking.LEVEL_NODE: [],
3509 def DeclareLocks(self, level):
3510 if level == locking.LEVEL_INSTANCE:
3511 assert not self.needed_locks[locking.LEVEL_INSTANCE]
3513 # Lock instances optimistically, needs verification once node and group
3514 # locks have been acquired
3515 self.needed_locks[locking.LEVEL_INSTANCE] = \
3516 self.cfg.GetNodeGroupInstances(self.group_uuid)
3518 elif level == locking.LEVEL_NODEGROUP:
3519 assert not self.needed_locks[locking.LEVEL_NODEGROUP]
3521 self.needed_locks[locking.LEVEL_NODEGROUP] = \
3522 set([self.group_uuid] +
3523 # Lock all groups used by instances optimistically; this requires
3524 # going via the node before it's locked, requiring verification
3527 for instance_name in self.owned_locks(locking.LEVEL_INSTANCE)
3528 for group_uuid in self.cfg.GetInstanceNodeGroups(instance_name)])
3530 elif level == locking.LEVEL_NODE:
3531 # This will only lock the nodes in the group to be verified which contain
3533 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
3534 self._LockInstancesNodes()
3536 # Lock all nodes in group to be verified
3537 assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
3538 member_nodes = self.cfg.GetNodeGroup(self.group_uuid).members
3539 self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
3541 def CheckPrereq(self):
3542 owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
3543 owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
3544 owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
3546 assert self.group_uuid in owned_groups
3548 # Check if locked instances are still correct
3549 _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
3551 # Get instance information
3552 self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
3554 # Check if node groups for locked instances are still correct
3555 _CheckInstancesNodeGroups(self.cfg, self.instances,
3556 owned_groups, owned_nodes, self.group_uuid)
3558 def Exec(self, feedback_fn):
3559 """Verify integrity of cluster disks.
3561 @rtype: tuple of three items
3562 @return: a tuple of (dict of node-to-node_error, list of instances
3563 which need activate-disks, dict of instance: (node, volume) for
3568 res_instances = set()
3571 nv_dict = _MapInstanceDisksToNodes([inst
3572 for inst in self.instances.values()
3573 if inst.admin_state == constants.ADMINST_UP])
3576 nodes = utils.NiceSort(set(self.owned_locks(locking.LEVEL_NODE)) &
3577 set(self.cfg.GetVmCapableNodeList()))
3579 node_lvs = self.rpc.call_lv_list(nodes, [])
3581 for (node, node_res) in node_lvs.items():
3582 if node_res.offline:
3585 msg = node_res.fail_msg
3587 logging.warning("Error enumerating LVs on node %s: %s", node, msg)
3588 res_nodes[node] = msg
3591 for lv_name, (_, _, lv_online) in node_res.payload.items():
3592 inst = nv_dict.pop((node, lv_name), None)
3593 if not (lv_online or inst is None):
3594 res_instances.add(inst)
3596 # any leftover items in nv_dict are missing LVs, let's arrange the data
3598 for key, inst in nv_dict.iteritems():
3599 res_missing.setdefault(inst, []).append(list(key))
3601 return (res_nodes, list(res_instances), res_missing)
3604 class LUClusterRepairDiskSizes(NoHooksLU):
3605 """Verifies the cluster disks sizes.
3610 def ExpandNames(self):
3611 if self.op.instances:
3612 self.wanted_names = _GetWantedInstances(self, self.op.instances)
3613 self.needed_locks = {
3614 locking.LEVEL_NODE_RES: [],
3615 locking.LEVEL_INSTANCE: self.wanted_names,
3617 self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
3619 self.wanted_names = None
3620 self.needed_locks = {
3621 locking.LEVEL_NODE_RES: locking.ALL_SET,
3622 locking.LEVEL_INSTANCE: locking.ALL_SET,
3624 self.share_locks = {
3625 locking.LEVEL_NODE_RES: 1,
3626 locking.LEVEL_INSTANCE: 0,
3629 def DeclareLocks(self, level):
3630 if level == locking.LEVEL_NODE_RES and self.wanted_names is not None:
3631 self._LockInstancesNodes(primary_only=True, level=level)
3633 def CheckPrereq(self):
3634 """Check prerequisites.
3636 This only checks the optional instance list against the existing names.
3639 if self.wanted_names is None:
3640 self.wanted_names = self.owned_locks(locking.LEVEL_INSTANCE)
3642 self.wanted_instances = \
3643 map(compat.snd, self.cfg.GetMultiInstanceInfo(self.wanted_names))
3645 def _EnsureChildSizes(self, disk):
3646 """Ensure children of the disk have the needed disk size.
3648 This is valid mainly for DRBD8 and fixes an issue where the
3649 children have smaller disk size.
3651 @param disk: an L{ganeti.objects.Disk} object
3654 if disk.dev_type == constants.LD_DRBD8:
3655 assert disk.children, "Empty children for DRBD8?"
3656 fchild = disk.children[0]
3657 mismatch = fchild.size < disk.size
3659 self.LogInfo("Child disk has size %d, parent %d, fixing",
3660 fchild.size, disk.size)
3661 fchild.size = disk.size
3663 # and we recurse on this child only, not on the metadev
3664 return self._EnsureChildSizes(fchild) or mismatch
3668 def Exec(self, feedback_fn):
3669 """Verify the size of cluster disks.
3672 # TODO: check child disks too
3673 # TODO: check differences in size between primary/secondary nodes
3675 for instance in self.wanted_instances:
3676 pnode = instance.primary_node
3677 if pnode not in per_node_disks:
3678 per_node_disks[pnode] = []
3679 for idx, disk in enumerate(instance.disks):
3680 per_node_disks[pnode].append((instance, idx, disk))
3682 assert not (frozenset(per_node_disks.keys()) -
3683 self.owned_locks(locking.LEVEL_NODE_RES)), \
3684 "Not owning correct locks"
3685 assert not self.owned_locks(locking.LEVEL_NODE)
3688 for node, dskl in per_node_disks.items():
3689 newl = [v[2].Copy() for v in dskl]
3691 self.cfg.SetDiskID(dsk, node)
3692 result = self.rpc.call_blockdev_getsize(node, newl)
3694 self.LogWarning("Failure in blockdev_getsize call to node"
3695 " %s, ignoring", node)
3697 if len(result.payload) != len(dskl):
3698 logging.warning("Invalid result from node %s: len(dksl)=%d,"
3699 " result.payload=%s", node, len(dskl), result.payload)
3700 self.LogWarning("Invalid result from node %s, ignoring node results",
3703 for ((instance, idx, disk), size) in zip(dskl, result.payload):
3705 self.LogWarning("Disk %d of instance %s did not return size"
3706 " information, ignoring", idx, instance.name)
3708 if not isinstance(size, (int, long)):
3709 self.LogWarning("Disk %d of instance %s did not return valid"
3710 " size information, ignoring", idx, instance.name)
3713 if size != disk.size:
3714 self.LogInfo("Disk %d of instance %s has mismatched size,"
3715 " correcting: recorded %d, actual %d", idx,
3716 instance.name, disk.size, size)
3718 self.cfg.Update(instance, feedback_fn)
3719 changed.append((instance.name, idx, size))
3720 if self._EnsureChildSizes(disk):
3721 self.cfg.Update(instance, feedback_fn)
3722 changed.append((instance.name, idx, disk.size))
3726 class LUClusterRename(LogicalUnit):
3727 """Rename the cluster.
3730 HPATH = "cluster-rename"
3731 HTYPE = constants.HTYPE_CLUSTER
3733 def BuildHooksEnv(self):
3738 "OP_TARGET": self.cfg.GetClusterName(),
3739 "NEW_NAME": self.op.name,
3742 def BuildHooksNodes(self):
3743 """Build hooks nodes.
3746 return ([self.cfg.GetMasterNode()], self.cfg.GetNodeList())
3748 def CheckPrereq(self):
3749 """Verify that the passed name is a valid one.
3752 hostname = netutils.GetHostname(name=self.op.name,
3753 family=self.cfg.GetPrimaryIPFamily())
3755 new_name = hostname.name
3756 self.ip = new_ip = hostname.ip
3757 old_name = self.cfg.GetClusterName()
3758 old_ip = self.cfg.GetMasterIP()
3759 if new_name == old_name and new_ip == old_ip:
3760 raise errors.OpPrereqError("Neither the name nor the IP address of the"
3761 " cluster has changed",
3763 if new_ip != old_ip:
3764 if netutils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
3765 raise errors.OpPrereqError("The given cluster IP address (%s) is"
3766 " reachable on the network" %
3767 new_ip, errors.ECODE_NOTUNIQUE)
3769 self.op.name = new_name
3771 def Exec(self, feedback_fn):
3772 """Rename the cluster.
3775 clustername = self.op.name
3778 # shutdown the master IP
3779 master_params = self.cfg.GetMasterNetworkParameters()
3780 ems = self.cfg.GetUseExternalMipScript()
3781 result = self.rpc.call_node_deactivate_master_ip(master_params.name,
3783 result.Raise("Could not disable the master role")
3786 cluster = self.cfg.GetClusterInfo()
3787 cluster.cluster_name = clustername
3788 cluster.master_ip = new_ip
3789 self.cfg.Update(cluster, feedback_fn)
3791 # update the known hosts file
3792 ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
3793 node_list = self.cfg.GetOnlineNodeList()
3795 node_list.remove(master_params.name)
3798 _UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
3800 master_params.ip = new_ip
3801 result = self.rpc.call_node_activate_master_ip(master_params.name,
3803 msg = result.fail_msg
3805 self.LogWarning("Could not re-enable the master role on"
3806 " the master, please restart manually: %s", msg)
3811 def _ValidateNetmask(cfg, netmask):
3812 """Checks if a netmask is valid.
3814 @type cfg: L{config.ConfigWriter}
3815 @param cfg: The cluster configuration
3817 @param netmask: the netmask to be verified
3818 @raise errors.OpPrereqError: if the validation fails
3821 ip_family = cfg.GetPrimaryIPFamily()
3823 ipcls = netutils.IPAddress.GetClassFromIpFamily(ip_family)
3824 except errors.ProgrammerError:
3825 raise errors.OpPrereqError("Invalid primary ip family: %s." %
3827 if not ipcls.ValidateNetmask(netmask):
3828 raise errors.OpPrereqError("CIDR netmask (%s) not valid" %
3832 class LUClusterSetParams(LogicalUnit):
3833 """Change the parameters of the cluster.
3836 HPATH = "cluster-modify"
3837 HTYPE = constants.HTYPE_CLUSTER
3840 def CheckArguments(self):
3844 if self.op.uid_pool:
3845 uidpool.CheckUidPool(self.op.uid_pool)
3847 if self.op.add_uids:
3848 uidpool.CheckUidPool(self.op.add_uids)
3850 if self.op.remove_uids:
3851 uidpool.CheckUidPool(self.op.remove_uids)
3853 if self.op.master_netmask is not None:
3854 _ValidateNetmask(self.cfg, self.op.master_netmask)
3856 if self.op.diskparams:
3857 for dt_params in self.op.diskparams.values():
3858 utils.ForceDictType(dt_params, constants.DISK_DT_TYPES)
3860 def ExpandNames(self):
3861 # FIXME: in the future maybe other cluster params won't require checking on
3862 # all nodes to be modified.
3863 self.needed_locks = {
3864 locking.LEVEL_NODE: locking.ALL_SET,
3865 locking.LEVEL_INSTANCE: locking.ALL_SET,
3866 locking.LEVEL_NODEGROUP: locking.ALL_SET,
3868 self.share_locks = {
3869 locking.LEVEL_NODE: 1,
3870 locking.LEVEL_INSTANCE: 1,
3871 locking.LEVEL_NODEGROUP: 1,
3874 def BuildHooksEnv(self):
3879 "OP_TARGET": self.cfg.GetClusterName(),
3880 "NEW_VG_NAME": self.op.vg_name,
3883 def BuildHooksNodes(self):
3884 """Build hooks nodes.
3887 mn = self.cfg.GetMasterNode()
3890 def CheckPrereq(self):
3891 """Check prerequisites.
3893 This checks whether the given params don't conflict and
3894 if the given volume group is valid.
3897 if self.op.vg_name is not None and not self.op.vg_name:
3898 if self.cfg.HasAnyDiskOfType(constants.LD_LV):
3899 raise errors.OpPrereqError("Cannot disable lvm storage while lvm-based"
3900 " instances exist", errors.ECODE_INVAL)
3902 if self.op.drbd_helper is not None and not self.op.drbd_helper:
3903 if self.cfg.HasAnyDiskOfType(constants.LD_DRBD8):
3904 raise errors.OpPrereqError("Cannot disable drbd helper while"
3905 " drbd-based instances exist",
3908 node_list = self.owned_locks(locking.LEVEL_NODE)
3910 # if vg_name not None, checks given volume group on all nodes
3912 vglist = self.rpc.call_vg_list(node_list)
3913 for node in node_list:
3914 msg = vglist[node].fail_msg
3916 # ignoring down node
3917 self.LogWarning("Error while gathering data on node %s"
3918 " (ignoring node): %s", node, msg)
3920 vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
3922 constants.MIN_VG_SIZE)
3924 raise errors.OpPrereqError("Error on node '%s': %s" %
3925 (node, vgstatus), errors.ECODE_ENVIRON)
3927 if self.op.drbd_helper:
3928 # checks given drbd helper on all nodes
3929 helpers = self.rpc.call_drbd_helper(node_list)
3930 for (node, ninfo) in self.cfg.GetMultiNodeInfo(node_list):
3932 self.LogInfo("Not checking drbd helper on offline node %s", node)
3934 msg = helpers[node].fail_msg
3936 raise errors.OpPrereqError("Error checking drbd helper on node"
3937 " '%s': %s" % (node, msg),
3938 errors.ECODE_ENVIRON)
3939 node_helper = helpers[node].payload
3940 if node_helper != self.op.drbd_helper:
3941 raise errors.OpPrereqError("Error on node '%s': drbd helper is %s" %
3942 (node, node_helper), errors.ECODE_ENVIRON)
3944 self.cluster = cluster = self.cfg.GetClusterInfo()
3945 # validate params changes
3946 if self.op.beparams:
3947 objects.UpgradeBeParams(self.op.beparams)
3948 utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
3949 self.new_beparams = cluster.SimpleFillBE(self.op.beparams)
3951 if self.op.ndparams:
3952 utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
3953 self.new_ndparams = cluster.SimpleFillND(self.op.ndparams)
3955 # TODO: we need a more general way to handle resetting
3956 # cluster-level parameters to default values
3957 if self.new_ndparams["oob_program"] == "":
3958 self.new_ndparams["oob_program"] = \
3959 constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
3961 if self.op.hv_state:
3962 new_hv_state = _MergeAndVerifyHvState(self.op.hv_state,
3963 self.cluster.hv_state_static)
3964 self.new_hv_state = dict((hv, cluster.SimpleFillHvState(values))
3965 for hv, values in new_hv_state.items())
3967 if self.op.disk_state:
3968 new_disk_state = _MergeAndVerifyDiskState(self.op.disk_state,
3969 self.cluster.disk_state_static)
3970 self.new_disk_state = \
3971 dict((storage, dict((name, cluster.SimpleFillDiskState(values))
3972 for name, values in svalues.items()))
3973 for storage, svalues in new_disk_state.items())
3976 self.new_ipolicy = _GetUpdatedIPolicy(cluster.ipolicy, self.op.ipolicy,
3979 all_instances = self.cfg.GetAllInstancesInfo().values()
3981 for group in self.cfg.GetAllNodeGroupsInfo().values():
3982 instances = frozenset([inst for inst in all_instances
3983 if compat.any(node in group.members
3984 for node in inst.all_nodes)])
3985 new_ipolicy = objects.FillIPolicy(self.new_ipolicy, group.ipolicy)
3986 new = _ComputeNewInstanceViolations(_CalculateGroupIPolicy(cluster,
3988 new_ipolicy, instances)
3990 violations.update(new)
3993 self.LogWarning("After the ipolicy change the following instances"
3994 " violate them: %s",
3995 utils.CommaJoin(utils.NiceSort(violations)))
3997 if self.op.nicparams:
3998 utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
3999 self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
4000 objects.NIC.CheckParameterSyntax(self.new_nicparams)
4003 # check all instances for consistency
4004 for instance in self.cfg.GetAllInstancesInfo().values():
4005 for nic_idx, nic in enumerate(instance.nics):
4006 params_copy = copy.deepcopy(nic.nicparams)
4007 params_filled = objects.FillDict(self.new_nicparams, params_copy)
4009 # check parameter syntax
4011 objects.NIC.CheckParameterSyntax(params_filled)
4012 except errors.ConfigurationError, err:
4013 nic_errors.append("Instance %s, nic/%d: %s" %
4014 (instance.name, nic_idx, err))
4016 # if we're moving instances to routed, check that they have an ip
4017 target_mode = params_filled[constants.NIC_MODE]
4018 if target_mode == constants.NIC_MODE_ROUTED and not nic.ip:
4019 nic_errors.append("Instance %s, nic/%d: routed NIC with no ip"
4020 " address" % (instance.name, nic_idx))
4022 raise errors.OpPrereqError("Cannot apply the change, errors:\n%s" %
4023 "\n".join(nic_errors))
4025 # hypervisor list/parameters
4026 self.new_hvparams = new_hvp = objects.FillDict(cluster.hvparams, {})
4027 if self.op.hvparams:
4028 for hv_name, hv_dict in self.op.hvparams.items():
4029 if hv_name not in self.new_hvparams:
4030 self.new_hvparams[hv_name] = hv_dict
4032 self.new_hvparams[hv_name].update(hv_dict)
4034 # disk template parameters
4035 self.new_diskparams = objects.FillDict(cluster.diskparams, {})
4036 if self.op.diskparams:
4037 for dt_name, dt_params in self.op.diskparams.items():
4038 if dt_name not in self.op.diskparams:
4039 self.new_diskparams[dt_name] = dt_params
4041 self.new_diskparams[dt_name].update(dt_params)
4043 # os hypervisor parameters
4044 self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
4046 for os_name, hvs in self.op.os_hvp.items():
4047 if os_name not in self.new_os_hvp:
4048 self.new_os_hvp[os_name] = hvs
4050 for hv_name, hv_dict in hvs.items():
4051 if hv_name not in self.new_os_hvp[os_name]:
4052 self.new_os_hvp[os_name][hv_name] = hv_dict
4054 self.new_os_hvp[os_name][hv_name].update(hv_dict)
4057 self.new_osp = objects.FillDict(cluster.osparams, {})
4058 if self.op.osparams:
4059 for os_name, osp in self.op.osparams.items():
4060 if os_name not in self.new_osp:
4061 self.new_osp[os_name] = {}
4063 self.new_osp[os_name] = _GetUpdatedParams(self.new_osp[os_name], osp,
4066 if not self.new_osp[os_name]:
4067 # we removed all parameters
4068 del self.new_osp[os_name]
4070 # check the parameter validity (remote check)
4071 _CheckOSParams(self, False, [self.cfg.GetMasterNode()],
4072 os_name, self.new_osp[os_name])
4074 # changes to the hypervisor list
4075 if self.op.enabled_hypervisors is not None:
4076 self.hv_list = self.op.enabled_hypervisors
4077 for hv in self.hv_list:
4078 # if the hypervisor doesn't already exist in the cluster
4079 # hvparams, we initialize it to empty, and then (in both
4080 # cases) we make sure to fill the defaults, as we might not
4081 # have a complete defaults list if the hypervisor wasn't
4083 if hv not in new_hvp:
4085 new_hvp[hv] = objects.FillDict(constants.HVC_DEFAULTS[hv], new_hvp[hv])
4086 utils.ForceDictType(new_hvp[hv], constants.HVS_PARAMETER_TYPES)
4088 self.hv_list = cluster.enabled_hypervisors
4090 if self.op.hvparams or self.op.enabled_hypervisors is not None:
4091 # either the enabled list has changed, or the parameters have, validate
4092 for hv_name, hv_params in self.new_hvparams.items():
4093 if ((self.op.hvparams and hv_name in self.op.hvparams) or
4094 (self.op.enabled_hypervisors and
4095 hv_name in self.op.enabled_hypervisors)):
4096 # either this is a new hypervisor, or its parameters have changed
4097 hv_class = hypervisor.GetHypervisor(hv_name)
4098 utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
4099 hv_class.CheckParameterSyntax(hv_params)
4100 _CheckHVParams(self, node_list, hv_name, hv_params)
4103 # no need to check any newly-enabled hypervisors, since the
4104 # defaults have already been checked in the above code-block
4105 for os_name, os_hvp in self.new_os_hvp.items():
4106 for hv_name, hv_params in os_hvp.items():
4107 utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
4108 # we need to fill in the new os_hvp on top of the actual hv_p
4109 cluster_defaults = self.new_hvparams.get(hv_name, {})
4110 new_osp = objects.FillDict(cluster_defaults, hv_params)
4111 hv_class = hypervisor.GetHypervisor(hv_name)
4112 hv_class.CheckParameterSyntax(new_osp)
4113 _CheckHVParams(self, node_list, hv_name, new_osp)
4115 if self.op.default_iallocator:
4116 alloc_script = utils.FindFile(self.op.default_iallocator,
4117 constants.IALLOCATOR_SEARCH_PATH,
4119 if alloc_script is None:
4120 raise errors.OpPrereqError("Invalid default iallocator script '%s'"
4121 " specified" % self.op.default_iallocator,
4124 def Exec(self, feedback_fn):
4125 """Change the parameters of the cluster.
4128 if self.op.vg_name is not None:
4129 new_volume = self.op.vg_name
4132 if new_volume != self.cfg.GetVGName():
4133 self.cfg.SetVGName(new_volume)
4135 feedback_fn("Cluster LVM configuration already in desired"
4136 " state, not changing")
4137 if self.op.drbd_helper is not None:
4138 new_helper = self.op.drbd_helper
4141 if new_helper != self.cfg.GetDRBDHelper():
4142 self.cfg.SetDRBDHelper(new_helper)
4144 feedback_fn("Cluster DRBD helper already in desired state,"
4146 if self.op.hvparams:
4147 self.cluster.hvparams = self.new_hvparams
4149 self.cluster.os_hvp = self.new_os_hvp
4150 if self.op.enabled_hypervisors is not None:
4151 self.cluster.hvparams = self.new_hvparams
4152 self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
4153 if self.op.beparams:
4154 self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
4155 if self.op.nicparams:
4156 self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
4158 self.cluster.ipolicy = self.new_ipolicy
4159 if self.op.osparams:
4160 self.cluster.osparams = self.new_osp
4161 if self.op.ndparams:
4162 self.cluster.ndparams = self.new_ndparams
4163 if self.op.diskparams:
4164 self.cluster.diskparams = self.new_diskparams
4165 if self.op.hv_state:
4166 self.cluster.hv_state_static = self.new_hv_state
4167 if self.op.disk_state:
4168 self.cluster.disk_state_static = self.new_disk_state
4170 if self.op.candidate_pool_size is not None:
4171 self.cluster.candidate_pool_size = self.op.candidate_pool_size
4172 # we need to update the pool size here, otherwise the save will fail
4173 _AdjustCandidatePool(self, [])
4175 if self.op.maintain_node_health is not None:
4176 if self.op.maintain_node_health and not constants.ENABLE_CONFD:
4177 feedback_fn("Note: CONFD was disabled at build time, node health"
4178 " maintenance is not useful (still enabling it)")
4179 self.cluster.maintain_node_health = self.op.maintain_node_health
4181 if self.op.prealloc_wipe_disks is not None:
4182 self.cluster.prealloc_wipe_disks = self.op.prealloc_wipe_disks
4184 if self.op.add_uids is not None:
4185 uidpool.AddToUidPool(self.cluster.uid_pool, self.op.add_uids)
4187 if self.op.remove_uids is not None:
4188 uidpool.RemoveFromUidPool(self.cluster.uid_pool, self.op.remove_uids)
4190 if self.op.uid_pool is not None:
4191 self.cluster.uid_pool = self.op.uid_pool
4193 if self.op.default_iallocator is not None:
4194 self.cluster.default_iallocator = self.op.default_iallocator
4196 if self.op.reserved_lvs is not None:
4197 self.cluster.reserved_lvs = self.op.reserved_lvs
4199 if self.op.use_external_mip_script is not None:
4200 self.cluster.use_external_mip_script = self.op.use_external_mip_script
4202 def helper_os(aname, mods, desc):
4204 lst = getattr(self.cluster, aname)
4205 for key, val in mods:
4206 if key == constants.DDM_ADD:
4208 feedback_fn("OS %s already in %s, ignoring" % (val, desc))
4211 elif key == constants.DDM_REMOVE:
4215 feedback_fn("OS %s not found in %s, ignoring" % (val, desc))
4217 raise errors.ProgrammerError("Invalid modification '%s'" % key)
4219 if self.op.hidden_os:
4220 helper_os("hidden_os", self.op.hidden_os, "hidden")
4222 if self.op.blacklisted_os:
4223 helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
4225 if self.op.master_netdev:
4226 master_params = self.cfg.GetMasterNetworkParameters()
4227 ems = self.cfg.GetUseExternalMipScript()
4228 feedback_fn("Shutting down master ip on the current netdev (%s)" %
4229 self.cluster.master_netdev)
4230 result = self.rpc.call_node_deactivate_master_ip(master_params.name,
4232 result.Raise("Could not disable the master ip")
4233 feedback_fn("Changing master_netdev from %s to %s" %
4234 (master_params.netdev, self.op.master_netdev))
4235 self.cluster.master_netdev = self.op.master_netdev
4237 if self.op.master_netmask:
4238 master_params = self.cfg.GetMasterNetworkParameters()
4239 feedback_fn("Changing master IP netmask to %s" % self.op.master_netmask)
4240 result = self.rpc.call_node_change_master_netmask(master_params.name,
4241 master_params.netmask,
4242 self.op.master_netmask,
4244 master_params.netdev)
4246 msg = "Could not change the master IP netmask: %s" % result.fail_msg
4249 self.cluster.master_netmask = self.op.master_netmask
4251 self.cfg.Update(self.cluster, feedback_fn)
4253 if self.op.master_netdev:
4254 master_params = self.cfg.GetMasterNetworkParameters()
4255 feedback_fn("Starting the master ip on the new master netdev (%s)" %
4256 self.op.master_netdev)
4257 ems = self.cfg.GetUseExternalMipScript()
4258 result = self.rpc.call_node_activate_master_ip(master_params.name,
4261 self.LogWarning("Could not re-enable the master ip on"
4262 " the master, please restart manually: %s",
4266 def _UploadHelper(lu, nodes, fname):
4267 """Helper for uploading a file and showing warnings.
4270 if os.path.exists(fname):
4271 result = lu.rpc.call_upload_file(nodes, fname)
4272 for to_node, to_result in result.items():
4273 msg = to_result.fail_msg
4275 msg = ("Copy of file %s to node %s failed: %s" %
4276 (fname, to_node, msg))
4277 lu.proc.LogWarning(msg)
4280 def _ComputeAncillaryFiles(cluster, redist):
4281 """Compute files external to Ganeti which need to be consistent.
4283 @type redist: boolean
4284 @param redist: Whether to include files which need to be redistributed
4287 # Compute files for all nodes
4289 constants.SSH_KNOWN_HOSTS_FILE,
4290 constants.CONFD_HMAC_KEY,
4291 constants.CLUSTER_DOMAIN_SECRET_FILE,
4292 constants.SPICE_CERT_FILE,
4293 constants.SPICE_CACERT_FILE,
4294 constants.RAPI_USERS_FILE,
4298 files_all.update(constants.ALL_CERT_FILES)
4299 files_all.update(ssconf.SimpleStore().GetFileList())
4301 # we need to ship at least the RAPI certificate
4302 files_all.add(constants.RAPI_CERT_FILE)
4304 if cluster.modify_etc_hosts:
4305 files_all.add(constants.ETC_HOSTS)
4307 # Files which are optional, these must:
4308 # - be present in one other category as well
4309 # - either exist or not exist on all nodes of that category (mc, vm all)
4311 constants.RAPI_USERS_FILE,
4314 # Files which should only be on master candidates
4318 files_mc.add(constants.CLUSTER_CONF_FILE)
4320 # FIXME: this should also be replicated but Ganeti doesn't support files_mc
4322 files_mc.add(constants.DEFAULT_MASTER_SETUP_SCRIPT)
4324 # Files which should only be on VM-capable nodes
4325 files_vm = set(filename
4326 for hv_name in cluster.enabled_hypervisors
4327 for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[0])
4329 files_opt |= set(filename
4330 for hv_name in cluster.enabled_hypervisors
4331 for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[1])
4333 # Filenames in each category must be unique
4334 all_files_set = files_all | files_mc | files_vm
4335 assert (len(all_files_set) ==
4336 sum(map(len, [files_all, files_mc, files_vm]))), \
4337 "Found file listed in more than one file list"
4339 # Optional files must be present in one other category
4340 assert all_files_set.issuperset(files_opt), \
4341 "Optional file not in a different required list"
4343 return (files_all, files_opt, files_mc, files_vm)
4346 def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
4347 """Distribute additional files which are part of the cluster configuration.
4349 ConfigWriter takes care of distributing the config and ssconf files, but
4350 there are more files which should be distributed to all nodes. This function
4351 makes sure those are copied.
4353 @param lu: calling logical unit
4354 @param additional_nodes: list of nodes not in the config to distribute to
4355 @type additional_vm: boolean
4356 @param additional_vm: whether the additional nodes are vm-capable or not
4359 # Gather target nodes
4360 cluster = lu.cfg.GetClusterInfo()
4361 master_info = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
4363 online_nodes = lu.cfg.GetOnlineNodeList()
4364 vm_nodes = lu.cfg.GetVmCapableNodeList()
4366 if additional_nodes is not None:
4367 online_nodes.extend(additional_nodes)
4369 vm_nodes.extend(additional_nodes)
4371 # Never distribute to master node
4372 for nodelist in [online_nodes, vm_nodes]:
4373 if master_info.name in nodelist:
4374 nodelist.remove(master_info.name)
4377 (files_all, _, files_mc, files_vm) = \
4378 _ComputeAncillaryFiles(cluster, True)
4380 # Never re-distribute configuration file from here
4381 assert not (constants.CLUSTER_CONF_FILE in files_all or
4382 constants.CLUSTER_CONF_FILE in files_vm)
4383 assert not files_mc, "Master candidates not handled in this function"
4386 (online_nodes, files_all),
4387 (vm_nodes, files_vm),
4391 for (node_list, files) in filemap:
4393 _UploadHelper(lu, node_list, fname)
4396 class LUClusterRedistConf(NoHooksLU):
4397 """Force the redistribution of cluster configuration.
4399 This is a very simple LU.
4404 def ExpandNames(self):
4405 self.needed_locks = {
4406 locking.LEVEL_NODE: locking.ALL_SET,
4408 self.share_locks[locking.LEVEL_NODE] = 1
4410 def Exec(self, feedback_fn):
4411 """Redistribute the configuration.
4414 self.cfg.Update(self.cfg.GetClusterInfo(), feedback_fn)
4415 _RedistributeAncillaryFiles(self)
4418 class LUClusterActivateMasterIp(NoHooksLU):
4419 """Activate the master IP on the master node.
4422 def Exec(self, feedback_fn):
4423 """Activate the master IP.
4426 master_params = self.cfg.GetMasterNetworkParameters()
4427 ems = self.cfg.GetUseExternalMipScript()
4428 result = self.rpc.call_node_activate_master_ip(master_params.name,
4430 result.Raise("Could not activate the master IP")
4433 class LUClusterDeactivateMasterIp(NoHooksLU):
4434 """Deactivate the master IP on the master node.
4437 def Exec(self, feedback_fn):
4438 """Deactivate the master IP.
4441 master_params = self.cfg.GetMasterNetworkParameters()
4442 ems = self.cfg.GetUseExternalMipScript()
4443 result = self.rpc.call_node_deactivate_master_ip(master_params.name,
4445 result.Raise("Could not deactivate the master IP")
4448 def _WaitForSync(lu, instance, disks=None, oneshot=False):
4449 """Sleep and poll for an instance's disk to sync.
4452 if not instance.disks or disks is not None and not disks:
4455 disks = _ExpandCheckDisks(instance, disks)
4458 lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
4460 node = instance.primary_node
4463 lu.cfg.SetDiskID(dev, node)
4465 # TODO: Convert to utils.Retry
4468 degr_retries = 10 # in seconds, as we sleep 1 second each time
4472 cumul_degraded = False
4473 rstats = lu.rpc.call_blockdev_getmirrorstatus(node, disks)
4474 msg = rstats.fail_msg
4476 lu.LogWarning("Can't get any data from node %s: %s", node, msg)
4479 raise errors.RemoteError("Can't contact node %s for mirror data,"
4480 " aborting." % node)
4483 rstats = rstats.payload
4485 for i, mstat in enumerate(rstats):
4487 lu.LogWarning("Can't compute data for node %s/%s",
4488 node, disks[i].iv_name)
4491 cumul_degraded = (cumul_degraded or
4492 (mstat.is_degraded and mstat.sync_percent is None))
4493 if mstat.sync_percent is not None:
4495 if mstat.estimated_time is not None:
4496 rem_time = ("%s remaining (estimated)" %
4497 utils.FormatSeconds(mstat.estimated_time))
4498 max_time = mstat.estimated_time
4500 rem_time = "no time estimate"
4501 lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
4502 (disks[i].iv_name, mstat.sync_percent, rem_time))
4504 # if we're done but degraded, let's do a few small retries, to
4505 # make sure we see a stable and not transient situation; therefore
4506 # we force restart of the loop
4507 if (done or oneshot) and cumul_degraded and degr_retries > 0:
4508 logging.info("Degraded disks found, %d retries left", degr_retries)
4516 time.sleep(min(60, max_time))
4519 lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
4520 return not cumul_degraded
4523 def _BlockdevFind(lu, node, dev, instance):
4524 """Wrapper around call_blockdev_find to annotate diskparams.
4526 @param lu: A reference to the lu object
4527 @param node: The node to call out
4528 @param dev: The device to find
4529 @param instance: The instance object the device belongs to
4530 @returns The result of the rpc call
4533 (disk,) = _AnnotateDiskParams(instance, [dev], lu.cfg)
4534 return lu.rpc.call_blockdev_find(node, disk)
4537 def _CheckDiskConsistency(lu, instance, dev, node, on_primary, ldisk=False):
4538 """Wrapper around L{_CheckDiskConsistencyInner}.
4541 (disk,) = _AnnotateDiskParams(instance, [dev], lu.cfg)
4542 return _CheckDiskConsistencyInner(lu, instance, disk, node, on_primary,
4546 def _CheckDiskConsistencyInner(lu, instance, dev, node, on_primary,
4548 """Check that mirrors are not degraded.
4550 @attention: The device has to be annotated already.
4552 The ldisk parameter, if True, will change the test from the
4553 is_degraded attribute (which represents overall non-ok status for
4554 the device(s)) to the ldisk (representing the local storage status).
4557 lu.cfg.SetDiskID(dev, node)
4561 if on_primary or dev.AssembleOnSecondary():
4562 rstats = lu.rpc.call_blockdev_find(node, dev)
4563 msg = rstats.fail_msg
4565 lu.LogWarning("Can't find disk on node %s: %s", node, msg)
4567 elif not rstats.payload:
4568 lu.LogWarning("Can't find disk on node %s", node)
4572 result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
4574 result = result and not rstats.payload.is_degraded
4577 for child in dev.children:
4578 result = result and _CheckDiskConsistencyInner(lu, instance, child, node,
4584 class LUOobCommand(NoHooksLU):
4585 """Logical unit for OOB handling.
4589 _SKIP_MASTER = (constants.OOB_POWER_OFF, constants.OOB_POWER_CYCLE)
4591 def ExpandNames(self):
4592 """Gather locks we need.
4595 if self.op.node_names:
4596 self.op.node_names = _GetWantedNodes(self, self.op.node_names)
4597 lock_names = self.op.node_names
4599 lock_names = locking.ALL_SET
4601 self.needed_locks = {
4602 locking.LEVEL_NODE: lock_names,
4605 def CheckPrereq(self):
4606 """Check prerequisites.
4609 - the node exists in the configuration
4612 Any errors are signaled by raising errors.OpPrereqError.
4616 self.master_node = self.cfg.GetMasterNode()
4618 assert self.op.power_delay >= 0.0
4620 if self.op.node_names:
4621 if (self.op.command in self._SKIP_MASTER and
4622 self.master_node in self.op.node_names):
4623 master_node_obj = self.cfg.GetNodeInfo(self.master_node)
4624 master_oob_handler = _SupportsOob(self.cfg, master_node_obj)
4626 if master_oob_handler:
4627 additional_text = ("run '%s %s %s' if you want to operate on the"
4628 " master regardless") % (master_oob_handler,
4632 additional_text = "it does not support out-of-band operations"
4634 raise errors.OpPrereqError(("Operating on the master node %s is not"
4635 " allowed for %s; %s") %
4636 (self.master_node, self.op.command,
4637 additional_text), errors.ECODE_INVAL)
4639 self.op.node_names = self.cfg.GetNodeList()
4640 if self.op.command in self._SKIP_MASTER:
4641 self.op.node_names.remove(self.master_node)
4643 if self.op.command in self._SKIP_MASTER:
4644 assert self.master_node not in self.op.node_names
4646 for (node_name, node) in self.cfg.GetMultiNodeInfo(self.op.node_names):
4648 raise errors.OpPrereqError("Node %s not found" % node_name,
4651 self.nodes.append(node)
4653 if (not self.op.ignore_status and
4654 (self.op.command == constants.OOB_POWER_OFF and not node.offline)):
4655 raise errors.OpPrereqError(("Cannot power off node %s because it is"
4656 " not marked offline") % node_name,
4659 def Exec(self, feedback_fn):
4660 """Execute OOB and return result if we expect any.
4663 master_node = self.master_node
4666 for idx, node in enumerate(utils.NiceSort(self.nodes,
4667 key=lambda node: node.name)):
4668 node_entry = [(constants.RS_NORMAL, node.name)]
4669 ret.append(node_entry)
4671 oob_program = _SupportsOob(self.cfg, node)
4674 node_entry.append((constants.RS_UNAVAIL, None))
4677 logging.info("Executing out-of-band command '%s' using '%s' on %s",
4678 self.op.command, oob_program, node.name)
4679 result = self.rpc.call_run_oob(master_node, oob_program,
4680 self.op.command, node.name,
4684 self.LogWarning("Out-of-band RPC failed on node '%s': %s",
4685 node.name, result.fail_msg)
4686 node_entry.append((constants.RS_NODATA, None))
4689 self._CheckPayload(result)
4690 except errors.OpExecError, err:
4691 self.LogWarning("Payload returned by node '%s' is not valid: %s",
4693 node_entry.append((constants.RS_NODATA, None))
4695 if self.op.command == constants.OOB_HEALTH:
4696 # For health we should log important events
4697 for item, status in result.payload:
4698 if status in [constants.OOB_STATUS_WARNING,
4699 constants.OOB_STATUS_CRITICAL]:
4700 self.LogWarning("Item '%s' on node '%s' has status '%s'",
4701 item, node.name, status)
4703 if self.op.command == constants.OOB_POWER_ON:
4705 elif self.op.command == constants.OOB_POWER_OFF:
4706 node.powered = False
4707 elif self.op.command == constants.OOB_POWER_STATUS:
4708 powered = result.payload[constants.OOB_POWER_STATUS_POWERED]
4709 if powered != node.powered:
4710 logging.warning(("Recorded power state (%s) of node '%s' does not"
4711 " match actual power state (%s)"), node.powered,
4714 # For configuration changing commands we should update the node
4715 if self.op.command in (constants.OOB_POWER_ON,
4716 constants.OOB_POWER_OFF):
4717 self.cfg.Update(node, feedback_fn)
4719 node_entry.append((constants.RS_NORMAL, result.payload))
4721 if (self.op.command == constants.OOB_POWER_ON and
4722 idx < len(self.nodes) - 1):
4723 time.sleep(self.op.power_delay)
4727 def _CheckPayload(self, result):
4728 """Checks if the payload is valid.
4730 @param result: RPC result
4731 @raises errors.OpExecError: If payload is not valid
4735 if self.op.command == constants.OOB_HEALTH:
4736 if not isinstance(result.payload, list):
4737 errs.append("command 'health' is expected to return a list but got %s" %
4738 type(result.payload))
4740 for item, status in result.payload:
4741 if status not in constants.OOB_STATUSES:
4742 errs.append("health item '%s' has invalid status '%s'" %
4745 if self.op.command == constants.OOB_POWER_STATUS:
4746 if not isinstance(result.payload, dict):
4747 errs.append("power-status is expected to return a dict but got %s" %
4748 type(result.payload))
4750 if self.op.command in [
4751 constants.OOB_POWER_ON,
4752 constants.OOB_POWER_OFF,
4753 constants.OOB_POWER_CYCLE,
4755 if result.payload is not None:
4756 errs.append("%s is expected to not return payload but got '%s'" %
4757 (self.op.command, result.payload))
4760 raise errors.OpExecError("Check of out-of-band payload failed due to %s" %
4761 utils.CommaJoin(errs))
4764 class _OsQuery(_QueryBase):
4765 FIELDS = query.OS_FIELDS
4767 def ExpandNames(self, lu):
4768 # Lock all nodes in shared mode
4769 # Temporary removal of locks, should be reverted later
4770 # TODO: reintroduce locks when they are lighter-weight
4771 lu.needed_locks = {}
4772 #self.share_locks[locking.LEVEL_NODE] = 1
4773 #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4775 # The following variables interact with _QueryBase._GetNames
4777 self.wanted = self.names
4779 self.wanted = locking.ALL_SET
4781 self.do_locking = self.use_locking
4783 def DeclareLocks(self, lu, level):
4787 def _DiagnoseByOS(rlist):
4788 """Remaps a per-node return list into an a per-os per-node dictionary
4790 @param rlist: a map with node names as keys and OS objects as values
4793 @return: a dictionary with osnames as keys and as value another
4794 map, with nodes as keys and tuples of (path, status, diagnose,
4795 variants, parameters, api_versions) as values, eg::
4797 {"debian-etch": {"node1": [(/usr/lib/..., True, "", [], []),
4798 (/srv/..., False, "invalid api")],
4799 "node2": [(/srv/..., True, "", [], [])]}
4804 # we build here the list of nodes that didn't fail the RPC (at RPC
4805 # level), so that nodes with a non-responding node daemon don't
4806 # make all OSes invalid
4807 good_nodes = [node_name for node_name in rlist
4808 if not rlist[node_name].fail_msg]
4809 for node_name, nr in rlist.items():
4810 if nr.fail_msg or not nr.payload:
4812 for (name, path, status, diagnose, variants,
4813 params, api_versions) in nr.payload:
4814 if name not in all_os:
4815 # build a list of nodes for this os containing empty lists
4816 # for each node in node_list
4818 for nname in good_nodes:
4819 all_os[name][nname] = []
4820 # convert params from [name, help] to (name, help)
4821 params = [tuple(v) for v in params]
4822 all_os[name][node_name].append((path, status, diagnose,
4823 variants, params, api_versions))
4826 def _GetQueryData(self, lu):
4827 """Computes the list of nodes and their attributes.
4830 # Locking is not used
4831 assert not (compat.any(lu.glm.is_owned(level)
4832 for level in locking.LEVELS
4833 if level != locking.LEVEL_CLUSTER) or
4834 self.do_locking or self.use_locking)
4836 valid_nodes = [node.name
4837 for node in lu.cfg.GetAllNodesInfo().values()
4838 if not node.offline and node.vm_capable]
4839 pol = self._DiagnoseByOS(lu.rpc.call_os_diagnose(valid_nodes))
4840 cluster = lu.cfg.GetClusterInfo()
4844 for (os_name, os_data) in pol.items():
4845 info = query.OsInfo(name=os_name, valid=True, node_status=os_data,
4846 hidden=(os_name in cluster.hidden_os),
4847 blacklisted=(os_name in cluster.blacklisted_os))
4851 api_versions = set()
4853 for idx, osl in enumerate(os_data.values()):
4854 info.valid = bool(info.valid and osl and osl[0][1])
4858 (node_variants, node_params, node_api) = osl[0][3:6]
4861 variants.update(node_variants)
4862 parameters.update(node_params)
4863 api_versions.update(node_api)
4865 # Filter out inconsistent values
4866 variants.intersection_update(node_variants)
4867 parameters.intersection_update(node_params)
4868 api_versions.intersection_update(node_api)
4870 info.variants = list(variants)
4871 info.parameters = list(parameters)
4872 info.api_versions = list(api_versions)
4874 data[os_name] = info
4876 # Prepare data in requested order
4877 return [data[name] for name in self._GetNames(lu, pol.keys(), None)
4881 class LUOsDiagnose(NoHooksLU):
4882 """Logical unit for OS diagnose/query.
4888 def _BuildFilter(fields, names):
4889 """Builds a filter for querying OSes.
4892 name_filter = qlang.MakeSimpleFilter("name", names)
4894 # Legacy behaviour: Hide hidden, blacklisted or invalid OSes if the
4895 # respective field is not requested
4896 status_filter = [[qlang.OP_NOT, [qlang.OP_TRUE, fname]]
4897 for fname in ["hidden", "blacklisted"]
4898 if fname not in fields]
4899 if "valid" not in fields:
4900 status_filter.append([qlang.OP_TRUE, "valid"])
4903 status_filter.insert(0, qlang.OP_AND)
4905 status_filter = None
4907 if name_filter and status_filter:
4908 return [qlang.OP_AND, name_filter, status_filter]
4912 return status_filter
4914 def CheckArguments(self):
4915 self.oq = _OsQuery(self._BuildFilter(self.op.output_fields, self.op.names),
4916 self.op.output_fields, False)
4918 def ExpandNames(self):
4919 self.oq.ExpandNames(self)
4921 def Exec(self, feedback_fn):
4922 return self.oq.OldStyleQuery(self)
4925 class LUNodeRemove(LogicalUnit):
4926 """Logical unit for removing a node.
4929 HPATH = "node-remove"
4930 HTYPE = constants.HTYPE_NODE
4932 def BuildHooksEnv(self):
4937 "OP_TARGET": self.op.node_name,
4938 "NODE_NAME": self.op.node_name,
4941 def BuildHooksNodes(self):
4942 """Build hooks nodes.
4944 This doesn't run on the target node in the pre phase as a failed
4945 node would then be impossible to remove.
4948 all_nodes = self.cfg.GetNodeList()
4950 all_nodes.remove(self.op.node_name)
4953 return (all_nodes, all_nodes)
4955 def CheckPrereq(self):
4956 """Check prerequisites.
4959 - the node exists in the configuration
4960 - it does not have primary or secondary instances
4961 - it's not the master
4963 Any errors are signaled by raising errors.OpPrereqError.
4966 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4967 node = self.cfg.GetNodeInfo(self.op.node_name)
4968 assert node is not None
4970 masternode = self.cfg.GetMasterNode()
4971 if node.name == masternode:
4972 raise errors.OpPrereqError("Node is the master node, failover to another"
4973 " node is required", errors.ECODE_INVAL)
4975 for instance_name, instance in self.cfg.GetAllInstancesInfo().items():
4976 if node.name in instance.all_nodes:
4977 raise errors.OpPrereqError("Instance %s is still running on the node,"
4978 " please remove first" % instance_name,
4980 self.op.node_name = node.name
4983 def Exec(self, feedback_fn):
4984 """Removes the node from the cluster.
4988 logging.info("Stopping the node daemon and removing configs from node %s",
4991 modify_ssh_setup = self.cfg.GetClusterInfo().modify_ssh_setup
4993 assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER), \
4996 # Promote nodes to master candidate as needed
4997 _AdjustCandidatePool(self, exceptions=[node.name])
4998 self.context.RemoveNode(node.name)
5000 # Run post hooks on the node before it's removed
5001 _RunPostHook(self, node.name)
5003 result = self.rpc.call_node_leave_cluster(node.name, modify_ssh_setup)
5004 msg = result.fail_msg
5006 self.LogWarning("Errors encountered on the remote node while leaving"
5007 " the cluster: %s", msg)
5009 # Remove node from our /etc/hosts
5010 if self.cfg.GetClusterInfo().modify_etc_hosts:
5011 master_node = self.cfg.GetMasterNode()
5012 result = self.rpc.call_etc_hosts_modify(master_node,
5013 constants.ETC_HOSTS_REMOVE,
5015 result.Raise("Can't update hosts file with new host data")
5016 _RedistributeAncillaryFiles(self)
5019 class _NodeQuery(_QueryBase):
5020 FIELDS = query.NODE_FIELDS
5022 def ExpandNames(self, lu):
5023 lu.needed_locks = {}
5024 lu.share_locks = _ShareAll()
5027 self.wanted = _GetWantedNodes(lu, self.names)
5029 self.wanted = locking.ALL_SET
5031 self.do_locking = (self.use_locking and
5032 query.NQ_LIVE in self.requested_data)
5035 # If any non-static field is requested we need to lock the nodes
5036 lu.needed_locks[locking.LEVEL_NODE] = self.wanted
5038 def DeclareLocks(self, lu, level):
5041 def _GetQueryData(self, lu):
5042 """Computes the list of nodes and their attributes.
5045 all_info = lu.cfg.GetAllNodesInfo()
5047 nodenames = self._GetNames(lu, all_info.keys(), locking.LEVEL_NODE)
5049 # Gather data as requested
5050 if query.NQ_LIVE in self.requested_data:
5051 # filter out non-vm_capable nodes
5052 toquery_nodes = [name for name in nodenames if all_info[name].vm_capable]
5054 node_data = lu.rpc.call_node_info(toquery_nodes, [lu.cfg.GetVGName()],
5055 [lu.cfg.GetHypervisorType()])
5056 live_data = dict((name, _MakeLegacyNodeInfo(nresult.payload))
5057 for (name, nresult) in node_data.items()
5058 if not nresult.fail_msg and nresult.payload)
5062 if query.NQ_INST in self.requested_data:
5063 node_to_primary = dict([(name, set()) for name in nodenames])
5064 node_to_secondary = dict([(name, set()) for name in nodenames])
5066 inst_data = lu.cfg.GetAllInstancesInfo()
5068 for inst in inst_data.values():
5069 if inst.primary_node in node_to_primary:
5070 node_to_primary[inst.primary_node].add(inst.name)
5071 for secnode in inst.secondary_nodes:
5072 if secnode in node_to_secondary:
5073 node_to_secondary[secnode].add(inst.name)
5075 node_to_primary = None
5076 node_to_secondary = None
5078 if query.NQ_OOB in self.requested_data:
5079 oob_support = dict((name, bool(_SupportsOob(lu.cfg, node)))
5080 for name, node in all_info.iteritems())
5084 if query.NQ_GROUP in self.requested_data:
5085 groups = lu.cfg.GetAllNodeGroupsInfo()
5089 return query.NodeQueryData([all_info[name] for name in nodenames],
5090 live_data, lu.cfg.GetMasterNode(),
5091 node_to_primary, node_to_secondary, groups,
5092 oob_support, lu.cfg.GetClusterInfo())
5095 class LUNodeQuery(NoHooksLU):
5096 """Logical unit for querying nodes.
5099 # pylint: disable=W0142
5102 def CheckArguments(self):
5103 self.nq = _NodeQuery(qlang.MakeSimpleFilter("name", self.op.names),
5104 self.op.output_fields, self.op.use_locking)
5106 def ExpandNames(self):
5107 self.nq.ExpandNames(self)
5109 def DeclareLocks(self, level):
5110 self.nq.DeclareLocks(self, level)
5112 def Exec(self, feedback_fn):
5113 return self.nq.OldStyleQuery(self)
5116 class LUNodeQueryvols(NoHooksLU):
5117 """Logical unit for getting volumes on node(s).
5121 _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
5122 _FIELDS_STATIC = utils.FieldSet("node")
5124 def CheckArguments(self):
5125 _CheckOutputFields(static=self._FIELDS_STATIC,
5126 dynamic=self._FIELDS_DYNAMIC,
5127 selected=self.op.output_fields)
5129 def ExpandNames(self):
5130 self.share_locks = _ShareAll()
5131 self.needed_locks = {}
5133 if not self.op.nodes:
5134 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
5136 self.needed_locks[locking.LEVEL_NODE] = \
5137 _GetWantedNodes(self, self.op.nodes)
5139 def Exec(self, feedback_fn):
5140 """Computes the list of nodes and their attributes.
5143 nodenames = self.owned_locks(locking.LEVEL_NODE)
5144 volumes = self.rpc.call_node_volumes(nodenames)
5146 ilist = self.cfg.GetAllInstancesInfo()
5147 vol2inst = _MapInstanceDisksToNodes(ilist.values())
5150 for node in nodenames:
5151 nresult = volumes[node]
5154 msg = nresult.fail_msg
5156 self.LogWarning("Can't compute volume data on node %s: %s", node, msg)
5159 node_vols = sorted(nresult.payload,
5160 key=operator.itemgetter("dev"))
5162 for vol in node_vols:
5164 for field in self.op.output_fields:
5167 elif field == "phys":
5171 elif field == "name":
5173 elif field == "size":
5174 val = int(float(vol["size"]))
5175 elif field == "instance":
5176 val = vol2inst.get((node, vol["vg"] + "/" + vol["name"]), "-")
5178 raise errors.ParameterError(field)
5179 node_output.append(str(val))
5181 output.append(node_output)
5186 class LUNodeQueryStorage(NoHooksLU):
5187 """Logical unit for getting information on storage units on node(s).
5190 _FIELDS_STATIC = utils.FieldSet(constants.SF_NODE)
5193 def CheckArguments(self):
5194 _CheckOutputFields(static=self._FIELDS_STATIC,
5195 dynamic=utils.FieldSet(*constants.VALID_STORAGE_FIELDS),
5196 selected=self.op.output_fields)
5198 def ExpandNames(self):
5199 self.share_locks = _ShareAll()
5200 self.needed_locks = {}
5203 self.needed_locks[locking.LEVEL_NODE] = \
5204 _GetWantedNodes(self, self.op.nodes)
5206 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
5208 def Exec(self, feedback_fn):
5209 """Computes the list of nodes and their attributes.
5212 self.nodes = self.owned_locks(locking.LEVEL_NODE)
5214 # Always get name to sort by
5215 if constants.SF_NAME in self.op.output_fields:
5216 fields = self.op.output_fields[:]
5218 fields = [constants.SF_NAME] + self.op.output_fields
5220 # Never ask for node or type as it's only known to the LU
5221 for extra in [constants.SF_NODE, constants.SF_TYPE]:
5222 while extra in fields:
5223 fields.remove(extra)
5225 field_idx = dict([(name, idx) for (idx, name) in enumerate(fields)])
5226 name_idx = field_idx[constants.SF_NAME]
5228 st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
5229 data = self.rpc.call_storage_list(self.nodes,
5230 self.op.storage_type, st_args,
5231 self.op.name, fields)
5235 for node in utils.NiceSort(self.nodes):
5236 nresult = data[node]
5240 msg = nresult.fail_msg
5242 self.LogWarning("Can't get storage data from node %s: %s", node, msg)
5245 rows = dict([(row[name_idx], row) for row in nresult.payload])
5247 for name in utils.NiceSort(rows.keys()):
5252 for field in self.op.output_fields:
5253 if field == constants.SF_NODE:
5255 elif field == constants.SF_TYPE:
5256 val = self.op.storage_type
5257 elif field in field_idx:
5258 val = row[field_idx[field]]
5260 raise errors.ParameterError(field)
5269 class _InstanceQuery(_QueryBase):
5270 FIELDS = query.INSTANCE_FIELDS
5272 def ExpandNames(self, lu):
5273 lu.needed_locks = {}
5274 lu.share_locks = _ShareAll()
5277 self.wanted = _GetWantedInstances(lu, self.names)
5279 self.wanted = locking.ALL_SET
5281 self.do_locking = (self.use_locking and
5282 query.IQ_LIVE in self.requested_data)
5284 lu.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
5285 lu.needed_locks[locking.LEVEL_NODEGROUP] = []
5286 lu.needed_locks[locking.LEVEL_NODE] = []
5287 lu.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5289 self.do_grouplocks = (self.do_locking and
5290 query.IQ_NODES in self.requested_data)
5292 def DeclareLocks(self, lu, level):
5294 if level == locking.LEVEL_NODEGROUP and self.do_grouplocks:
5295 assert not lu.needed_locks[locking.LEVEL_NODEGROUP]
5297 # Lock all groups used by instances optimistically; this requires going
5298 # via the node before it's locked, requiring verification later on
5299 lu.needed_locks[locking.LEVEL_NODEGROUP] = \
5301 for instance_name in lu.owned_locks(locking.LEVEL_INSTANCE)
5302 for group_uuid in lu.cfg.GetInstanceNodeGroups(instance_name))
5303 elif level == locking.LEVEL_NODE:
5304 lu._LockInstancesNodes() # pylint: disable=W0212
5307 def _CheckGroupLocks(lu):
5308 owned_instances = frozenset(lu.owned_locks(locking.LEVEL_INSTANCE))
5309 owned_groups = frozenset(lu.owned_locks(locking.LEVEL_NODEGROUP))
5311 # Check if node groups for locked instances are still correct
5312 for instance_name in owned_instances:
5313 _CheckInstanceNodeGroups(lu.cfg, instance_name, owned_groups)
5315 def _GetQueryData(self, lu):
5316 """Computes the list of instances and their attributes.
5319 if self.do_grouplocks:
5320 self._CheckGroupLocks(lu)
5322 cluster = lu.cfg.GetClusterInfo()
5323 all_info = lu.cfg.GetAllInstancesInfo()
5325 instance_names = self._GetNames(lu, all_info.keys(), locking.LEVEL_INSTANCE)
5327 instance_list = [all_info[name] for name in instance_names]
5328 nodes = frozenset(itertools.chain(*(inst.all_nodes
5329 for inst in instance_list)))
5330 hv_list = list(set([inst.hypervisor for inst in instance_list]))
5333 wrongnode_inst = set()
5335 # Gather data as requested
5336 if self.requested_data & set([query.IQ_LIVE, query.IQ_CONSOLE]):
5338 node_data = lu.rpc.call_all_instances_info(nodes, hv_list)
5340 result = node_data[name]
5342 # offline nodes will be in both lists
5343 assert result.fail_msg
5344 offline_nodes.append(name)
5346 bad_nodes.append(name)
5347 elif result.payload:
5348 for inst in result.payload:
5349 if inst in all_info:
5350 if all_info[inst].primary_node == name:
5351 live_data.update(result.payload)
5353 wrongnode_inst.add(inst)
5355 # orphan instance; we don't list it here as we don't
5356 # handle this case yet in the output of instance listing
5357 logging.warning("Orphan instance '%s' found on node %s",
5359 # else no instance is alive
5363 if query.IQ_DISKUSAGE in self.requested_data:
5364 disk_usage = dict((inst.name,
5365 _ComputeDiskSize(inst.disk_template,
5366 [{constants.IDISK_SIZE: disk.size}
5367 for disk in inst.disks]))
5368 for inst in instance_list)
5372 if query.IQ_CONSOLE in self.requested_data:
5374 for inst in instance_list:
5375 if inst.name in live_data:
5376 # Instance is running
5377 consinfo[inst.name] = _GetInstanceConsole(cluster, inst)
5379 consinfo[inst.name] = None
5380 assert set(consinfo.keys()) == set(instance_names)
5384 if query.IQ_NODES in self.requested_data:
5385 node_names = set(itertools.chain(*map(operator.attrgetter("all_nodes"),
5387 nodes = dict(lu.cfg.GetMultiNodeInfo(node_names))
5388 groups = dict((uuid, lu.cfg.GetNodeGroup(uuid))
5389 for uuid in set(map(operator.attrgetter("group"),
5395 return query.InstanceQueryData(instance_list, lu.cfg.GetClusterInfo(),
5396 disk_usage, offline_nodes, bad_nodes,
5397 live_data, wrongnode_inst, consinfo,
5401 class LUQuery(NoHooksLU):
5402 """Query for resources/items of a certain kind.
5405 # pylint: disable=W0142
5408 def CheckArguments(self):
5409 qcls = _GetQueryImplementation(self.op.what)
5411 self.impl = qcls(self.op.qfilter, self.op.fields, self.op.use_locking)
5413 def ExpandNames(self):
5414 self.impl.ExpandNames(self)
5416 def DeclareLocks(self, level):
5417 self.impl.DeclareLocks(self, level)
5419 def Exec(self, feedback_fn):
5420 return self.impl.NewStyleQuery(self)
5423 class LUQueryFields(NoHooksLU):
5424 """Query for resources/items of a certain kind.
5427 # pylint: disable=W0142
5430 def CheckArguments(self):
5431 self.qcls = _GetQueryImplementation(self.op.what)
5433 def ExpandNames(self):
5434 self.needed_locks = {}
5436 def Exec(self, feedback_fn):
5437 return query.QueryFields(self.qcls.FIELDS, self.op.fields)
5440 class LUNodeModifyStorage(NoHooksLU):
5441 """Logical unit for modifying a storage volume on a node.
5446 def CheckArguments(self):
5447 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
5449 storage_type = self.op.storage_type
5452 modifiable = constants.MODIFIABLE_STORAGE_FIELDS[storage_type]
5454 raise errors.OpPrereqError("Storage units of type '%s' can not be"
5455 " modified" % storage_type,
5458 diff = set(self.op.changes.keys()) - modifiable
5460 raise errors.OpPrereqError("The following fields can not be modified for"
5461 " storage units of type '%s': %r" %
5462 (storage_type, list(diff)),
5465 def ExpandNames(self):
5466 self.needed_locks = {
5467 locking.LEVEL_NODE: self.op.node_name,
5470 def Exec(self, feedback_fn):
5471 """Computes the list of nodes and their attributes.
5474 st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
5475 result = self.rpc.call_storage_modify(self.op.node_name,
5476 self.op.storage_type, st_args,
5477 self.op.name, self.op.changes)
5478 result.Raise("Failed to modify storage unit '%s' on %s" %
5479 (self.op.name, self.op.node_name))
5482 class LUNodeAdd(LogicalUnit):
5483 """Logical unit for adding node to the cluster.
5487 HTYPE = constants.HTYPE_NODE
5488 _NFLAGS = ["master_capable", "vm_capable"]
5490 def CheckArguments(self):
5491 self.primary_ip_family = self.cfg.GetPrimaryIPFamily()
5492 # validate/normalize the node name
5493 self.hostname = netutils.GetHostname(name=self.op.node_name,
5494 family=self.primary_ip_family)
5495 self.op.node_name = self.hostname.name
5497 if self.op.readd and self.op.node_name == self.cfg.GetMasterNode():
5498 raise errors.OpPrereqError("Cannot readd the master node",
5501 if self.op.readd and self.op.group:
5502 raise errors.OpPrereqError("Cannot pass a node group when a node is"
5503 " being readded", errors.ECODE_INVAL)
5505 def BuildHooksEnv(self):
5508 This will run on all nodes before, and on all nodes + the new node after.
5512 "OP_TARGET": self.op.node_name,
5513 "NODE_NAME": self.op.node_name,
5514 "NODE_PIP": self.op.primary_ip,
5515 "NODE_SIP": self.op.secondary_ip,
5516 "MASTER_CAPABLE": str(self.op.master_capable),
5517 "VM_CAPABLE": str(self.op.vm_capable),
5520 def BuildHooksNodes(self):
5521 """Build hooks nodes.
5524 # Exclude added node
5525 pre_nodes = list(set(self.cfg.GetNodeList()) - set([self.op.node_name]))
5526 post_nodes = pre_nodes + [self.op.node_name, ]
5528 return (pre_nodes, post_nodes)
5530 def CheckPrereq(self):
5531 """Check prerequisites.
5534 - the new node is not already in the config
5536 - its parameters (single/dual homed) matches the cluster
5538 Any errors are signaled by raising errors.OpPrereqError.
5542 hostname = self.hostname
5543 node = hostname.name
5544 primary_ip = self.op.primary_ip = hostname.ip
5545 if self.op.secondary_ip is None:
5546 if self.primary_ip_family == netutils.IP6Address.family:
5547 raise errors.OpPrereqError("When using a IPv6 primary address, a valid"
5548 " IPv4 address must be given as secondary",
5550 self.op.secondary_ip = primary_ip
5552 secondary_ip = self.op.secondary_ip
5553 if not netutils.IP4Address.IsValid(secondary_ip):
5554 raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
5555 " address" % secondary_ip, errors.ECODE_INVAL)
5557 node_list = cfg.GetNodeList()
5558 if not self.op.readd and node in node_list:
5559 raise errors.OpPrereqError("Node %s is already in the configuration" %
5560 node, errors.ECODE_EXISTS)
5561 elif self.op.readd and node not in node_list:
5562 raise errors.OpPrereqError("Node %s is not in the configuration" % node,
5565 self.changed_primary_ip = False
5567 for existing_node_name, existing_node in cfg.GetMultiNodeInfo(node_list):
5568 if self.op.readd and node == existing_node_name:
5569 if existing_node.secondary_ip != secondary_ip:
5570 raise errors.OpPrereqError("Readded node doesn't have the same IP"
5571 " address configuration as before",
5573 if existing_node.primary_ip != primary_ip:
5574 self.changed_primary_ip = True
5578 if (existing_node.primary_ip == primary_ip or
5579 existing_node.secondary_ip == primary_ip or
5580 existing_node.primary_ip == secondary_ip or
5581 existing_node.secondary_ip == secondary_ip):
5582 raise errors.OpPrereqError("New node ip address(es) conflict with"
5583 " existing node %s" % existing_node.name,
5584 errors.ECODE_NOTUNIQUE)
5586 # After this 'if' block, None is no longer a valid value for the
5587 # _capable op attributes
5589 old_node = self.cfg.GetNodeInfo(node)
5590 assert old_node is not None, "Can't retrieve locked node %s" % node
5591 for attr in self._NFLAGS:
5592 if getattr(self.op, attr) is None:
5593 setattr(self.op, attr, getattr(old_node, attr))
5595 for attr in self._NFLAGS:
5596 if getattr(self.op, attr) is None:
5597 setattr(self.op, attr, True)
5599 if self.op.readd and not self.op.vm_capable:
5600 pri, sec = cfg.GetNodeInstances(node)
5602 raise errors.OpPrereqError("Node %s being re-added with vm_capable"
5603 " flag set to false, but it already holds"
5604 " instances" % node,
5607 # check that the type of the node (single versus dual homed) is the
5608 # same as for the master
5609 myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
5610 master_singlehomed = myself.secondary_ip == myself.primary_ip
5611 newbie_singlehomed = secondary_ip == primary_ip
5612 if master_singlehomed != newbie_singlehomed:
5613 if master_singlehomed:
5614 raise errors.OpPrereqError("The master has no secondary ip but the"
5615 " new node has one",
5618 raise errors.OpPrereqError("The master has a secondary ip but the"
5619 " new node doesn't have one",
5622 # checks reachability
5623 if not netutils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
5624 raise errors.OpPrereqError("Node not reachable by ping",
5625 errors.ECODE_ENVIRON)
5627 if not newbie_singlehomed:
5628 # check reachability from my secondary ip to newbie's secondary ip
5629 if not netutils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
5630 source=myself.secondary_ip):
5631 raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
5632 " based ping to node daemon port",
5633 errors.ECODE_ENVIRON)
5640 if self.op.master_capable:
5641 self.master_candidate = _DecideSelfPromotion(self, exceptions=exceptions)
5643 self.master_candidate = False
5646 self.new_node = old_node
5648 node_group = cfg.LookupNodeGroup(self.op.group)
5649 self.new_node = objects.Node(name=node,
5650 primary_ip=primary_ip,
5651 secondary_ip=secondary_ip,
5652 master_candidate=self.master_candidate,
5653 offline=False, drained=False,
5656 if self.op.ndparams:
5657 utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
5659 if self.op.hv_state:
5660 self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state, None)
5662 if self.op.disk_state:
5663 self.new_disk_state = _MergeAndVerifyDiskState(self.op.disk_state, None)
5665 # TODO: If we need to have multiple DnsOnlyRunner we probably should make
5666 # it a property on the base class.
5667 result = rpc.DnsOnlyRunner().call_version([node])[node]
5668 result.Raise("Can't get version information from node %s" % node)
5669 if constants.PROTOCOL_VERSION == result.payload:
5670 logging.info("Communication to node %s fine, sw version %s match",
5671 node, result.payload)
5673 raise errors.OpPrereqError("Version mismatch master version %s,"
5674 " node version %s" %
5675 (constants.PROTOCOL_VERSION, result.payload),
5676 errors.ECODE_ENVIRON)
5678 def Exec(self, feedback_fn):
5679 """Adds the new node to the cluster.
5682 new_node = self.new_node
5683 node = new_node.name
5685 assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER), \
5688 # We adding a new node so we assume it's powered
5689 new_node.powered = True
5691 # for re-adds, reset the offline/drained/master-candidate flags;
5692 # we need to reset here, otherwise offline would prevent RPC calls
5693 # later in the procedure; this also means that if the re-add
5694 # fails, we are left with a non-offlined, broken node
5696 new_node.drained = new_node.offline = False # pylint: disable=W0201
5697 self.LogInfo("Readding a node, the offline/drained flags were reset")
5698 # if we demote the node, we do cleanup later in the procedure
5699 new_node.master_candidate = self.master_candidate
5700 if self.changed_primary_ip:
5701 new_node.primary_ip = self.op.primary_ip
5703 # copy the master/vm_capable flags
5704 for attr in self._NFLAGS:
5705 setattr(new_node, attr, getattr(self.op, attr))
5707 # notify the user about any possible mc promotion
5708 if new_node.master_candidate:
5709 self.LogInfo("Node will be a master candidate")
5711 if self.op.ndparams:
5712 new_node.ndparams = self.op.ndparams
5714 new_node.ndparams = {}
5716 if self.op.hv_state:
5717 new_node.hv_state_static = self.new_hv_state
5719 if self.op.disk_state:
5720 new_node.disk_state_static = self.new_disk_state
5722 # Add node to our /etc/hosts, and add key to known_hosts
5723 if self.cfg.GetClusterInfo().modify_etc_hosts:
5724 master_node = self.cfg.GetMasterNode()
5725 result = self.rpc.call_etc_hosts_modify(master_node,
5726 constants.ETC_HOSTS_ADD,
5729 result.Raise("Can't update hosts file with new host data")
5731 if new_node.secondary_ip != new_node.primary_ip:
5732 _CheckNodeHasSecondaryIP(self, new_node.name, new_node.secondary_ip,
5735 node_verify_list = [self.cfg.GetMasterNode()]
5736 node_verify_param = {
5737 constants.NV_NODELIST: ([node], {}),
5738 # TODO: do a node-net-test as well?
5741 result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
5742 self.cfg.GetClusterName())
5743 for verifier in node_verify_list:
5744 result[verifier].Raise("Cannot communicate with node %s" % verifier)
5745 nl_payload = result[verifier].payload[constants.NV_NODELIST]
5747 for failed in nl_payload:
5748 feedback_fn("ssh/hostname verification failed"
5749 " (checking from %s): %s" %
5750 (verifier, nl_payload[failed]))
5751 raise errors.OpExecError("ssh/hostname verification failed")
5754 _RedistributeAncillaryFiles(self)
5755 self.context.ReaddNode(new_node)
5756 # make sure we redistribute the config
5757 self.cfg.Update(new_node, feedback_fn)
5758 # and make sure the new node will not have old files around
5759 if not new_node.master_candidate:
5760 result = self.rpc.call_node_demote_from_mc(new_node.name)
5761 msg = result.fail_msg
5763 self.LogWarning("Node failed to demote itself from master"
5764 " candidate status: %s" % msg)
5766 _RedistributeAncillaryFiles(self, additional_nodes=[node],
5767 additional_vm=self.op.vm_capable)
5768 self.context.AddNode(new_node, self.proc.GetECId())
5771 class LUNodeSetParams(LogicalUnit):
5772 """Modifies the parameters of a node.
5774 @cvar _F2R: a dictionary from tuples of flags (mc, drained, offline)
5775 to the node role (as _ROLE_*)
5776 @cvar _R2F: a dictionary from node role to tuples of flags
5777 @cvar _FLAGS: a list of attribute names corresponding to the flags
5780 HPATH = "node-modify"
5781 HTYPE = constants.HTYPE_NODE
5783 (_ROLE_CANDIDATE, _ROLE_DRAINED, _ROLE_OFFLINE, _ROLE_REGULAR) = range(4)
5785 (True, False, False): _ROLE_CANDIDATE,
5786 (False, True, False): _ROLE_DRAINED,
5787 (False, False, True): _ROLE_OFFLINE,
5788 (False, False, False): _ROLE_REGULAR,
5790 _R2F = dict((v, k) for k, v in _F2R.items())
5791 _FLAGS = ["master_candidate", "drained", "offline"]
5793 def CheckArguments(self):
5794 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
5795 all_mods = [self.op.offline, self.op.master_candidate, self.op.drained,
5796 self.op.master_capable, self.op.vm_capable,
5797 self.op.secondary_ip, self.op.ndparams, self.op.hv_state,
5799 if all_mods.count(None) == len(all_mods):
5800 raise errors.OpPrereqError("Please pass at least one modification",
5802 if all_mods.count(True) > 1:
5803 raise errors.OpPrereqError("Can't set the node into more than one"
5804 " state at the same time",
5807 # Boolean value that tells us whether we might be demoting from MC
5808 self.might_demote = (self.op.master_candidate == False or
5809 self.op.offline == True or
5810 self.op.drained == True or
5811 self.op.master_capable == False)
5813 if self.op.secondary_ip:
5814 if not netutils.IP4Address.IsValid(self.op.secondary_ip):
5815 raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
5816 " address" % self.op.secondary_ip,
5819 self.lock_all = self.op.auto_promote and self.might_demote
5820 self.lock_instances = self.op.secondary_ip is not None
5822 def _InstanceFilter(self, instance):
5823 """Filter for getting affected instances.
5826 return (instance.disk_template in constants.DTS_INT_MIRROR and
5827 self.op.node_name in instance.all_nodes)
5829 def ExpandNames(self):
5831 self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET}
5833 self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
5835 # Since modifying a node can have severe effects on currently running
5836 # operations the resource lock is at least acquired in shared mode
5837 self.needed_locks[locking.LEVEL_NODE_RES] = \
5838 self.needed_locks[locking.LEVEL_NODE]
5840 # Get node resource and instance locks in shared mode; they are not used
5841 # for anything but read-only access
5842 self.share_locks[locking.LEVEL_NODE_RES] = 1
5843 self.share_locks[locking.LEVEL_INSTANCE] = 1
5845 if self.lock_instances:
5846 self.needed_locks[locking.LEVEL_INSTANCE] = \
5847 frozenset(self.cfg.GetInstancesInfoByFilter(self._InstanceFilter))
5849 def BuildHooksEnv(self):
5852 This runs on the master node.
5856 "OP_TARGET": self.op.node_name,
5857 "MASTER_CANDIDATE": str(self.op.master_candidate),
5858 "OFFLINE": str(self.op.offline),
5859 "DRAINED": str(self.op.drained),
5860 "MASTER_CAPABLE": str(self.op.master_capable),
5861 "VM_CAPABLE": str(self.op.vm_capable),
5864 def BuildHooksNodes(self):
5865 """Build hooks nodes.
5868 nl = [self.cfg.GetMasterNode(), self.op.node_name]
5871 def CheckPrereq(self):
5872 """Check prerequisites.
5874 This only checks the instance list against the existing names.
5877 node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
5879 if self.lock_instances:
5880 affected_instances = \
5881 self.cfg.GetInstancesInfoByFilter(self._InstanceFilter)
5883 # Verify instance locks
5884 owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
5885 wanted_instances = frozenset(affected_instances.keys())
5886 if wanted_instances - owned_instances:
5887 raise errors.OpPrereqError("Instances affected by changing node %s's"
5888 " secondary IP address have changed since"
5889 " locks were acquired, wanted '%s', have"
5890 " '%s'; retry the operation" %
5892 utils.CommaJoin(wanted_instances),
5893 utils.CommaJoin(owned_instances)),
5896 affected_instances = None
5898 if (self.op.master_candidate is not None or
5899 self.op.drained is not None or
5900 self.op.offline is not None):
5901 # we can't change the master's node flags
5902 if self.op.node_name == self.cfg.GetMasterNode():
5903 raise errors.OpPrereqError("The master role can be changed"
5904 " only via master-failover",
5907 if self.op.master_candidate and not node.master_capable:
5908 raise errors.OpPrereqError("Node %s is not master capable, cannot make"
5909 " it a master candidate" % node.name,
5912 if self.op.vm_capable == False:
5913 (ipri, isec) = self.cfg.GetNodeInstances(self.op.node_name)
5915 raise errors.OpPrereqError("Node %s hosts instances, cannot unset"
5916 " the vm_capable flag" % node.name,
5919 if node.master_candidate and self.might_demote and not self.lock_all:
5920 assert not self.op.auto_promote, "auto_promote set but lock_all not"
5921 # check if after removing the current node, we're missing master
5923 (mc_remaining, mc_should, _) = \
5924 self.cfg.GetMasterCandidateStats(exceptions=[node.name])
5925 if mc_remaining < mc_should:
5926 raise errors.OpPrereqError("Not enough master candidates, please"
5927 " pass auto promote option to allow"
5928 " promotion (--auto-promote or RAPI"
5929 " auto_promote=True)", errors.ECODE_STATE)
5931 self.old_flags = old_flags = (node.master_candidate,
5932 node.drained, node.offline)
5933 assert old_flags in self._F2R, "Un-handled old flags %s" % str(old_flags)
5934 self.old_role = old_role = self._F2R[old_flags]
5936 # Check for ineffective changes
5937 for attr in self._FLAGS:
5938 if (getattr(self.op, attr) == False and getattr(node, attr) == False):
5939 self.LogInfo("Ignoring request to unset flag %s, already unset", attr)
5940 setattr(self.op, attr, None)
5942 # Past this point, any flag change to False means a transition
5943 # away from the respective state, as only real changes are kept
5945 # TODO: We might query the real power state if it supports OOB
5946 if _SupportsOob(self.cfg, node):
5947 if self.op.offline is False and not (node.powered or
5948 self.op.powered == True):
5949 raise errors.OpPrereqError(("Node %s needs to be turned on before its"
5950 " offline status can be reset") %
5952 elif self.op.powered is not None:
5953 raise errors.OpPrereqError(("Unable to change powered state for node %s"
5954 " as it does not support out-of-band"
5955 " handling") % self.op.node_name)
5957 # If we're being deofflined/drained, we'll MC ourself if needed
5958 if (self.op.drained == False or self.op.offline == False or
5959 (self.op.master_capable and not node.master_capable)):
5960 if _DecideSelfPromotion(self):
5961 self.op.master_candidate = True
5962 self.LogInfo("Auto-promoting node to master candidate")
5964 # If we're no longer master capable, we'll demote ourselves from MC
5965 if self.op.master_capable == False and node.master_candidate:
5966 self.LogInfo("Demoting from master candidate")
5967 self.op.master_candidate = False
5970 assert [getattr(self.op, attr) for attr in self._FLAGS].count(True) <= 1
5971 if self.op.master_candidate:
5972 new_role = self._ROLE_CANDIDATE
5973 elif self.op.drained:
5974 new_role = self._ROLE_DRAINED
5975 elif self.op.offline:
5976 new_role = self._ROLE_OFFLINE
5977 elif False in [self.op.master_candidate, self.op.drained, self.op.offline]:
5978 # False is still in new flags, which means we're un-setting (the
5980 new_role = self._ROLE_REGULAR
5981 else: # no new flags, nothing, keep old role
5984 self.new_role = new_role
5986 if old_role == self._ROLE_OFFLINE and new_role != old_role:
5987 # Trying to transition out of offline status
5988 result = self.rpc.call_version([node.name])[node.name]
5990 raise errors.OpPrereqError("Node %s is being de-offlined but fails"
5991 " to report its version: %s" %
5992 (node.name, result.fail_msg),
5995 self.LogWarning("Transitioning node from offline to online state"
5996 " without using re-add. Please make sure the node"
5999 if self.op.secondary_ip:
6000 # Ok even without locking, because this can't be changed by any LU
6001 master = self.cfg.GetNodeInfo(self.cfg.GetMasterNode())
6002 master_singlehomed = master.secondary_ip == master.primary_ip
6003 if master_singlehomed and self.op.secondary_ip:
6004 raise errors.OpPrereqError("Cannot change the secondary ip on a single"
6005 " homed cluster", errors.ECODE_INVAL)
6007 assert not (frozenset(affected_instances) -
6008 self.owned_locks(locking.LEVEL_INSTANCE))
6011 if affected_instances:
6012 raise errors.OpPrereqError("Cannot change secondary IP address:"
6013 " offline node has instances (%s)"
6014 " configured to use it" %
6015 utils.CommaJoin(affected_instances.keys()))
6017 # On online nodes, check that no instances are running, and that
6018 # the node has the new ip and we can reach it.
6019 for instance in affected_instances.values():
6020 _CheckInstanceState(self, instance, INSTANCE_DOWN,
6021 msg="cannot change secondary ip")
6023 _CheckNodeHasSecondaryIP(self, node.name, self.op.secondary_ip, True)
6024 if master.name != node.name:
6025 # check reachability from master secondary ip to new secondary ip
6026 if not netutils.TcpPing(self.op.secondary_ip,
6027 constants.DEFAULT_NODED_PORT,
6028 source=master.secondary_ip):
6029 raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
6030 " based ping to node daemon port",
6031 errors.ECODE_ENVIRON)
6033 if self.op.ndparams:
6034 new_ndparams = _GetUpdatedParams(self.node.ndparams, self.op.ndparams)
6035 utils.ForceDictType(new_ndparams, constants.NDS_PARAMETER_TYPES)
6036 self.new_ndparams = new_ndparams
6038 if self.op.hv_state:
6039 self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state,
6040 self.node.hv_state_static)
6042 if self.op.disk_state:
6043 self.new_disk_state = \
6044 _MergeAndVerifyDiskState(self.op.disk_state,
6045 self.node.disk_state_static)
6047 def Exec(self, feedback_fn):
6052 old_role = self.old_role
6053 new_role = self.new_role
6057 if self.op.ndparams:
6058 node.ndparams = self.new_ndparams
6060 if self.op.powered is not None:
6061 node.powered = self.op.powered
6063 if self.op.hv_state:
6064 node.hv_state_static = self.new_hv_state
6066 if self.op.disk_state:
6067 node.disk_state_static = self.new_disk_state
6069 for attr in ["master_capable", "vm_capable"]:
6070 val = getattr(self.op, attr)
6072 setattr(node, attr, val)
6073 result.append((attr, str(val)))
6075 if new_role != old_role:
6076 # Tell the node to demote itself, if no longer MC and not offline
6077 if old_role == self._ROLE_CANDIDATE and new_role != self._ROLE_OFFLINE:
6078 msg = self.rpc.call_node_demote_from_mc(node.name).fail_msg
6080 self.LogWarning("Node failed to demote itself: %s", msg)
6082 new_flags = self._R2F[new_role]
6083 for of, nf, desc in zip(self.old_flags, new_flags, self._FLAGS):
6085 result.append((desc, str(nf)))
6086 (node.master_candidate, node.drained, node.offline) = new_flags
6088 # we locked all nodes, we adjust the CP before updating this node
6090 _AdjustCandidatePool(self, [node.name])
6092 if self.op.secondary_ip:
6093 node.secondary_ip = self.op.secondary_ip
6094 result.append(("secondary_ip", self.op.secondary_ip))
6096 # this will trigger configuration file update, if needed
6097 self.cfg.Update(node, feedback_fn)
6099 # this will trigger job queue propagation or cleanup if the mc
6101 if [old_role, new_role].count(self._ROLE_CANDIDATE) == 1:
6102 self.context.ReaddNode(node)
6107 class LUNodePowercycle(NoHooksLU):
6108 """Powercycles a node.
6113 def CheckArguments(self):
6114 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
6115 if self.op.node_name == self.cfg.GetMasterNode() and not self.op.force:
6116 raise errors.OpPrereqError("The node is the master and the force"
6117 " parameter was not set",
6120 def ExpandNames(self):
6121 """Locking for PowercycleNode.
6123 This is a last-resort option and shouldn't block on other
6124 jobs. Therefore, we grab no locks.
6127 self.needed_locks = {}
6129 def Exec(self, feedback_fn):
6133 result = self.rpc.call_node_powercycle(self.op.node_name,
6134 self.cfg.GetHypervisorType())
6135 result.Raise("Failed to schedule the reboot")
6136 return result.payload
6139 class LUClusterQuery(NoHooksLU):
6140 """Query cluster configuration.
6145 def ExpandNames(self):
6146 self.needed_locks = {}
6148 def Exec(self, feedback_fn):
6149 """Return cluster config.
6152 cluster = self.cfg.GetClusterInfo()
6155 # Filter just for enabled hypervisors
6156 for os_name, hv_dict in cluster.os_hvp.items():
6157 os_hvp[os_name] = {}
6158 for hv_name, hv_params in hv_dict.items():
6159 if hv_name in cluster.enabled_hypervisors:
6160 os_hvp[os_name][hv_name] = hv_params
6162 # Convert ip_family to ip_version
6163 primary_ip_version = constants.IP4_VERSION
6164 if cluster.primary_ip_family == netutils.IP6Address.family:
6165 primary_ip_version = constants.IP6_VERSION
6168 "software_version": constants.RELEASE_VERSION,
6169 "protocol_version": constants.PROTOCOL_VERSION,
6170 "config_version": constants.CONFIG_VERSION,
6171 "os_api_version": max(constants.OS_API_VERSIONS),
6172 "export_version": constants.EXPORT_VERSION,
6173 "architecture": runtime.GetArchInfo(),
6174 "name": cluster.cluster_name,
6175 "master": cluster.master_node,
6176 "default_hypervisor": cluster.primary_hypervisor,
6177 "enabled_hypervisors": cluster.enabled_hypervisors,
6178 "hvparams": dict([(hypervisor_name, cluster.hvparams[hypervisor_name])
6179 for hypervisor_name in cluster.enabled_hypervisors]),
6181 "beparams": cluster.beparams,
6182 "osparams": cluster.osparams,
6183 "ipolicy": cluster.ipolicy,
6184 "nicparams": cluster.nicparams,
6185 "ndparams": cluster.ndparams,
6186 "diskparams": cluster.diskparams,
6187 "candidate_pool_size": cluster.candidate_pool_size,
6188 "master_netdev": cluster.master_netdev,
6189 "master_netmask": cluster.master_netmask,
6190 "use_external_mip_script": cluster.use_external_mip_script,
6191 "volume_group_name": cluster.volume_group_name,
6192 "drbd_usermode_helper": cluster.drbd_usermode_helper,
6193 "file_storage_dir": cluster.file_storage_dir,
6194 "shared_file_storage_dir": cluster.shared_file_storage_dir,
6195 "maintain_node_health": cluster.maintain_node_health,
6196 "ctime": cluster.ctime,
6197 "mtime": cluster.mtime,
6198 "uuid": cluster.uuid,
6199 "tags": list(cluster.GetTags()),
6200 "uid_pool": cluster.uid_pool,
6201 "default_iallocator": cluster.default_iallocator,
6202 "reserved_lvs": cluster.reserved_lvs,
6203 "primary_ip_version": primary_ip_version,
6204 "prealloc_wipe_disks": cluster.prealloc_wipe_disks,
6205 "hidden_os": cluster.hidden_os,
6206 "blacklisted_os": cluster.blacklisted_os,
6212 class LUClusterConfigQuery(NoHooksLU):
6213 """Return configuration values.
6218 def CheckArguments(self):
6219 self.cq = _ClusterQuery(None, self.op.output_fields, False)
6221 def ExpandNames(self):
6222 self.cq.ExpandNames(self)
6224 def DeclareLocks(self, level):
6225 self.cq.DeclareLocks(self, level)
6227 def Exec(self, feedback_fn):
6228 result = self.cq.OldStyleQuery(self)
6230 assert len(result) == 1
6235 class _ClusterQuery(_QueryBase):
6236 FIELDS = query.CLUSTER_FIELDS
6238 #: Do not sort (there is only one item)
6241 def ExpandNames(self, lu):
6242 lu.needed_locks = {}
6244 # The following variables interact with _QueryBase._GetNames
6245 self.wanted = locking.ALL_SET
6246 self.do_locking = self.use_locking
6249 raise errors.OpPrereqError("Can not use locking for cluster queries",
6252 def DeclareLocks(self, lu, level):
6255 def _GetQueryData(self, lu):
6256 """Computes the list of nodes and their attributes.
6259 # Locking is not used
6260 assert not (compat.any(lu.glm.is_owned(level)
6261 for level in locking.LEVELS
6262 if level != locking.LEVEL_CLUSTER) or
6263 self.do_locking or self.use_locking)
6265 if query.CQ_CONFIG in self.requested_data:
6266 cluster = lu.cfg.GetClusterInfo()
6268 cluster = NotImplemented
6270 if query.CQ_QUEUE_DRAINED in self.requested_data:
6271 drain_flag = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
6273 drain_flag = NotImplemented
6275 if query.CQ_WATCHER_PAUSE in self.requested_data:
6276 watcher_pause = utils.ReadWatcherPauseFile(constants.WATCHER_PAUSEFILE)
6278 watcher_pause = NotImplemented
6280 return query.ClusterQueryData(cluster, drain_flag, watcher_pause)
6283 class LUInstanceActivateDisks(NoHooksLU):
6284 """Bring up an instance's disks.
6289 def ExpandNames(self):
6290 self._ExpandAndLockInstance()
6291 self.needed_locks[locking.LEVEL_NODE] = []
6292 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6294 def DeclareLocks(self, level):
6295 if level == locking.LEVEL_NODE:
6296 self._LockInstancesNodes()
6298 def CheckPrereq(self):
6299 """Check prerequisites.
6301 This checks that the instance is in the cluster.
6304 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6305 assert self.instance is not None, \
6306 "Cannot retrieve locked instance %s" % self.op.instance_name
6307 _CheckNodeOnline(self, self.instance.primary_node)
6309 def Exec(self, feedback_fn):
6310 """Activate the disks.
6313 disks_ok, disks_info = \
6314 _AssembleInstanceDisks(self, self.instance,
6315 ignore_size=self.op.ignore_size)
6317 raise errors.OpExecError("Cannot activate block devices")
6322 def _AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False,
6324 """Prepare the block devices for an instance.
6326 This sets up the block devices on all nodes.
6328 @type lu: L{LogicalUnit}
6329 @param lu: the logical unit on whose behalf we execute
6330 @type instance: L{objects.Instance}
6331 @param instance: the instance for whose disks we assemble
6332 @type disks: list of L{objects.Disk} or None
6333 @param disks: which disks to assemble (or all, if None)
6334 @type ignore_secondaries: boolean
6335 @param ignore_secondaries: if true, errors on secondary nodes
6336 won't result in an error return from the function
6337 @type ignore_size: boolean
6338 @param ignore_size: if true, the current known size of the disk
6339 will not be used during the disk activation, useful for cases
6340 when the size is wrong
6341 @return: False if the operation failed, otherwise a list of
6342 (host, instance_visible_name, node_visible_name)
6343 with the mapping from node devices to instance devices
6348 iname = instance.name
6349 disks = _ExpandCheckDisks(instance, disks)
6351 # With the two passes mechanism we try to reduce the window of
6352 # opportunity for the race condition of switching DRBD to primary
6353 # before handshaking occured, but we do not eliminate it
6355 # The proper fix would be to wait (with some limits) until the
6356 # connection has been made and drbd transitions from WFConnection
6357 # into any other network-connected state (Connected, SyncTarget,
6360 # 1st pass, assemble on all nodes in secondary mode
6361 for idx, inst_disk in enumerate(disks):
6362 for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
6364 node_disk = node_disk.Copy()
6365 node_disk.UnsetSize()
6366 lu.cfg.SetDiskID(node_disk, node)
6367 result = lu.rpc.call_blockdev_assemble(node, (node_disk, instance), iname,
6369 msg = result.fail_msg
6371 lu.proc.LogWarning("Could not prepare block device %s on node %s"
6372 " (is_primary=False, pass=1): %s",
6373 inst_disk.iv_name, node, msg)
6374 if not ignore_secondaries:
6377 # FIXME: race condition on drbd migration to primary
6379 # 2nd pass, do only the primary node
6380 for idx, inst_disk in enumerate(disks):
6383 for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
6384 if node != instance.primary_node:
6387 node_disk = node_disk.Copy()
6388 node_disk.UnsetSize()
6389 lu.cfg.SetDiskID(node_disk, node)
6390 result = lu.rpc.call_blockdev_assemble(node, (node_disk, instance), iname,
6392 msg = result.fail_msg
6394 lu.proc.LogWarning("Could not prepare block device %s on node %s"
6395 " (is_primary=True, pass=2): %s",
6396 inst_disk.iv_name, node, msg)
6399 dev_path = result.payload
6401 device_info.append((instance.primary_node, inst_disk.iv_name, dev_path))
6403 # leave the disks configured for the primary node
6404 # this is a workaround that would be fixed better by
6405 # improving the logical/physical id handling
6407 lu.cfg.SetDiskID(disk, instance.primary_node)
6409 return disks_ok, device_info
6412 def _StartInstanceDisks(lu, instance, force):
6413 """Start the disks of an instance.
6416 disks_ok, _ = _AssembleInstanceDisks(lu, instance,
6417 ignore_secondaries=force)
6419 _ShutdownInstanceDisks(lu, instance)
6420 if force is not None and not force:
6421 lu.proc.LogWarning("", hint="If the message above refers to a"
6423 " you can retry the operation using '--force'.")
6424 raise errors.OpExecError("Disk consistency error")
6427 class LUInstanceDeactivateDisks(NoHooksLU):
6428 """Shutdown an instance's disks.
6433 def ExpandNames(self):
6434 self._ExpandAndLockInstance()
6435 self.needed_locks[locking.LEVEL_NODE] = []
6436 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6438 def DeclareLocks(self, level):
6439 if level == locking.LEVEL_NODE:
6440 self._LockInstancesNodes()
6442 def CheckPrereq(self):
6443 """Check prerequisites.
6445 This checks that the instance is in the cluster.
6448 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6449 assert self.instance is not None, \
6450 "Cannot retrieve locked instance %s" % self.op.instance_name
6452 def Exec(self, feedback_fn):
6453 """Deactivate the disks
6456 instance = self.instance
6458 _ShutdownInstanceDisks(self, instance)
6460 _SafeShutdownInstanceDisks(self, instance)
6463 def _SafeShutdownInstanceDisks(lu, instance, disks=None):
6464 """Shutdown block devices of an instance.
6466 This function checks if an instance is running, before calling
6467 _ShutdownInstanceDisks.
6470 _CheckInstanceState(lu, instance, INSTANCE_DOWN, msg="cannot shutdown disks")
6471 _ShutdownInstanceDisks(lu, instance, disks=disks)
6474 def _ExpandCheckDisks(instance, disks):
6475 """Return the instance disks selected by the disks list
6477 @type disks: list of L{objects.Disk} or None
6478 @param disks: selected disks
6479 @rtype: list of L{objects.Disk}
6480 @return: selected instance disks to act on
6484 return instance.disks
6486 if not set(disks).issubset(instance.disks):
6487 raise errors.ProgrammerError("Can only act on disks belonging to the"
6492 def _ShutdownInstanceDisks(lu, instance, disks=None, ignore_primary=False):
6493 """Shutdown block devices of an instance.
6495 This does the shutdown on all nodes of the instance.
6497 If the ignore_primary is false, errors on the primary node are
6502 disks = _ExpandCheckDisks(instance, disks)
6505 for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
6506 lu.cfg.SetDiskID(top_disk, node)
6507 result = lu.rpc.call_blockdev_shutdown(node, top_disk)
6508 msg = result.fail_msg
6510 lu.LogWarning("Could not shutdown block device %s on node %s: %s",
6511 disk.iv_name, node, msg)
6512 if ((node == instance.primary_node and not ignore_primary) or
6513 (node != instance.primary_node and not result.offline)):
6518 def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
6519 """Checks if a node has enough free memory.
6521 This function check if a given node has the needed amount of free
6522 memory. In case the node has less memory or we cannot get the
6523 information from the node, this function raise an OpPrereqError
6526 @type lu: C{LogicalUnit}
6527 @param lu: a logical unit from which we get configuration data
6529 @param node: the node to check
6530 @type reason: C{str}
6531 @param reason: string to use in the error message
6532 @type requested: C{int}
6533 @param requested: the amount of memory in MiB to check for
6534 @type hypervisor_name: C{str}
6535 @param hypervisor_name: the hypervisor to ask for memory stats
6537 @return: node current free memory
6538 @raise errors.OpPrereqError: if the node doesn't have enough memory, or
6539 we cannot check the node
6542 nodeinfo = lu.rpc.call_node_info([node], None, [hypervisor_name])
6543 nodeinfo[node].Raise("Can't get data from node %s" % node,
6544 prereq=True, ecode=errors.ECODE_ENVIRON)
6545 (_, _, (hv_info, )) = nodeinfo[node].payload
6547 free_mem = hv_info.get("memory_free", None)
6548 if not isinstance(free_mem, int):
6549 raise errors.OpPrereqError("Can't compute free memory on node %s, result"
6550 " was '%s'" % (node, free_mem),
6551 errors.ECODE_ENVIRON)
6552 if requested > free_mem:
6553 raise errors.OpPrereqError("Not enough memory on node %s for %s:"
6554 " needed %s MiB, available %s MiB" %
6555 (node, reason, requested, free_mem),
6560 def _CheckNodesFreeDiskPerVG(lu, nodenames, req_sizes):
6561 """Checks if nodes have enough free disk space in the all VGs.
6563 This function check if all given nodes have the needed amount of
6564 free disk. In case any node has less disk or we cannot get the
6565 information from the node, this function raise an OpPrereqError
6568 @type lu: C{LogicalUnit}
6569 @param lu: a logical unit from which we get configuration data
6570 @type nodenames: C{list}
6571 @param nodenames: the list of node names to check
6572 @type req_sizes: C{dict}
6573 @param req_sizes: the hash of vg and corresponding amount of disk in
6575 @raise errors.OpPrereqError: if the node doesn't have enough disk,
6576 or we cannot check the node
6579 for vg, req_size in req_sizes.items():
6580 _CheckNodesFreeDiskOnVG(lu, nodenames, vg, req_size)
6583 def _CheckNodesFreeDiskOnVG(lu, nodenames, vg, requested):
6584 """Checks if nodes have enough free disk space in the specified VG.
6586 This function check if all given nodes have the needed amount of
6587 free disk. In case any node has less disk or we cannot get the
6588 information from the node, this function raise an OpPrereqError
6591 @type lu: C{LogicalUnit}
6592 @param lu: a logical unit from which we get configuration data
6593 @type nodenames: C{list}
6594 @param nodenames: the list of node names to check
6596 @param vg: the volume group to check
6597 @type requested: C{int}
6598 @param requested: the amount of disk in MiB to check for
6599 @raise errors.OpPrereqError: if the node doesn't have enough disk,
6600 or we cannot check the node
6603 nodeinfo = lu.rpc.call_node_info(nodenames, [vg], None)
6604 for node in nodenames:
6605 info = nodeinfo[node]
6606 info.Raise("Cannot get current information from node %s" % node,
6607 prereq=True, ecode=errors.ECODE_ENVIRON)
6608 (_, (vg_info, ), _) = info.payload
6609 vg_free = vg_info.get("vg_free", None)
6610 if not isinstance(vg_free, int):
6611 raise errors.OpPrereqError("Can't compute free disk space on node"
6612 " %s for vg %s, result was '%s'" %
6613 (node, vg, vg_free), errors.ECODE_ENVIRON)
6614 if requested > vg_free:
6615 raise errors.OpPrereqError("Not enough disk space on target node %s"
6616 " vg %s: required %d MiB, available %d MiB" %
6617 (node, vg, requested, vg_free),
6621 def _CheckNodesPhysicalCPUs(lu, nodenames, requested, hypervisor_name):
6622 """Checks if nodes have enough physical CPUs
6624 This function checks if all given nodes have the needed number of
6625 physical CPUs. In case any node has less CPUs or we cannot get the
6626 information from the node, this function raises an OpPrereqError
6629 @type lu: C{LogicalUnit}
6630 @param lu: a logical unit from which we get configuration data
6631 @type nodenames: C{list}
6632 @param nodenames: the list of node names to check
6633 @type requested: C{int}
6634 @param requested: the minimum acceptable number of physical CPUs
6635 @raise errors.OpPrereqError: if the node doesn't have enough CPUs,
6636 or we cannot check the node
6639 nodeinfo = lu.rpc.call_node_info(nodenames, None, [hypervisor_name])
6640 for node in nodenames:
6641 info = nodeinfo[node]
6642 info.Raise("Cannot get current information from node %s" % node,
6643 prereq=True, ecode=errors.ECODE_ENVIRON)
6644 (_, _, (hv_info, )) = info.payload
6645 num_cpus = hv_info.get("cpu_total", None)
6646 if not isinstance(num_cpus, int):
6647 raise errors.OpPrereqError("Can't compute the number of physical CPUs"
6648 " on node %s, result was '%s'" %
6649 (node, num_cpus), errors.ECODE_ENVIRON)
6650 if requested > num_cpus:
6651 raise errors.OpPrereqError("Node %s has %s physical CPUs, but %s are "
6652 "required" % (node, num_cpus, requested),
6656 class LUInstanceStartup(LogicalUnit):
6657 """Starts an instance.
6660 HPATH = "instance-start"
6661 HTYPE = constants.HTYPE_INSTANCE
6664 def CheckArguments(self):
6666 if self.op.beparams:
6667 # fill the beparams dict
6668 objects.UpgradeBeParams(self.op.beparams)
6669 utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
6671 def ExpandNames(self):
6672 self._ExpandAndLockInstance()
6673 self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
6675 def DeclareLocks(self, level):
6676 if level == locking.LEVEL_NODE_RES:
6677 self._LockInstancesNodes(primary_only=True, level=locking.LEVEL_NODE_RES)
6679 def BuildHooksEnv(self):
6682 This runs on master, primary and secondary nodes of the instance.
6686 "FORCE": self.op.force,
6689 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6693 def BuildHooksNodes(self):
6694 """Build hooks nodes.
6697 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6700 def CheckPrereq(self):
6701 """Check prerequisites.
6703 This checks that the instance is in the cluster.
6706 self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6707 assert self.instance is not None, \
6708 "Cannot retrieve locked instance %s" % self.op.instance_name
6711 if self.op.hvparams:
6712 # check hypervisor parameter syntax (locally)
6713 cluster = self.cfg.GetClusterInfo()
6714 utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
6715 filled_hvp = cluster.FillHV(instance)
6716 filled_hvp.update(self.op.hvparams)
6717 hv_type = hypervisor.GetHypervisor(instance.hypervisor)
6718 hv_type.CheckParameterSyntax(filled_hvp)
6719 _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
6721 _CheckInstanceState(self, instance, INSTANCE_ONLINE)
6723 self.primary_offline = self.cfg.GetNodeInfo(instance.primary_node).offline
6725 if self.primary_offline and self.op.ignore_offline_nodes:
6726 self.proc.LogWarning("Ignoring offline primary node")
6728 if self.op.hvparams or self.op.beparams:
6729 self.proc.LogWarning("Overridden parameters are ignored")
6731 _CheckNodeOnline(self, instance.primary_node)
6733 bep = self.cfg.GetClusterInfo().FillBE(instance)
6734 bep.update(self.op.beparams)
6736 # check bridges existence
6737 _CheckInstanceBridgesExist(self, instance)
6739 remote_info = self.rpc.call_instance_info(instance.primary_node,
6741 instance.hypervisor)
6742 remote_info.Raise("Error checking node %s" % instance.primary_node,
6743 prereq=True, ecode=errors.ECODE_ENVIRON)
6744 if not remote_info.payload: # not running already
6745 _CheckNodeFreeMemory(self, instance.primary_node,
6746 "starting instance %s" % instance.name,
6747 bep[constants.BE_MINMEM], instance.hypervisor)
6749 def Exec(self, feedback_fn):
6750 """Start the instance.
6753 instance = self.instance
6754 force = self.op.force
6756 if not self.op.no_remember:
6757 self.cfg.MarkInstanceUp(instance.name)
6759 if self.primary_offline:
6760 assert self.op.ignore_offline_nodes
6761 self.proc.LogInfo("Primary node offline, marked instance as started")
6763 node_current = instance.primary_node
6765 _StartInstanceDisks(self, instance, force)
6768 self.rpc.call_instance_start(node_current,
6769 (instance, self.op.hvparams,
6771 self.op.startup_paused)
6772 msg = result.fail_msg
6774 _ShutdownInstanceDisks(self, instance)
6775 raise errors.OpExecError("Could not start instance: %s" % msg)
6778 class LUInstanceReboot(LogicalUnit):
6779 """Reboot an instance.
6782 HPATH = "instance-reboot"
6783 HTYPE = constants.HTYPE_INSTANCE
6786 def ExpandNames(self):
6787 self._ExpandAndLockInstance()
6789 def BuildHooksEnv(self):
6792 This runs on master, primary and secondary nodes of the instance.
6796 "IGNORE_SECONDARIES": self.op.ignore_secondaries,
6797 "REBOOT_TYPE": self.op.reboot_type,
6798 "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
6801 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6805 def BuildHooksNodes(self):
6806 """Build hooks nodes.
6809 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6812 def CheckPrereq(self):
6813 """Check prerequisites.
6815 This checks that the instance is in the cluster.
6818 self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6819 assert self.instance is not None, \
6820 "Cannot retrieve locked instance %s" % self.op.instance_name
6821 _CheckInstanceState(self, instance, INSTANCE_ONLINE)
6822 _CheckNodeOnline(self, instance.primary_node)
6824 # check bridges existence
6825 _CheckInstanceBridgesExist(self, instance)
6827 def Exec(self, feedback_fn):
6828 """Reboot the instance.
6831 instance = self.instance
6832 ignore_secondaries = self.op.ignore_secondaries
6833 reboot_type = self.op.reboot_type
6835 remote_info = self.rpc.call_instance_info(instance.primary_node,
6837 instance.hypervisor)
6838 remote_info.Raise("Error checking node %s" % instance.primary_node)
6839 instance_running = bool(remote_info.payload)
6841 node_current = instance.primary_node
6843 if instance_running and reboot_type in [constants.INSTANCE_REBOOT_SOFT,
6844 constants.INSTANCE_REBOOT_HARD]:
6845 for disk in instance.disks:
6846 self.cfg.SetDiskID(disk, node_current)
6847 result = self.rpc.call_instance_reboot(node_current, instance,
6849 self.op.shutdown_timeout)
6850 result.Raise("Could not reboot instance")
6852 if instance_running:
6853 result = self.rpc.call_instance_shutdown(node_current, instance,
6854 self.op.shutdown_timeout)
6855 result.Raise("Could not shutdown instance for full reboot")
6856 _ShutdownInstanceDisks(self, instance)
6858 self.LogInfo("Instance %s was already stopped, starting now",
6860 _StartInstanceDisks(self, instance, ignore_secondaries)
6861 result = self.rpc.call_instance_start(node_current,
6862 (instance, None, None), False)
6863 msg = result.fail_msg
6865 _ShutdownInstanceDisks(self, instance)
6866 raise errors.OpExecError("Could not start instance for"
6867 " full reboot: %s" % msg)
6869 self.cfg.MarkInstanceUp(instance.name)
6872 class LUInstanceShutdown(LogicalUnit):
6873 """Shutdown an instance.
6876 HPATH = "instance-stop"
6877 HTYPE = constants.HTYPE_INSTANCE
6880 def ExpandNames(self):
6881 self._ExpandAndLockInstance()
6883 def BuildHooksEnv(self):
6886 This runs on master, primary and secondary nodes of the instance.
6889 env = _BuildInstanceHookEnvByObject(self, self.instance)
6890 env["TIMEOUT"] = self.op.timeout
6893 def BuildHooksNodes(self):
6894 """Build hooks nodes.
6897 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6900 def CheckPrereq(self):
6901 """Check prerequisites.
6903 This checks that the instance is in the cluster.
6906 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6907 assert self.instance is not None, \
6908 "Cannot retrieve locked instance %s" % self.op.instance_name
6910 _CheckInstanceState(self, self.instance, INSTANCE_ONLINE)
6912 self.primary_offline = \
6913 self.cfg.GetNodeInfo(self.instance.primary_node).offline
6915 if self.primary_offline and self.op.ignore_offline_nodes:
6916 self.proc.LogWarning("Ignoring offline primary node")
6918 _CheckNodeOnline(self, self.instance.primary_node)
6920 def Exec(self, feedback_fn):
6921 """Shutdown the instance.
6924 instance = self.instance
6925 node_current = instance.primary_node
6926 timeout = self.op.timeout
6928 if not self.op.no_remember:
6929 self.cfg.MarkInstanceDown(instance.name)
6931 if self.primary_offline:
6932 assert self.op.ignore_offline_nodes
6933 self.proc.LogInfo("Primary node offline, marked instance as stopped")
6935 result = self.rpc.call_instance_shutdown(node_current, instance, timeout)
6936 msg = result.fail_msg
6938 self.proc.LogWarning("Could not shutdown instance: %s" % msg)
6940 _ShutdownInstanceDisks(self, instance)
6943 class LUInstanceReinstall(LogicalUnit):
6944 """Reinstall an instance.
6947 HPATH = "instance-reinstall"
6948 HTYPE = constants.HTYPE_INSTANCE
6951 def ExpandNames(self):
6952 self._ExpandAndLockInstance()
6954 def BuildHooksEnv(self):
6957 This runs on master, primary and secondary nodes of the instance.
6960 return _BuildInstanceHookEnvByObject(self, self.instance)
6962 def BuildHooksNodes(self):
6963 """Build hooks nodes.
6966 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6969 def CheckPrereq(self):
6970 """Check prerequisites.
6972 This checks that the instance is in the cluster and is not running.
6975 instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6976 assert instance is not None, \
6977 "Cannot retrieve locked instance %s" % self.op.instance_name
6978 _CheckNodeOnline(self, instance.primary_node, "Instance primary node"
6979 " offline, cannot reinstall")
6980 for node in instance.secondary_nodes:
6981 _CheckNodeOnline(self, node, "Instance secondary node offline,"
6982 " cannot reinstall")
6984 if instance.disk_template == constants.DT_DISKLESS:
6985 raise errors.OpPrereqError("Instance '%s' has no disks" %
6986 self.op.instance_name,
6988 _CheckInstanceState(self, instance, INSTANCE_DOWN, msg="cannot reinstall")
6990 if self.op.os_type is not None:
6992 pnode = _ExpandNodeName(self.cfg, instance.primary_node)
6993 _CheckNodeHasOS(self, pnode, self.op.os_type, self.op.force_variant)
6994 instance_os = self.op.os_type
6996 instance_os = instance.os
6998 nodelist = list(instance.all_nodes)
7000 if self.op.osparams:
7001 i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
7002 _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
7003 self.os_inst = i_osdict # the new dict (without defaults)
7007 self.instance = instance
7009 def Exec(self, feedback_fn):
7010 """Reinstall the instance.
7013 inst = self.instance
7015 if self.op.os_type is not None:
7016 feedback_fn("Changing OS to '%s'..." % self.op.os_type)
7017 inst.os = self.op.os_type
7018 # Write to configuration
7019 self.cfg.Update(inst, feedback_fn)
7021 _StartInstanceDisks(self, inst, None)
7023 feedback_fn("Running the instance OS create scripts...")
7024 # FIXME: pass debug option from opcode to backend
7025 result = self.rpc.call_instance_os_add(inst.primary_node,
7026 (inst, self.os_inst), True,
7027 self.op.debug_level)
7028 result.Raise("Could not install OS for instance %s on node %s" %
7029 (inst.name, inst.primary_node))
7031 _ShutdownInstanceDisks(self, inst)
7034 class LUInstanceRecreateDisks(LogicalUnit):
7035 """Recreate an instance's missing disks.
7038 HPATH = "instance-recreate-disks"
7039 HTYPE = constants.HTYPE_INSTANCE
7042 _MODIFYABLE = frozenset([
7043 constants.IDISK_SIZE,
7044 constants.IDISK_MODE,
7047 # New or changed disk parameters may have different semantics
7048 assert constants.IDISK_PARAMS == (_MODIFYABLE | frozenset([
7049 constants.IDISK_ADOPT,
7051 # TODO: Implement support changing VG while recreating
7053 constants.IDISK_METAVG,
7056 def CheckArguments(self):
7057 if self.op.disks and ht.TPositiveInt(self.op.disks[0]):
7058 # Normalize and convert deprecated list of disk indices
7059 self.op.disks = [(idx, {}) for idx in sorted(frozenset(self.op.disks))]
7061 duplicates = utils.FindDuplicates(map(compat.fst, self.op.disks))
7063 raise errors.OpPrereqError("Some disks have been specified more than"
7064 " once: %s" % utils.CommaJoin(duplicates),
7067 for (idx, params) in self.op.disks:
7068 utils.ForceDictType(params, constants.IDISK_PARAMS_TYPES)
7069 unsupported = frozenset(params.keys()) - self._MODIFYABLE
7071 raise errors.OpPrereqError("Parameters for disk %s try to change"
7072 " unmodifyable parameter(s): %s" %
7073 (idx, utils.CommaJoin(unsupported)),
7076 def ExpandNames(self):
7077 self._ExpandAndLockInstance()
7078 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
7080 self.op.nodes = [_ExpandNodeName(self.cfg, n) for n in self.op.nodes]
7081 self.needed_locks[locking.LEVEL_NODE] = list(self.op.nodes)
7083 self.needed_locks[locking.LEVEL_NODE] = []
7084 self.needed_locks[locking.LEVEL_NODE_RES] = []
7086 def DeclareLocks(self, level):
7087 if level == locking.LEVEL_NODE:
7088 # if we replace the nodes, we only need to lock the old primary,
7089 # otherwise we need to lock all nodes for disk re-creation
7090 primary_only = bool(self.op.nodes)
7091 self._LockInstancesNodes(primary_only=primary_only)
7092 elif level == locking.LEVEL_NODE_RES:
7094 self.needed_locks[locking.LEVEL_NODE_RES] = \
7095 self.needed_locks[locking.LEVEL_NODE][:]
7097 def BuildHooksEnv(self):
7100 This runs on master, primary and secondary nodes of the instance.
7103 return _BuildInstanceHookEnvByObject(self, self.instance)
7105 def BuildHooksNodes(self):
7106 """Build hooks nodes.
7109 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
7112 def CheckPrereq(self):
7113 """Check prerequisites.
7115 This checks that the instance is in the cluster and is not running.
7118 instance = self.cfg.GetInstanceInfo(self.op.instance_name)
7119 assert instance is not None, \
7120 "Cannot retrieve locked instance %s" % self.op.instance_name
7122 if len(self.op.nodes) != len(instance.all_nodes):
7123 raise errors.OpPrereqError("Instance %s currently has %d nodes, but"
7124 " %d replacement nodes were specified" %
7125 (instance.name, len(instance.all_nodes),
7126 len(self.op.nodes)),
7128 assert instance.disk_template != constants.DT_DRBD8 or \
7129 len(self.op.nodes) == 2
7130 assert instance.disk_template != constants.DT_PLAIN or \
7131 len(self.op.nodes) == 1
7132 primary_node = self.op.nodes[0]
7134 primary_node = instance.primary_node
7135 _CheckNodeOnline(self, primary_node)
7137 if instance.disk_template == constants.DT_DISKLESS:
7138 raise errors.OpPrereqError("Instance '%s' has no disks" %
7139 self.op.instance_name, errors.ECODE_INVAL)
7141 # if we replace nodes *and* the old primary is offline, we don't
7143 assert instance.primary_node in self.owned_locks(locking.LEVEL_NODE)
7144 assert instance.primary_node in self.owned_locks(locking.LEVEL_NODE_RES)
7145 old_pnode = self.cfg.GetNodeInfo(instance.primary_node)
7146 if not (self.op.nodes and old_pnode.offline):
7147 _CheckInstanceState(self, instance, INSTANCE_NOT_RUNNING,
7148 msg="cannot recreate disks")
7151 self.disks = dict(self.op.disks)
7153 self.disks = dict((idx, {}) for idx in range(len(instance.disks)))
7155 maxidx = max(self.disks.keys())
7156 if maxidx >= len(instance.disks):
7157 raise errors.OpPrereqError("Invalid disk index '%s'" % maxidx,
7160 if (self.op.nodes and
7161 sorted(self.disks.keys()) != range(len(instance.disks))):
7162 raise errors.OpPrereqError("Can't recreate disks partially and"
7163 " change the nodes at the same time",
7166 self.instance = instance
7168 def Exec(self, feedback_fn):
7169 """Recreate the disks.
7172 instance = self.instance
7174 assert (self.owned_locks(locking.LEVEL_NODE) ==
7175 self.owned_locks(locking.LEVEL_NODE_RES))
7178 mods = [] # keeps track of needed changes
7180 for idx, disk in enumerate(instance.disks):
7182 changes = self.disks[idx]
7184 # Disk should not be recreated
7188 # update secondaries for disks, if needed
7189 if self.op.nodes and disk.dev_type == constants.LD_DRBD8:
7190 # need to update the nodes and minors
7191 assert len(self.op.nodes) == 2
7192 assert len(disk.logical_id) == 6 # otherwise disk internals
7194 (_, _, old_port, _, _, old_secret) = disk.logical_id
7195 new_minors = self.cfg.AllocateDRBDMinor(self.op.nodes, instance.name)
7196 new_id = (self.op.nodes[0], self.op.nodes[1], old_port,
7197 new_minors[0], new_minors[1], old_secret)
7198 assert len(disk.logical_id) == len(new_id)
7202 mods.append((idx, new_id, changes))
7204 # now that we have passed all asserts above, we can apply the mods
7205 # in a single run (to avoid partial changes)
7206 for idx, new_id, changes in mods:
7207 disk = instance.disks[idx]
7208 if new_id is not None:
7209 assert disk.dev_type == constants.LD_DRBD8
7210 disk.logical_id = new_id
7212 disk.Update(size=changes.get(constants.IDISK_SIZE, None),
7213 mode=changes.get(constants.IDISK_MODE, None))
7215 # change primary node, if needed
7217 instance.primary_node = self.op.nodes[0]
7218 self.LogWarning("Changing the instance's nodes, you will have to"
7219 " remove any disks left on the older nodes manually")
7222 self.cfg.Update(instance, feedback_fn)
7224 _CreateDisks(self, instance, to_skip=to_skip)
7227 class LUInstanceRename(LogicalUnit):
7228 """Rename an instance.
7231 HPATH = "instance-rename"
7232 HTYPE = constants.HTYPE_INSTANCE
7234 def CheckArguments(self):
7238 if self.op.ip_check and not self.op.name_check:
7239 # TODO: make the ip check more flexible and not depend on the name check
7240 raise errors.OpPrereqError("IP address check requires a name check",
7243 def BuildHooksEnv(self):
7246 This runs on master, primary and secondary nodes of the instance.
7249 env = _BuildInstanceHookEnvByObject(self, self.instance)
7250 env["INSTANCE_NEW_NAME"] = self.op.new_name
7253 def BuildHooksNodes(self):
7254 """Build hooks nodes.
7257 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
7260 def CheckPrereq(self):
7261 """Check prerequisites.
7263 This checks that the instance is in the cluster and is not running.
7266 self.op.instance_name = _ExpandInstanceName(self.cfg,
7267 self.op.instance_name)
7268 instance = self.cfg.GetInstanceInfo(self.op.instance_name)
7269 assert instance is not None
7270 _CheckNodeOnline(self, instance.primary_node)
7271 _CheckInstanceState(self, instance, INSTANCE_NOT_RUNNING,
7272 msg="cannot rename")
7273 self.instance = instance
7275 new_name = self.op.new_name
7276 if self.op.name_check:
7277 hostname = netutils.GetHostname(name=new_name)
7278 if hostname.name != new_name:
7279 self.LogInfo("Resolved given name '%s' to '%s'", new_name,
7281 if not utils.MatchNameComponent(self.op.new_name, [hostname.name]):
7282 raise errors.OpPrereqError(("Resolved hostname '%s' does not look the"
7283 " same as given hostname '%s'") %
7284 (hostname.name, self.op.new_name),
7286 new_name = self.op.new_name = hostname.name
7287 if (self.op.ip_check and
7288 netutils.TcpPing(hostname.ip, constants.DEFAULT_NODED_PORT)):
7289 raise errors.OpPrereqError("IP %s of instance %s already in use" %
7290 (hostname.ip, new_name),
7291 errors.ECODE_NOTUNIQUE)
7293 instance_list = self.cfg.GetInstanceList()
7294 if new_name in instance_list and new_name != instance.name:
7295 raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
7296 new_name, errors.ECODE_EXISTS)
7298 def Exec(self, feedback_fn):
7299 """Rename the instance.
7302 inst = self.instance
7303 old_name = inst.name
7305 rename_file_storage = False
7306 if (inst.disk_template in constants.DTS_FILEBASED and
7307 self.op.new_name != inst.name):
7308 old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
7309 rename_file_storage = True
7311 self.cfg.RenameInstance(inst.name, self.op.new_name)
7312 # Change the instance lock. This is definitely safe while we hold the BGL.
7313 # Otherwise the new lock would have to be added in acquired mode.
7315 self.glm.remove(locking.LEVEL_INSTANCE, old_name)
7316 self.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
7318 # re-read the instance from the configuration after rename
7319 inst = self.cfg.GetInstanceInfo(self.op.new_name)
7321 if rename_file_storage:
7322 new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
7323 result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
7324 old_file_storage_dir,
7325 new_file_storage_dir)
7326 result.Raise("Could not rename on node %s directory '%s' to '%s'"
7327 " (but the instance has been renamed in Ganeti)" %
7328 (inst.primary_node, old_file_storage_dir,
7329 new_file_storage_dir))
7331 _StartInstanceDisks(self, inst, None)
7333 result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
7334 old_name, self.op.debug_level)
7335 msg = result.fail_msg
7337 msg = ("Could not run OS rename script for instance %s on node %s"
7338 " (but the instance has been renamed in Ganeti): %s" %
7339 (inst.name, inst.primary_node, msg))
7340 self.proc.LogWarning(msg)
7342 _ShutdownInstanceDisks(self, inst)
7347 class LUInstanceRemove(LogicalUnit):
7348 """Remove an instance.
7351 HPATH = "instance-remove"
7352 HTYPE = constants.HTYPE_INSTANCE
7355 def ExpandNames(self):
7356 self._ExpandAndLockInstance()
7357 self.needed_locks[locking.LEVEL_NODE] = []
7358 self.needed_locks[locking.LEVEL_NODE_RES] = []
7359 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
7361 def DeclareLocks(self, level):
7362 if level == locking.LEVEL_NODE:
7363 self._LockInstancesNodes()
7364 elif level == locking.LEVEL_NODE_RES:
7366 self.needed_locks[locking.LEVEL_NODE_RES] = \
7367 self.needed_locks[locking.LEVEL_NODE][:]
7369 def BuildHooksEnv(self):
7372 This runs on master, primary and secondary nodes of the instance.
7375 env = _BuildInstanceHookEnvByObject(self, self.instance)
7376 env["SHUTDOWN_TIMEOUT"] = self.op.shutdown_timeout
7379 def BuildHooksNodes(self):
7380 """Build hooks nodes.
7383 nl = [self.cfg.GetMasterNode()]
7384 nl_post = list(self.instance.all_nodes) + nl
7385 return (nl, nl_post)
7387 def CheckPrereq(self):
7388 """Check prerequisites.
7390 This checks that the instance is in the cluster.
7393 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
7394 assert self.instance is not None, \
7395 "Cannot retrieve locked instance %s" % self.op.instance_name
7397 def Exec(self, feedback_fn):
7398 """Remove the instance.
7401 instance = self.instance
7402 logging.info("Shutting down instance %s on node %s",
7403 instance.name, instance.primary_node)
7405 result = self.rpc.call_instance_shutdown(instance.primary_node, instance,
7406 self.op.shutdown_timeout)
7407 msg = result.fail_msg
7409 if self.op.ignore_failures:
7410 feedback_fn("Warning: can't shutdown instance: %s" % msg)
7412 raise errors.OpExecError("Could not shutdown instance %s on"
7414 (instance.name, instance.primary_node, msg))
7416 assert (self.owned_locks(locking.LEVEL_NODE) ==
7417 self.owned_locks(locking.LEVEL_NODE_RES))
7418 assert not (set(instance.all_nodes) -
7419 self.owned_locks(locking.LEVEL_NODE)), \
7420 "Not owning correct locks"
7422 _RemoveInstance(self, feedback_fn, instance, self.op.ignore_failures)
7425 def _RemoveInstance(lu, feedback_fn, instance, ignore_failures):
7426 """Utility function to remove an instance.
7429 logging.info("Removing block devices for instance %s", instance.name)
7431 if not _RemoveDisks(lu, instance, ignore_failures=ignore_failures):
7432 if not ignore_failures:
7433 raise errors.OpExecError("Can't remove instance's disks")
7434 feedback_fn("Warning: can't remove instance's disks")
7436 logging.info("Removing instance %s out of cluster config", instance.name)
7438 lu.cfg.RemoveInstance(instance.name)
7440 assert not lu.remove_locks.get(locking.LEVEL_INSTANCE), \
7441 "Instance lock removal conflict"
7443 # Remove lock for the instance
7444 lu.remove_locks[locking.LEVEL_INSTANCE] = instance.name
7447 class LUInstanceQuery(NoHooksLU):
7448 """Logical unit for querying instances.
7451 # pylint: disable=W0142
7454 def CheckArguments(self):
7455 self.iq = _InstanceQuery(qlang.MakeSimpleFilter("name", self.op.names),
7456 self.op.output_fields, self.op.use_locking)
7458 def ExpandNames(self):
7459 self.iq.ExpandNames(self)
7461 def DeclareLocks(self, level):
7462 self.iq.DeclareLocks(self, level)
7464 def Exec(self, feedback_fn):
7465 return self.iq.OldStyleQuery(self)
7468 class LUInstanceFailover(LogicalUnit):
7469 """Failover an instance.
7472 HPATH = "instance-failover"
7473 HTYPE = constants.HTYPE_INSTANCE
7476 def CheckArguments(self):
7477 """Check the arguments.
7480 self.iallocator = getattr(self.op, "iallocator", None)
7481 self.target_node = getattr(self.op, "target_node", None)
7483 def ExpandNames(self):
7484 self._ExpandAndLockInstance()
7486 if self.op.target_node is not None:
7487 self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
7489 self.needed_locks[locking.LEVEL_NODE] = []
7490 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
7492 self.needed_locks[locking.LEVEL_NODE_RES] = []
7493 self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
7495 ignore_consistency = self.op.ignore_consistency
7496 shutdown_timeout = self.op.shutdown_timeout
7497 self._migrater = TLMigrateInstance(self, self.op.instance_name,
7500 ignore_consistency=ignore_consistency,
7501 shutdown_timeout=shutdown_timeout,
7502 ignore_ipolicy=self.op.ignore_ipolicy)
7503 self.tasklets = [self._migrater]
7505 def DeclareLocks(self, level):
7506 if level == locking.LEVEL_NODE:
7507 instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
7508 if instance.disk_template in constants.DTS_EXT_MIRROR:
7509 if self.op.target_node is None:
7510 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7512 self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
7513 self.op.target_node]
7514 del self.recalculate_locks[locking.LEVEL_NODE]
7516 self._LockInstancesNodes()
7517 elif level == locking.LEVEL_NODE_RES:
7519 self.needed_locks[locking.LEVEL_NODE_RES] = \
7520 self.needed_locks[locking.LEVEL_NODE][:]
7522 def BuildHooksEnv(self):
7525 This runs on master, primary and secondary nodes of the instance.
7528 instance = self._migrater.instance
7529 source_node = instance.primary_node
7530 target_node = self.op.target_node
7532 "IGNORE_CONSISTENCY": self.op.ignore_consistency,
7533 "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
7534 "OLD_PRIMARY": source_node,
7535 "NEW_PRIMARY": target_node,
7538 if instance.disk_template in constants.DTS_INT_MIRROR:
7539 env["OLD_SECONDARY"] = instance.secondary_nodes[0]
7540 env["NEW_SECONDARY"] = source_node
7542 env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = ""
7544 env.update(_BuildInstanceHookEnvByObject(self, instance))
7548 def BuildHooksNodes(self):
7549 """Build hooks nodes.
7552 instance = self._migrater.instance
7553 nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
7554 return (nl, nl + [instance.primary_node])
7557 class LUInstanceMigrate(LogicalUnit):
7558 """Migrate an instance.
7560 This is migration without shutting down, compared to the failover,
7561 which is done with shutdown.
7564 HPATH = "instance-migrate"
7565 HTYPE = constants.HTYPE_INSTANCE
7568 def ExpandNames(self):
7569 self._ExpandAndLockInstance()
7571 if self.op.target_node is not None:
7572 self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
7574 self.needed_locks[locking.LEVEL_NODE] = []
7575 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
7577 self.needed_locks[locking.LEVEL_NODE] = []
7578 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
7581 TLMigrateInstance(self, self.op.instance_name,
7582 cleanup=self.op.cleanup,
7584 fallback=self.op.allow_failover,
7585 allow_runtime_changes=self.op.allow_runtime_changes,
7586 ignore_ipolicy=self.op.ignore_ipolicy)
7587 self.tasklets = [self._migrater]
7589 def DeclareLocks(self, level):
7590 if level == locking.LEVEL_NODE:
7591 instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
7592 if instance.disk_template in constants.DTS_EXT_MIRROR:
7593 if self.op.target_node is None:
7594 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7596 self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
7597 self.op.target_node]
7598 del self.recalculate_locks[locking.LEVEL_NODE]
7600 self._LockInstancesNodes()
7601 elif level == locking.LEVEL_NODE_RES:
7603 self.needed_locks[locking.LEVEL_NODE_RES] = \
7604 self.needed_locks[locking.LEVEL_NODE][:]
7606 def BuildHooksEnv(self):
7609 This runs on master, primary and secondary nodes of the instance.
7612 instance = self._migrater.instance
7613 source_node = instance.primary_node
7614 target_node = self.op.target_node
7615 env = _BuildInstanceHookEnvByObject(self, instance)
7617 "MIGRATE_LIVE": self._migrater.live,
7618 "MIGRATE_CLEANUP": self.op.cleanup,
7619 "OLD_PRIMARY": source_node,
7620 "NEW_PRIMARY": target_node,
7621 "ALLOW_RUNTIME_CHANGES": self.op.allow_runtime_changes,
7624 if instance.disk_template in constants.DTS_INT_MIRROR:
7625 env["OLD_SECONDARY"] = target_node
7626 env["NEW_SECONDARY"] = source_node
7628 env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = None
7632 def BuildHooksNodes(self):
7633 """Build hooks nodes.
7636 instance = self._migrater.instance
7637 nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
7638 return (nl, nl + [instance.primary_node])
7641 class LUInstanceMove(LogicalUnit):
7642 """Move an instance by data-copying.
7645 HPATH = "instance-move"
7646 HTYPE = constants.HTYPE_INSTANCE
7649 def ExpandNames(self):
7650 self._ExpandAndLockInstance()
7651 target_node = _ExpandNodeName(self.cfg, self.op.target_node)
7652 self.op.target_node = target_node
7653 self.needed_locks[locking.LEVEL_NODE] = [target_node]
7654 self.needed_locks[locking.LEVEL_NODE_RES] = []
7655 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
7657 def DeclareLocks(self, level):
7658 if level == locking.LEVEL_NODE:
7659 self._LockInstancesNodes(primary_only=True)
7660 elif level == locking.LEVEL_NODE_RES:
7662 self.needed_locks[locking.LEVEL_NODE_RES] = \
7663 self.needed_locks[locking.LEVEL_NODE][:]
7665 def BuildHooksEnv(self):
7668 This runs on master, primary and secondary nodes of the instance.
7672 "TARGET_NODE": self.op.target_node,
7673 "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
7675 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
7678 def BuildHooksNodes(self):
7679 """Build hooks nodes.
7683 self.cfg.GetMasterNode(),
7684 self.instance.primary_node,
7685 self.op.target_node,
7689 def CheckPrereq(self):
7690 """Check prerequisites.
7692 This checks that the instance is in the cluster.
7695 self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
7696 assert self.instance is not None, \
7697 "Cannot retrieve locked instance %s" % self.op.instance_name
7699 node = self.cfg.GetNodeInfo(self.op.target_node)
7700 assert node is not None, \
7701 "Cannot retrieve locked node %s" % self.op.target_node
7703 self.target_node = target_node = node.name
7705 if target_node == instance.primary_node:
7706 raise errors.OpPrereqError("Instance %s is already on the node %s" %
7707 (instance.name, target_node),
7710 bep = self.cfg.GetClusterInfo().FillBE(instance)
7712 for idx, dsk in enumerate(instance.disks):
7713 if dsk.dev_type not in (constants.LD_LV, constants.LD_FILE):
7714 raise errors.OpPrereqError("Instance disk %d has a complex layout,"
7715 " cannot copy" % idx, errors.ECODE_STATE)
7717 _CheckNodeOnline(self, target_node)
7718 _CheckNodeNotDrained(self, target_node)
7719 _CheckNodeVmCapable(self, target_node)
7720 ipolicy = _CalculateGroupIPolicy(self.cfg.GetClusterInfo(),
7721 self.cfg.GetNodeGroup(node.group))
7722 _CheckTargetNodeIPolicy(self, ipolicy, instance, node,
7723 ignore=self.op.ignore_ipolicy)
7725 if instance.admin_state == constants.ADMINST_UP:
7726 # check memory requirements on the secondary node
7727 _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
7728 instance.name, bep[constants.BE_MAXMEM],
7729 instance.hypervisor)
7731 self.LogInfo("Not checking memory on the secondary node as"
7732 " instance will not be started")
7734 # check bridge existance
7735 _CheckInstanceBridgesExist(self, instance, node=target_node)
7737 def Exec(self, feedback_fn):
7738 """Move an instance.
7740 The move is done by shutting it down on its present node, copying
7741 the data over (slow) and starting it on the new node.
7744 instance = self.instance
7746 source_node = instance.primary_node
7747 target_node = self.target_node
7749 self.LogInfo("Shutting down instance %s on source node %s",
7750 instance.name, source_node)
7752 assert (self.owned_locks(locking.LEVEL_NODE) ==
7753 self.owned_locks(locking.LEVEL_NODE_RES))
7755 result = self.rpc.call_instance_shutdown(source_node, instance,
7756 self.op.shutdown_timeout)
7757 msg = result.fail_msg
7759 if self.op.ignore_consistency:
7760 self.proc.LogWarning("Could not shutdown instance %s on node %s."
7761 " Proceeding anyway. Please make sure node"
7762 " %s is down. Error details: %s",
7763 instance.name, source_node, source_node, msg)
7765 raise errors.OpExecError("Could not shutdown instance %s on"
7767 (instance.name, source_node, msg))
7769 # create the target disks
7771 _CreateDisks(self, instance, target_node=target_node)
7772 except errors.OpExecError:
7773 self.LogWarning("Device creation failed, reverting...")
7775 _RemoveDisks(self, instance, target_node=target_node)
7777 self.cfg.ReleaseDRBDMinors(instance.name)
7780 cluster_name = self.cfg.GetClusterInfo().cluster_name
7783 # activate, get path, copy the data over
7784 for idx, disk in enumerate(instance.disks):
7785 self.LogInfo("Copying data for disk %d", idx)
7786 result = self.rpc.call_blockdev_assemble(target_node, (disk, instance),
7787 instance.name, True, idx)
7789 self.LogWarning("Can't assemble newly created disk %d: %s",
7790 idx, result.fail_msg)
7791 errs.append(result.fail_msg)
7793 dev_path = result.payload
7794 result = self.rpc.call_blockdev_export(source_node, (disk, instance),
7795 target_node, dev_path,
7798 self.LogWarning("Can't copy data over for disk %d: %s",
7799 idx, result.fail_msg)
7800 errs.append(result.fail_msg)
7804 self.LogWarning("Some disks failed to copy, aborting")
7806 _RemoveDisks(self, instance, target_node=target_node)
7808 self.cfg.ReleaseDRBDMinors(instance.name)
7809 raise errors.OpExecError("Errors during disk copy: %s" %
7812 instance.primary_node = target_node
7813 self.cfg.Update(instance, feedback_fn)
7815 self.LogInfo("Removing the disks on the original node")
7816 _RemoveDisks(self, instance, target_node=source_node)
7818 # Only start the instance if it's marked as up
7819 if instance.admin_state == constants.ADMINST_UP:
7820 self.LogInfo("Starting instance %s on node %s",
7821 instance.name, target_node)
7823 disks_ok, _ = _AssembleInstanceDisks(self, instance,
7824 ignore_secondaries=True)
7826 _ShutdownInstanceDisks(self, instance)
7827 raise errors.OpExecError("Can't activate the instance's disks")
7829 result = self.rpc.call_instance_start(target_node,
7830 (instance, None, None), False)
7831 msg = result.fail_msg
7833 _ShutdownInstanceDisks(self, instance)
7834 raise errors.OpExecError("Could not start instance %s on node %s: %s" %
7835 (instance.name, target_node, msg))
7838 class LUNodeMigrate(LogicalUnit):
7839 """Migrate all instances from a node.
7842 HPATH = "node-migrate"
7843 HTYPE = constants.HTYPE_NODE
7846 def CheckArguments(self):
7849 def ExpandNames(self):
7850 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
7852 self.share_locks = _ShareAll()
7853 self.needed_locks = {
7854 locking.LEVEL_NODE: [self.op.node_name],
7857 def BuildHooksEnv(self):
7860 This runs on the master, the primary and all the secondaries.
7864 "NODE_NAME": self.op.node_name,
7865 "ALLOW_RUNTIME_CHANGES": self.op.allow_runtime_changes,
7868 def BuildHooksNodes(self):
7869 """Build hooks nodes.
7872 nl = [self.cfg.GetMasterNode()]
7875 def CheckPrereq(self):
7878 def Exec(self, feedback_fn):
7879 # Prepare jobs for migration instances
7880 allow_runtime_changes = self.op.allow_runtime_changes
7882 [opcodes.OpInstanceMigrate(instance_name=inst.name,
7885 iallocator=self.op.iallocator,
7886 target_node=self.op.target_node,
7887 allow_runtime_changes=allow_runtime_changes,
7888 ignore_ipolicy=self.op.ignore_ipolicy)]
7889 for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name)
7892 # TODO: Run iallocator in this opcode and pass correct placement options to
7893 # OpInstanceMigrate. Since other jobs can modify the cluster between
7894 # running the iallocator and the actual migration, a good consistency model
7895 # will have to be found.
7897 assert (frozenset(self.owned_locks(locking.LEVEL_NODE)) ==
7898 frozenset([self.op.node_name]))
7900 return ResultWithJobs(jobs)
7903 class TLMigrateInstance(Tasklet):
7904 """Tasklet class for instance migration.
7907 @ivar live: whether the migration will be done live or non-live;
7908 this variable is initalized only after CheckPrereq has run
7909 @type cleanup: boolean
7910 @ivar cleanup: Wheater we cleanup from a failed migration
7911 @type iallocator: string
7912 @ivar iallocator: The iallocator used to determine target_node
7913 @type target_node: string
7914 @ivar target_node: If given, the target_node to reallocate the instance to
7915 @type failover: boolean
7916 @ivar failover: Whether operation results in failover or migration
7917 @type fallback: boolean
7918 @ivar fallback: Whether fallback to failover is allowed if migration not
7920 @type ignore_consistency: boolean
7921 @ivar ignore_consistency: Wheter we should ignore consistency between source
7923 @type shutdown_timeout: int
7924 @ivar shutdown_timeout: In case of failover timeout of the shutdown
7925 @type ignore_ipolicy: bool
7926 @ivar ignore_ipolicy: If true, we can ignore instance policy when migrating
7931 _MIGRATION_POLL_INTERVAL = 1 # seconds
7932 _MIGRATION_FEEDBACK_INTERVAL = 10 # seconds
7934 def __init__(self, lu, instance_name, cleanup=False,
7935 failover=False, fallback=False,
7936 ignore_consistency=False,
7937 allow_runtime_changes=True,
7938 shutdown_timeout=constants.DEFAULT_SHUTDOWN_TIMEOUT,
7939 ignore_ipolicy=False):
7940 """Initializes this class.
7943 Tasklet.__init__(self, lu)
7946 self.instance_name = instance_name
7947 self.cleanup = cleanup
7948 self.live = False # will be overridden later
7949 self.failover = failover
7950 self.fallback = fallback
7951 self.ignore_consistency = ignore_consistency
7952 self.shutdown_timeout = shutdown_timeout
7953 self.ignore_ipolicy = ignore_ipolicy
7954 self.allow_runtime_changes = allow_runtime_changes
7956 def CheckPrereq(self):
7957 """Check prerequisites.
7959 This checks that the instance is in the cluster.
7962 instance_name = _ExpandInstanceName(self.lu.cfg, self.instance_name)
7963 instance = self.cfg.GetInstanceInfo(instance_name)
7964 assert instance is not None
7965 self.instance = instance
7966 cluster = self.cfg.GetClusterInfo()
7968 if (not self.cleanup and
7969 not instance.admin_state == constants.ADMINST_UP and
7970 not self.failover and self.fallback):
7971 self.lu.LogInfo("Instance is marked down or offline, fallback allowed,"
7972 " switching to failover")
7973 self.failover = True
7975 if instance.disk_template not in constants.DTS_MIRRORED:
7980 raise errors.OpPrereqError("Instance's disk layout '%s' does not allow"
7981 " %s" % (instance.disk_template, text),
7984 if instance.disk_template in constants.DTS_EXT_MIRROR:
7985 _CheckIAllocatorOrNode(self.lu, "iallocator", "target_node")
7987 if self.lu.op.iallocator:
7988 self._RunAllocator()
7990 # We set set self.target_node as it is required by
7992 self.target_node = self.lu.op.target_node
7994 # Check that the target node is correct in terms of instance policy
7995 nodeinfo = self.cfg.GetNodeInfo(self.target_node)
7996 group_info = self.cfg.GetNodeGroup(nodeinfo.group)
7997 ipolicy = _CalculateGroupIPolicy(cluster, group_info)
7998 _CheckTargetNodeIPolicy(self.lu, ipolicy, instance, nodeinfo,
7999 ignore=self.ignore_ipolicy)
8001 # self.target_node is already populated, either directly or by the
8003 target_node = self.target_node
8004 if self.target_node == instance.primary_node:
8005 raise errors.OpPrereqError("Cannot migrate instance %s"
8006 " to its primary (%s)" %
8007 (instance.name, instance.primary_node))
8009 if len(self.lu.tasklets) == 1:
8010 # It is safe to release locks only when we're the only tasklet
8012 _ReleaseLocks(self.lu, locking.LEVEL_NODE,
8013 keep=[instance.primary_node, self.target_node])
8016 secondary_nodes = instance.secondary_nodes
8017 if not secondary_nodes:
8018 raise errors.ConfigurationError("No secondary node but using"
8019 " %s disk template" %
8020 instance.disk_template)
8021 target_node = secondary_nodes[0]
8022 if self.lu.op.iallocator or (self.lu.op.target_node and
8023 self.lu.op.target_node != target_node):
8025 text = "failed over"
8028 raise errors.OpPrereqError("Instances with disk template %s cannot"
8029 " be %s to arbitrary nodes"
8030 " (neither an iallocator nor a target"
8031 " node can be passed)" %
8032 (instance.disk_template, text),
8034 nodeinfo = self.cfg.GetNodeInfo(target_node)
8035 group_info = self.cfg.GetNodeGroup(nodeinfo.group)
8036 ipolicy = _CalculateGroupIPolicy(cluster, group_info)
8037 _CheckTargetNodeIPolicy(self.lu, ipolicy, instance, nodeinfo,
8038 ignore=self.ignore_ipolicy)
8040 i_be = cluster.FillBE(instance)
8042 # check memory requirements on the secondary node
8043 if (not self.cleanup and
8044 (not self.failover or instance.admin_state == constants.ADMINST_UP)):
8045 self.tgt_free_mem = _CheckNodeFreeMemory(self.lu, target_node,
8046 "migrating instance %s" %
8048 i_be[constants.BE_MINMEM],
8049 instance.hypervisor)
8051 self.lu.LogInfo("Not checking memory on the secondary node as"
8052 " instance will not be started")
8054 # check if failover must be forced instead of migration
8055 if (not self.cleanup and not self.failover and
8056 i_be[constants.BE_ALWAYS_FAILOVER]):
8058 self.lu.LogInfo("Instance configured to always failover; fallback"
8060 self.failover = True
8062 raise errors.OpPrereqError("This instance has been configured to"
8063 " always failover, please allow failover",
8066 # check bridge existance
8067 _CheckInstanceBridgesExist(self.lu, instance, node=target_node)
8069 if not self.cleanup:
8070 _CheckNodeNotDrained(self.lu, target_node)
8071 if not self.failover:
8072 result = self.rpc.call_instance_migratable(instance.primary_node,
8074 if result.fail_msg and self.fallback:
8075 self.lu.LogInfo("Can't migrate, instance offline, fallback to"
8077 self.failover = True
8079 result.Raise("Can't migrate, please use failover",
8080 prereq=True, ecode=errors.ECODE_STATE)
8082 assert not (self.failover and self.cleanup)
8084 if not self.failover:
8085 if self.lu.op.live is not None and self.lu.op.mode is not None:
8086 raise errors.OpPrereqError("Only one of the 'live' and 'mode'"
8087 " parameters are accepted",
8089 if self.lu.op.live is not None:
8091 self.lu.op.mode = constants.HT_MIGRATION_LIVE
8093 self.lu.op.mode = constants.HT_MIGRATION_NONLIVE
8094 # reset the 'live' parameter to None so that repeated
8095 # invocations of CheckPrereq do not raise an exception
8096 self.lu.op.live = None
8097 elif self.lu.op.mode is None:
8098 # read the default value from the hypervisor
8099 i_hv = cluster.FillHV(self.instance, skip_globals=False)
8100 self.lu.op.mode = i_hv[constants.HV_MIGRATION_MODE]
8102 self.live = self.lu.op.mode == constants.HT_MIGRATION_LIVE
8104 # Failover is never live
8107 if not (self.failover or self.cleanup):
8108 remote_info = self.rpc.call_instance_info(instance.primary_node,
8110 instance.hypervisor)
8111 remote_info.Raise("Error checking instance on node %s" %
8112 instance.primary_node)
8113 instance_running = bool(remote_info.payload)
8114 if instance_running:
8115 self.current_mem = int(remote_info.payload["memory"])
8117 def _RunAllocator(self):
8118 """Run the allocator based on input opcode.
8121 # FIXME: add a self.ignore_ipolicy option
8122 ial = IAllocator(self.cfg, self.rpc,
8123 mode=constants.IALLOCATOR_MODE_RELOC,
8124 name=self.instance_name,
8125 relocate_from=[self.instance.primary_node],
8128 ial.Run(self.lu.op.iallocator)
8131 raise errors.OpPrereqError("Can't compute nodes using"
8132 " iallocator '%s': %s" %
8133 (self.lu.op.iallocator, ial.info),
8135 if len(ial.result) != ial.required_nodes:
8136 raise errors.OpPrereqError("iallocator '%s' returned invalid number"
8137 " of nodes (%s), required %s" %
8138 (self.lu.op.iallocator, len(ial.result),
8139 ial.required_nodes), errors.ECODE_FAULT)
8140 self.target_node = ial.result[0]
8141 self.lu.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
8142 self.instance_name, self.lu.op.iallocator,
8143 utils.CommaJoin(ial.result))
8145 def _WaitUntilSync(self):
8146 """Poll with custom rpc for disk sync.
8148 This uses our own step-based rpc call.
8151 self.feedback_fn("* wait until resync is done")
8155 result = self.rpc.call_drbd_wait_sync(self.all_nodes,
8157 (self.instance.disks,
8160 for node, nres in result.items():
8161 nres.Raise("Cannot resync disks on node %s" % node)
8162 node_done, node_percent = nres.payload
8163 all_done = all_done and node_done
8164 if node_percent is not None:
8165 min_percent = min(min_percent, node_percent)
8167 if min_percent < 100:
8168 self.feedback_fn(" - progress: %.1f%%" % min_percent)
8171 def _EnsureSecondary(self, node):
8172 """Demote a node to secondary.
8175 self.feedback_fn("* switching node %s to secondary mode" % node)
8177 for dev in self.instance.disks:
8178 self.cfg.SetDiskID(dev, node)
8180 result = self.rpc.call_blockdev_close(node, self.instance.name,
8181 self.instance.disks)
8182 result.Raise("Cannot change disk to secondary on node %s" % node)
8184 def _GoStandalone(self):
8185 """Disconnect from the network.
8188 self.feedback_fn("* changing into standalone mode")
8189 result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
8190 self.instance.disks)
8191 for node, nres in result.items():
8192 nres.Raise("Cannot disconnect disks node %s" % node)
8194 def _GoReconnect(self, multimaster):
8195 """Reconnect to the network.
8201 msg = "single-master"
8202 self.feedback_fn("* changing disks into %s mode" % msg)
8203 result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
8204 (self.instance.disks, self.instance),
8205 self.instance.name, multimaster)
8206 for node, nres in result.items():
8207 nres.Raise("Cannot change disks config on node %s" % node)
8209 def _ExecCleanup(self):
8210 """Try to cleanup after a failed migration.
8212 The cleanup is done by:
8213 - check that the instance is running only on one node
8214 (and update the config if needed)
8215 - change disks on its secondary node to secondary
8216 - wait until disks are fully synchronized
8217 - disconnect from the network
8218 - change disks into single-master mode
8219 - wait again until disks are fully synchronized
8222 instance = self.instance
8223 target_node = self.target_node
8224 source_node = self.source_node
8226 # check running on only one node
8227 self.feedback_fn("* checking where the instance actually runs"
8228 " (if this hangs, the hypervisor might be in"
8230 ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
8231 for node, result in ins_l.items():
8232 result.Raise("Can't contact node %s" % node)
8234 runningon_source = instance.name in ins_l[source_node].payload
8235 runningon_target = instance.name in ins_l[target_node].payload
8237 if runningon_source and runningon_target:
8238 raise errors.OpExecError("Instance seems to be running on two nodes,"
8239 " or the hypervisor is confused; you will have"
8240 " to ensure manually that it runs only on one"
8241 " and restart this operation")
8243 if not (runningon_source or runningon_target):
8244 raise errors.OpExecError("Instance does not seem to be running at all;"
8245 " in this case it's safer to repair by"
8246 " running 'gnt-instance stop' to ensure disk"
8247 " shutdown, and then restarting it")
8249 if runningon_target:
8250 # the migration has actually succeeded, we need to update the config
8251 self.feedback_fn("* instance running on secondary node (%s),"
8252 " updating config" % target_node)
8253 instance.primary_node = target_node
8254 self.cfg.Update(instance, self.feedback_fn)
8255 demoted_node = source_node
8257 self.feedback_fn("* instance confirmed to be running on its"
8258 " primary node (%s)" % source_node)
8259 demoted_node = target_node
8261 if instance.disk_template in constants.DTS_INT_MIRROR:
8262 self._EnsureSecondary(demoted_node)
8264 self._WaitUntilSync()
8265 except errors.OpExecError:
8266 # we ignore here errors, since if the device is standalone, it
8267 # won't be able to sync
8269 self._GoStandalone()
8270 self._GoReconnect(False)
8271 self._WaitUntilSync()
8273 self.feedback_fn("* done")
8275 def _RevertDiskStatus(self):
8276 """Try to revert the disk status after a failed migration.
8279 target_node = self.target_node
8280 if self.instance.disk_template in constants.DTS_EXT_MIRROR:
8284 self._EnsureSecondary(target_node)
8285 self._GoStandalone()
8286 self._GoReconnect(False)
8287 self._WaitUntilSync()
8288 except errors.OpExecError, err:
8289 self.lu.LogWarning("Migration failed and I can't reconnect the drives,"
8290 " please try to recover the instance manually;"
8291 " error '%s'" % str(err))
8293 def _AbortMigration(self):
8294 """Call the hypervisor code to abort a started migration.
8297 instance = self.instance
8298 target_node = self.target_node
8299 source_node = self.source_node
8300 migration_info = self.migration_info
8302 abort_result = self.rpc.call_instance_finalize_migration_dst(target_node,
8306 abort_msg = abort_result.fail_msg
8308 logging.error("Aborting migration failed on target node %s: %s",
8309 target_node, abort_msg)
8310 # Don't raise an exception here, as we stil have to try to revert the
8311 # disk status, even if this step failed.
8313 abort_result = self.rpc.call_instance_finalize_migration_src(source_node,
8314 instance, False, self.live)
8315 abort_msg = abort_result.fail_msg
8317 logging.error("Aborting migration failed on source node %s: %s",
8318 source_node, abort_msg)
8320 def _ExecMigration(self):
8321 """Migrate an instance.
8323 The migrate is done by:
8324 - change the disks into dual-master mode
8325 - wait until disks are fully synchronized again
8326 - migrate the instance
8327 - change disks on the new secondary node (the old primary) to secondary
8328 - wait until disks are fully synchronized
8329 - change disks into single-master mode
8332 instance = self.instance
8333 target_node = self.target_node
8334 source_node = self.source_node
8336 # Check for hypervisor version mismatch and warn the user.
8337 nodeinfo = self.rpc.call_node_info([source_node, target_node],
8338 None, [self.instance.hypervisor])
8339 for ninfo in nodeinfo.values():
8340 ninfo.Raise("Unable to retrieve node information from node '%s'" %
8342 (_, _, (src_info, )) = nodeinfo[source_node].payload
8343 (_, _, (dst_info, )) = nodeinfo[target_node].payload
8345 if ((constants.HV_NODEINFO_KEY_VERSION in src_info) and
8346 (constants.HV_NODEINFO_KEY_VERSION in dst_info)):
8347 src_version = src_info[constants.HV_NODEINFO_KEY_VERSION]
8348 dst_version = dst_info[constants.HV_NODEINFO_KEY_VERSION]
8349 if src_version != dst_version:
8350 self.feedback_fn("* warning: hypervisor version mismatch between"
8351 " source (%s) and target (%s) node" %
8352 (src_version, dst_version))
8354 self.feedback_fn("* checking disk consistency between source and target")
8355 for (idx, dev) in enumerate(instance.disks):
8356 if not _CheckDiskConsistency(self.lu, instance, dev, target_node, False):
8357 raise errors.OpExecError("Disk %s is degraded or not fully"
8358 " synchronized on target node,"
8359 " aborting migration" % idx)
8361 if self.current_mem > self.tgt_free_mem:
8362 if not self.allow_runtime_changes:
8363 raise errors.OpExecError("Memory ballooning not allowed and not enough"
8364 " free memory to fit instance %s on target"
8365 " node %s (have %dMB, need %dMB)" %
8366 (instance.name, target_node,
8367 self.tgt_free_mem, self.current_mem))
8368 self.feedback_fn("* setting instance memory to %s" % self.tgt_free_mem)
8369 rpcres = self.rpc.call_instance_balloon_memory(instance.primary_node,
8372 rpcres.Raise("Cannot modify instance runtime memory")
8374 # First get the migration information from the remote node
8375 result = self.rpc.call_migration_info(source_node, instance)
8376 msg = result.fail_msg
8378 log_err = ("Failed fetching source migration information from %s: %s" %
8380 logging.error(log_err)
8381 raise errors.OpExecError(log_err)
8383 self.migration_info = migration_info = result.payload
8385 if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
8386 # Then switch the disks to master/master mode
8387 self._EnsureSecondary(target_node)
8388 self._GoStandalone()
8389 self._GoReconnect(True)
8390 self._WaitUntilSync()
8392 self.feedback_fn("* preparing %s to accept the instance" % target_node)
8393 result = self.rpc.call_accept_instance(target_node,
8396 self.nodes_ip[target_node])
8398 msg = result.fail_msg
8400 logging.error("Instance pre-migration failed, trying to revert"
8401 " disk status: %s", msg)
8402 self.feedback_fn("Pre-migration failed, aborting")
8403 self._AbortMigration()
8404 self._RevertDiskStatus()
8405 raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
8406 (instance.name, msg))
8408 self.feedback_fn("* migrating instance to %s" % target_node)
8409 result = self.rpc.call_instance_migrate(source_node, instance,
8410 self.nodes_ip[target_node],
8412 msg = result.fail_msg
8414 logging.error("Instance migration failed, trying to revert"
8415 " disk status: %s", msg)
8416 self.feedback_fn("Migration failed, aborting")
8417 self._AbortMigration()
8418 self._RevertDiskStatus()
8419 raise errors.OpExecError("Could not migrate instance %s: %s" %
8420 (instance.name, msg))
8422 self.feedback_fn("* starting memory transfer")
8423 last_feedback = time.time()
8425 result = self.rpc.call_instance_get_migration_status(source_node,
8427 msg = result.fail_msg
8428 ms = result.payload # MigrationStatus instance
8429 if msg or (ms.status in constants.HV_MIGRATION_FAILED_STATUSES):
8430 logging.error("Instance migration failed, trying to revert"
8431 " disk status: %s", msg)
8432 self.feedback_fn("Migration failed, aborting")
8433 self._AbortMigration()
8434 self._RevertDiskStatus()
8435 raise errors.OpExecError("Could not migrate instance %s: %s" %
8436 (instance.name, msg))
8438 if result.payload.status != constants.HV_MIGRATION_ACTIVE:
8439 self.feedback_fn("* memory transfer complete")
8442 if (utils.TimeoutExpired(last_feedback,
8443 self._MIGRATION_FEEDBACK_INTERVAL) and
8444 ms.transferred_ram is not None):
8445 mem_progress = 100 * float(ms.transferred_ram) / float(ms.total_ram)
8446 self.feedback_fn("* memory transfer progress: %.2f %%" % mem_progress)
8447 last_feedback = time.time()
8449 time.sleep(self._MIGRATION_POLL_INTERVAL)
8451 result = self.rpc.call_instance_finalize_migration_src(source_node,
8455 msg = result.fail_msg
8457 logging.error("Instance migration succeeded, but finalization failed"
8458 " on the source node: %s", msg)
8459 raise errors.OpExecError("Could not finalize instance migration: %s" %
8462 instance.primary_node = target_node
8464 # distribute new instance config to the other nodes
8465 self.cfg.Update(instance, self.feedback_fn)
8467 result = self.rpc.call_instance_finalize_migration_dst(target_node,
8471 msg = result.fail_msg
8473 logging.error("Instance migration succeeded, but finalization failed"
8474 " on the target node: %s", msg)
8475 raise errors.OpExecError("Could not finalize instance migration: %s" %
8478 if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
8479 self._EnsureSecondary(source_node)
8480 self._WaitUntilSync()
8481 self._GoStandalone()
8482 self._GoReconnect(False)
8483 self._WaitUntilSync()
8485 # If the instance's disk template is `rbd' and there was a successful
8486 # migration, unmap the device from the source node.
8487 if self.instance.disk_template == constants.DT_RBD:
8488 disks = _ExpandCheckDisks(instance, instance.disks)
8489 self.feedback_fn("* unmapping instance's disks from %s" % source_node)
8491 result = self.rpc.call_blockdev_shutdown(source_node, disk)
8492 msg = result.fail_msg
8494 logging.error("Migration was successful, but couldn't unmap the"
8495 " block device %s on source node %s: %s",
8496 disk.iv_name, source_node, msg)
8497 logging.error("You need to unmap the device %s manually on %s",
8498 disk.iv_name, source_node)
8500 self.feedback_fn("* done")
8502 def _ExecFailover(self):
8503 """Failover an instance.
8505 The failover is done by shutting it down on its present node and
8506 starting it on the secondary.
8509 instance = self.instance
8510 primary_node = self.cfg.GetNodeInfo(instance.primary_node)
8512 source_node = instance.primary_node
8513 target_node = self.target_node
8515 if instance.admin_state == constants.ADMINST_UP:
8516 self.feedback_fn("* checking disk consistency between source and target")
8517 for (idx, dev) in enumerate(instance.disks):
8518 # for drbd, these are drbd over lvm
8519 if not _CheckDiskConsistency(self.lu, instance, dev, target_node,
8521 if primary_node.offline:
8522 self.feedback_fn("Node %s is offline, ignoring degraded disk %s on"
8524 (primary_node.name, idx, target_node))
8525 elif not self.ignore_consistency:
8526 raise errors.OpExecError("Disk %s is degraded on target node,"
8527 " aborting failover" % idx)
8529 self.feedback_fn("* not checking disk consistency as instance is not"
8532 self.feedback_fn("* shutting down instance on source node")
8533 logging.info("Shutting down instance %s on node %s",
8534 instance.name, source_node)
8536 result = self.rpc.call_instance_shutdown(source_node, instance,
8537 self.shutdown_timeout)
8538 msg = result.fail_msg
8540 if self.ignore_consistency or primary_node.offline:
8541 self.lu.LogWarning("Could not shutdown instance %s on node %s,"
8542 " proceeding anyway; please make sure node"
8543 " %s is down; error details: %s",
8544 instance.name, source_node, source_node, msg)
8546 raise errors.OpExecError("Could not shutdown instance %s on"
8548 (instance.name, source_node, msg))
8550 self.feedback_fn("* deactivating the instance's disks on source node")
8551 if not _ShutdownInstanceDisks(self.lu, instance, ignore_primary=True):
8552 raise errors.OpExecError("Can't shut down the instance's disks")
8554 instance.primary_node = target_node
8555 # distribute new instance config to the other nodes
8556 self.cfg.Update(instance, self.feedback_fn)
8558 # Only start the instance if it's marked as up
8559 if instance.admin_state == constants.ADMINST_UP:
8560 self.feedback_fn("* activating the instance's disks on target node %s" %
8562 logging.info("Starting instance %s on node %s",
8563 instance.name, target_node)
8565 disks_ok, _ = _AssembleInstanceDisks(self.lu, instance,
8566 ignore_secondaries=True)
8568 _ShutdownInstanceDisks(self.lu, instance)
8569 raise errors.OpExecError("Can't activate the instance's disks")
8571 self.feedback_fn("* starting the instance on the target node %s" %
8573 result = self.rpc.call_instance_start(target_node, (instance, None, None),
8575 msg = result.fail_msg
8577 _ShutdownInstanceDisks(self.lu, instance)
8578 raise errors.OpExecError("Could not start instance %s on node %s: %s" %
8579 (instance.name, target_node, msg))
8581 def Exec(self, feedback_fn):
8582 """Perform the migration.
8585 self.feedback_fn = feedback_fn
8586 self.source_node = self.instance.primary_node
8588 # FIXME: if we implement migrate-to-any in DRBD, this needs fixing
8589 if self.instance.disk_template in constants.DTS_INT_MIRROR:
8590 self.target_node = self.instance.secondary_nodes[0]
8591 # Otherwise self.target_node has been populated either
8592 # directly, or through an iallocator.
8594 self.all_nodes = [self.source_node, self.target_node]
8595 self.nodes_ip = dict((name, node.secondary_ip) for (name, node)
8596 in self.cfg.GetMultiNodeInfo(self.all_nodes))
8599 feedback_fn("Failover instance %s" % self.instance.name)
8600 self._ExecFailover()
8602 feedback_fn("Migrating instance %s" % self.instance.name)
8605 return self._ExecCleanup()
8607 return self._ExecMigration()
8610 def _CreateBlockDev(lu, node, instance, device, force_create, info,
8612 """Wrapper around L{_CreateBlockDevInner}.
8614 This method annotates the root device first.
8617 (disk,) = _AnnotateDiskParams(instance, [device], lu.cfg)
8618 return _CreateBlockDevInner(lu, node, instance, disk, force_create, info,
8622 def _CreateBlockDevInner(lu, node, instance, device, force_create,
8624 """Create a tree of block devices on a given node.
8626 If this device type has to be created on secondaries, create it and
8629 If not, just recurse to children keeping the same 'force' value.
8631 @attention: The device has to be annotated already.
8633 @param lu: the lu on whose behalf we execute
8634 @param node: the node on which to create the device
8635 @type instance: L{objects.Instance}
8636 @param instance: the instance which owns the device
8637 @type device: L{objects.Disk}
8638 @param device: the device to create
8639 @type force_create: boolean
8640 @param force_create: whether to force creation of this device; this
8641 will be change to True whenever we find a device which has
8642 CreateOnSecondary() attribute
8643 @param info: the extra 'metadata' we should attach to the device
8644 (this will be represented as a LVM tag)
8645 @type force_open: boolean
8646 @param force_open: this parameter will be passes to the
8647 L{backend.BlockdevCreate} function where it specifies
8648 whether we run on primary or not, and it affects both
8649 the child assembly and the device own Open() execution
8652 if device.CreateOnSecondary():
8656 for child in device.children:
8657 _CreateBlockDevInner(lu, node, instance, child, force_create,
8660 if not force_create:
8663 _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
8666 def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
8667 """Create a single block device on a given node.
8669 This will not recurse over children of the device, so they must be
8672 @param lu: the lu on whose behalf we execute
8673 @param node: the node on which to create the device
8674 @type instance: L{objects.Instance}
8675 @param instance: the instance which owns the device
8676 @type device: L{objects.Disk}
8677 @param device: the device to create
8678 @param info: the extra 'metadata' we should attach to the device
8679 (this will be represented as a LVM tag)
8680 @type force_open: boolean
8681 @param force_open: this parameter will be passes to the
8682 L{backend.BlockdevCreate} function where it specifies
8683 whether we run on primary or not, and it affects both
8684 the child assembly and the device own Open() execution
8687 lu.cfg.SetDiskID(device, node)
8688 result = lu.rpc.call_blockdev_create(node, device, device.size,
8689 instance.name, force_open, info)
8690 result.Raise("Can't create block device %s on"
8691 " node %s for instance %s" % (device, node, instance.name))
8692 if device.physical_id is None:
8693 device.physical_id = result.payload
8696 def _GenerateUniqueNames(lu, exts):
8697 """Generate a suitable LV name.
8699 This will generate a logical volume name for the given instance.
8704 new_id = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
8705 results.append("%s%s" % (new_id, val))
8709 def _GenerateDRBD8Branch(lu, primary, secondary, size, vgnames, names,
8710 iv_name, p_minor, s_minor):
8711 """Generate a drbd8 device complete with its children.
8714 assert len(vgnames) == len(names) == 2
8715 port = lu.cfg.AllocatePort()
8716 shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
8718 dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
8719 logical_id=(vgnames[0], names[0]),
8721 dev_meta = objects.Disk(dev_type=constants.LD_LV, size=DRBD_META_SIZE,
8722 logical_id=(vgnames[1], names[1]),
8724 drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
8725 logical_id=(primary, secondary, port,
8728 children=[dev_data, dev_meta],
8729 iv_name=iv_name, params={})
8733 _DISK_TEMPLATE_NAME_PREFIX = {
8734 constants.DT_PLAIN: "",
8735 constants.DT_RBD: ".rbd",
8739 _DISK_TEMPLATE_DEVICE_TYPE = {
8740 constants.DT_PLAIN: constants.LD_LV,
8741 constants.DT_FILE: constants.LD_FILE,
8742 constants.DT_SHARED_FILE: constants.LD_FILE,
8743 constants.DT_BLOCK: constants.LD_BLOCKDEV,
8744 constants.DT_RBD: constants.LD_RBD,
8748 def _GenerateDiskTemplate(lu, template_name, instance_name, primary_node,
8749 secondary_nodes, disk_info, file_storage_dir, file_driver, base_index,
8750 feedback_fn, full_disk_params, _req_file_storage=opcodes.RequireFileStorage,
8751 _req_shr_file_storage=opcodes.RequireSharedFileStorage):
8752 """Generate the entire disk layout for a given template type.
8755 #TODO: compute space requirements
8757 vgname = lu.cfg.GetVGName()
8758 disk_count = len(disk_info)
8761 if template_name == constants.DT_DISKLESS:
8763 elif template_name == constants.DT_DRBD8:
8764 if len(secondary_nodes) != 1:
8765 raise errors.ProgrammerError("Wrong template configuration")
8766 remote_node = secondary_nodes[0]
8767 minors = lu.cfg.AllocateDRBDMinor(
8768 [primary_node, remote_node] * len(disk_info), instance_name)
8770 (drbd_params, _, _) = objects.Disk.ComputeLDParams(template_name,
8772 drbd_default_metavg = drbd_params[constants.LDP_DEFAULT_METAVG]
8775 for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
8776 for i in range(disk_count)]):
8777 names.append(lv_prefix + "_data")
8778 names.append(lv_prefix + "_meta")
8779 for idx, disk in enumerate(disk_info):
8780 disk_index = idx + base_index
8781 data_vg = disk.get(constants.IDISK_VG, vgname)
8782 meta_vg = disk.get(constants.IDISK_METAVG, drbd_default_metavg)
8783 disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
8784 disk[constants.IDISK_SIZE],
8786 names[idx * 2:idx * 2 + 2],
8787 "disk/%d" % disk_index,
8788 minors[idx * 2], minors[idx * 2 + 1])
8789 disk_dev.mode = disk[constants.IDISK_MODE]
8790 disks.append(disk_dev)
8793 raise errors.ProgrammerError("Wrong template configuration")
8795 if template_name == constants.DT_FILE:
8797 elif template_name == constants.DT_SHARED_FILE:
8798 _req_shr_file_storage()
8800 name_prefix = _DISK_TEMPLATE_NAME_PREFIX.get(template_name, None)
8801 if name_prefix is None:
8804 names = _GenerateUniqueNames(lu, ["%s.disk%s" %
8805 (name_prefix, base_index + i)
8806 for i in range(disk_count)])
8808 if template_name == constants.DT_PLAIN:
8809 def logical_id_fn(idx, _, disk):
8810 vg = disk.get(constants.IDISK_VG, vgname)
8811 return (vg, names[idx])
8812 elif template_name in (constants.DT_FILE, constants.DT_SHARED_FILE):
8814 lambda _, disk_index, disk: (file_driver,
8815 "%s/disk%d" % (file_storage_dir,
8817 elif template_name == constants.DT_BLOCK:
8819 lambda idx, disk_index, disk: (constants.BLOCKDEV_DRIVER_MANUAL,
8820 disk[constants.IDISK_ADOPT])
8821 elif template_name == constants.DT_RBD:
8822 logical_id_fn = lambda idx, _, disk: ("rbd", names[idx])
8824 raise errors.ProgrammerError("Unknown disk template '%s'" % template_name)
8826 dev_type = _DISK_TEMPLATE_DEVICE_TYPE[template_name]
8828 for idx, disk in enumerate(disk_info):
8829 disk_index = idx + base_index
8830 size = disk[constants.IDISK_SIZE]
8831 feedback_fn("* disk %s, size %s" %
8832 (disk_index, utils.FormatUnit(size, "h")))
8833 disks.append(objects.Disk(dev_type=dev_type, size=size,
8834 logical_id=logical_id_fn(idx, disk_index, disk),
8835 iv_name="disk/%d" % disk_index,
8836 mode=disk[constants.IDISK_MODE],
8842 def _GetInstanceInfoText(instance):
8843 """Compute that text that should be added to the disk's metadata.
8846 return "originstname+%s" % instance.name
8849 def _CalcEta(time_taken, written, total_size):
8850 """Calculates the ETA based on size written and total size.
8852 @param time_taken: The time taken so far
8853 @param written: amount written so far
8854 @param total_size: The total size of data to be written
8855 @return: The remaining time in seconds
8858 avg_time = time_taken / float(written)
8859 return (total_size - written) * avg_time
8862 def _WipeDisks(lu, instance):
8863 """Wipes instance disks.
8865 @type lu: L{LogicalUnit}
8866 @param lu: the logical unit on whose behalf we execute
8867 @type instance: L{objects.Instance}
8868 @param instance: the instance whose disks we should create
8869 @return: the success of the wipe
8872 node = instance.primary_node
8874 for device in instance.disks:
8875 lu.cfg.SetDiskID(device, node)
8877 logging.info("Pause sync of instance %s disks", instance.name)
8878 result = lu.rpc.call_blockdev_pause_resume_sync(node,
8879 (instance.disks, instance),
8882 for idx, success in enumerate(result.payload):
8884 logging.warn("pause-sync of instance %s for disks %d failed",
8888 for idx, device in enumerate(instance.disks):
8889 # The wipe size is MIN_WIPE_CHUNK_PERCENT % of the instance disk but
8890 # MAX_WIPE_CHUNK at max
8891 wipe_chunk_size = min(constants.MAX_WIPE_CHUNK, device.size / 100.0 *
8892 constants.MIN_WIPE_CHUNK_PERCENT)
8893 # we _must_ make this an int, otherwise rounding errors will
8895 wipe_chunk_size = int(wipe_chunk_size)
8897 lu.LogInfo("* Wiping disk %d", idx)
8898 logging.info("Wiping disk %d for instance %s, node %s using"
8899 " chunk size %s", idx, instance.name, node, wipe_chunk_size)
8904 start_time = time.time()
8906 while offset < size:
8907 wipe_size = min(wipe_chunk_size, size - offset)
8908 logging.debug("Wiping disk %d, offset %s, chunk %s",
8909 idx, offset, wipe_size)
8910 result = lu.rpc.call_blockdev_wipe(node, (device, instance), offset,
8912 result.Raise("Could not wipe disk %d at offset %d for size %d" %
8913 (idx, offset, wipe_size))
8916 if now - last_output >= 60:
8917 eta = _CalcEta(now - start_time, offset, size)
8918 lu.LogInfo(" - done: %.1f%% ETA: %s" %
8919 (offset / float(size) * 100, utils.FormatSeconds(eta)))
8922 logging.info("Resume sync of instance %s disks", instance.name)
8924 result = lu.rpc.call_blockdev_pause_resume_sync(node,
8925 (instance.disks, instance),
8928 for idx, success in enumerate(result.payload):
8930 lu.LogWarning("Resume sync of disk %d failed, please have a"
8931 " look at the status and troubleshoot the issue", idx)
8932 logging.warn("resume-sync of instance %s for disks %d failed",
8936 def _CreateDisks(lu, instance, to_skip=None, target_node=None):
8937 """Create all disks for an instance.
8939 This abstracts away some work from AddInstance.
8941 @type lu: L{LogicalUnit}
8942 @param lu: the logical unit on whose behalf we execute
8943 @type instance: L{objects.Instance}
8944 @param instance: the instance whose disks we should create
8946 @param to_skip: list of indices to skip
8947 @type target_node: string
8948 @param target_node: if passed, overrides the target node for creation
8950 @return: the success of the creation
8953 info = _GetInstanceInfoText(instance)
8954 if target_node is None:
8955 pnode = instance.primary_node
8956 all_nodes = instance.all_nodes
8961 if instance.disk_template in constants.DTS_FILEBASED:
8962 file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
8963 result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
8965 result.Raise("Failed to create directory '%s' on"
8966 " node %s" % (file_storage_dir, pnode))
8968 # Note: this needs to be kept in sync with adding of disks in
8969 # LUInstanceSetParams
8970 for idx, device in enumerate(instance.disks):
8971 if to_skip and idx in to_skip:
8973 logging.info("Creating disk %s for instance '%s'", idx, instance.name)
8975 for node in all_nodes:
8976 f_create = node == pnode
8977 _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
8980 def _RemoveDisks(lu, instance, target_node=None, ignore_failures=False):
8981 """Remove all disks for an instance.
8983 This abstracts away some work from `AddInstance()` and
8984 `RemoveInstance()`. Note that in case some of the devices couldn't
8985 be removed, the removal will continue with the other ones (compare
8986 with `_CreateDisks()`).
8988 @type lu: L{LogicalUnit}
8989 @param lu: the logical unit on whose behalf we execute
8990 @type instance: L{objects.Instance}
8991 @param instance: the instance whose disks we should remove
8992 @type target_node: string
8993 @param target_node: used to override the node on which to remove the disks
8995 @return: the success of the removal
8998 logging.info("Removing block devices for instance %s", instance.name)
9001 ports_to_release = set()
9002 for (idx, device) in enumerate(instance.disks):
9004 edata = [(target_node, device)]
9006 edata = device.ComputeNodeTree(instance.primary_node)
9007 for node, disk in edata:
9008 lu.cfg.SetDiskID(disk, node)
9009 msg = lu.rpc.call_blockdev_remove(node, disk).fail_msg
9011 lu.LogWarning("Could not remove disk %s on node %s,"
9012 " continuing anyway: %s", idx, node, msg)
9015 # if this is a DRBD disk, return its port to the pool
9016 if device.dev_type in constants.LDS_DRBD:
9017 ports_to_release.add(device.logical_id[2])
9019 if all_result or ignore_failures:
9020 for port in ports_to_release:
9021 lu.cfg.AddTcpUdpPort(port)
9023 if instance.disk_template == constants.DT_FILE:
9024 file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
9028 tgt = instance.primary_node
9029 result = lu.rpc.call_file_storage_dir_remove(tgt, file_storage_dir)
9031 lu.LogWarning("Could not remove directory '%s' on node %s: %s",
9032 file_storage_dir, instance.primary_node, result.fail_msg)
9038 def _ComputeDiskSizePerVG(disk_template, disks):
9039 """Compute disk size requirements in the volume group
9042 def _compute(disks, payload):
9043 """Universal algorithm.
9048 vgs[disk[constants.IDISK_VG]] = \
9049 vgs.get(constants.IDISK_VG, 0) + disk[constants.IDISK_SIZE] + payload
9053 # Required free disk space as a function of disk and swap space
9055 constants.DT_DISKLESS: {},
9056 constants.DT_PLAIN: _compute(disks, 0),
9057 # 128 MB are added for drbd metadata for each disk
9058 constants.DT_DRBD8: _compute(disks, DRBD_META_SIZE),
9059 constants.DT_FILE: {},
9060 constants.DT_SHARED_FILE: {},
9063 if disk_template not in req_size_dict:
9064 raise errors.ProgrammerError("Disk template '%s' size requirement"
9065 " is unknown" % disk_template)
9067 return req_size_dict[disk_template]
9070 def _ComputeDiskSize(disk_template, disks):
9071 """Compute disk size requirements in the volume group
9074 # Required free disk space as a function of disk and swap space
9076 constants.DT_DISKLESS: None,
9077 constants.DT_PLAIN: sum(d[constants.IDISK_SIZE] for d in disks),
9078 # 128 MB are added for drbd metadata for each disk
9080 sum(d[constants.IDISK_SIZE] + DRBD_META_SIZE for d in disks),
9081 constants.DT_FILE: None,
9082 constants.DT_SHARED_FILE: 0,
9083 constants.DT_BLOCK: 0,
9084 constants.DT_RBD: 0,
9087 if disk_template not in req_size_dict:
9088 raise errors.ProgrammerError("Disk template '%s' size requirement"
9089 " is unknown" % disk_template)
9091 return req_size_dict[disk_template]
9094 def _FilterVmNodes(lu, nodenames):
9095 """Filters out non-vm_capable nodes from a list.
9097 @type lu: L{LogicalUnit}
9098 @param lu: the logical unit for which we check
9099 @type nodenames: list
9100 @param nodenames: the list of nodes on which we should check
9102 @return: the list of vm-capable nodes
9105 vm_nodes = frozenset(lu.cfg.GetNonVmCapableNodeList())
9106 return [name for name in nodenames if name not in vm_nodes]
9109 def _CheckHVParams(lu, nodenames, hvname, hvparams):
9110 """Hypervisor parameter validation.
9112 This function abstract the hypervisor parameter validation to be
9113 used in both instance create and instance modify.
9115 @type lu: L{LogicalUnit}
9116 @param lu: the logical unit for which we check
9117 @type nodenames: list
9118 @param nodenames: the list of nodes on which we should check
9119 @type hvname: string
9120 @param hvname: the name of the hypervisor we should use
9121 @type hvparams: dict
9122 @param hvparams: the parameters which we need to check
9123 @raise errors.OpPrereqError: if the parameters are not valid
9126 nodenames = _FilterVmNodes(lu, nodenames)
9128 cluster = lu.cfg.GetClusterInfo()
9129 hvfull = objects.FillDict(cluster.hvparams.get(hvname, {}), hvparams)
9131 hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames, hvname, hvfull)
9132 for node in nodenames:
9136 info.Raise("Hypervisor parameter validation failed on node %s" % node)
9139 def _CheckOSParams(lu, required, nodenames, osname, osparams):
9140 """OS parameters validation.
9142 @type lu: L{LogicalUnit}
9143 @param lu: the logical unit for which we check
9144 @type required: boolean
9145 @param required: whether the validation should fail if the OS is not
9147 @type nodenames: list
9148 @param nodenames: the list of nodes on which we should check
9149 @type osname: string
9150 @param osname: the name of the hypervisor we should use
9151 @type osparams: dict
9152 @param osparams: the parameters which we need to check
9153 @raise errors.OpPrereqError: if the parameters are not valid
9156 nodenames = _FilterVmNodes(lu, nodenames)
9157 result = lu.rpc.call_os_validate(nodenames, required, osname,
9158 [constants.OS_VALIDATE_PARAMETERS],
9160 for node, nres in result.items():
9161 # we don't check for offline cases since this should be run only
9162 # against the master node and/or an instance's nodes
9163 nres.Raise("OS Parameters validation failed on node %s" % node)
9164 if not nres.payload:
9165 lu.LogInfo("OS %s not found on node %s, validation skipped",
9169 class LUInstanceCreate(LogicalUnit):
9170 """Create an instance.
9173 HPATH = "instance-add"
9174 HTYPE = constants.HTYPE_INSTANCE
9177 def CheckArguments(self):
9181 # do not require name_check to ease forward/backward compatibility
9183 if self.op.no_install and self.op.start:
9184 self.LogInfo("No-installation mode selected, disabling startup")
9185 self.op.start = False
9186 # validate/normalize the instance name
9187 self.op.instance_name = \
9188 netutils.Hostname.GetNormalizedName(self.op.instance_name)
9190 if self.op.ip_check and not self.op.name_check:
9191 # TODO: make the ip check more flexible and not depend on the name check
9192 raise errors.OpPrereqError("Cannot do IP address check without a name"
9193 " check", errors.ECODE_INVAL)
9195 # check nics' parameter names
9196 for nic in self.op.nics:
9197 utils.ForceDictType(nic, constants.INIC_PARAMS_TYPES)
9199 # check disks. parameter names and consistent adopt/no-adopt strategy
9200 has_adopt = has_no_adopt = False
9201 for disk in self.op.disks:
9202 utils.ForceDictType(disk, constants.IDISK_PARAMS_TYPES)
9203 if constants.IDISK_ADOPT in disk:
9207 if has_adopt and has_no_adopt:
9208 raise errors.OpPrereqError("Either all disks are adopted or none is",
9211 if self.op.disk_template not in constants.DTS_MAY_ADOPT:
9212 raise errors.OpPrereqError("Disk adoption is not supported for the"
9213 " '%s' disk template" %
9214 self.op.disk_template,
9216 if self.op.iallocator is not None:
9217 raise errors.OpPrereqError("Disk adoption not allowed with an"
9218 " iallocator script", errors.ECODE_INVAL)
9219 if self.op.mode == constants.INSTANCE_IMPORT:
9220 raise errors.OpPrereqError("Disk adoption not allowed for"
9221 " instance import", errors.ECODE_INVAL)
9223 if self.op.disk_template in constants.DTS_MUST_ADOPT:
9224 raise errors.OpPrereqError("Disk template %s requires disk adoption,"
9225 " but no 'adopt' parameter given" %
9226 self.op.disk_template,
9229 self.adopt_disks = has_adopt
9231 # instance name verification
9232 if self.op.name_check:
9233 self.hostname1 = netutils.GetHostname(name=self.op.instance_name)
9234 self.op.instance_name = self.hostname1.name
9235 # used in CheckPrereq for ip ping check
9236 self.check_ip = self.hostname1.ip
9238 self.check_ip = None
9240 # file storage checks
9241 if (self.op.file_driver and
9242 not self.op.file_driver in constants.FILE_DRIVER):
9243 raise errors.OpPrereqError("Invalid file driver name '%s'" %
9244 self.op.file_driver, errors.ECODE_INVAL)
9246 if self.op.disk_template == constants.DT_FILE:
9247 opcodes.RequireFileStorage()
9248 elif self.op.disk_template == constants.DT_SHARED_FILE:
9249 opcodes.RequireSharedFileStorage()
9251 ### Node/iallocator related checks
9252 _CheckIAllocatorOrNode(self, "iallocator", "pnode")
9254 if self.op.pnode is not None:
9255 if self.op.disk_template in constants.DTS_INT_MIRROR:
9256 if self.op.snode is None:
9257 raise errors.OpPrereqError("The networked disk templates need"
9258 " a mirror node", errors.ECODE_INVAL)
9260 self.LogWarning("Secondary node will be ignored on non-mirrored disk"
9262 self.op.snode = None
9264 self._cds = _GetClusterDomainSecret()
9266 if self.op.mode == constants.INSTANCE_IMPORT:
9267 # On import force_variant must be True, because if we forced it at
9268 # initial install, our only chance when importing it back is that it
9270 self.op.force_variant = True
9272 if self.op.no_install:
9273 self.LogInfo("No-installation mode has no effect during import")
9275 elif self.op.mode == constants.INSTANCE_CREATE:
9276 if self.op.os_type is None:
9277 raise errors.OpPrereqError("No guest OS specified",
9279 if self.op.os_type in self.cfg.GetClusterInfo().blacklisted_os:
9280 raise errors.OpPrereqError("Guest OS '%s' is not allowed for"
9281 " installation" % self.op.os_type,
9283 if self.op.disk_template is None:
9284 raise errors.OpPrereqError("No disk template specified",
9287 elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
9288 # Check handshake to ensure both clusters have the same domain secret
9289 src_handshake = self.op.source_handshake
9290 if not src_handshake:
9291 raise errors.OpPrereqError("Missing source handshake",
9294 errmsg = masterd.instance.CheckRemoteExportHandshake(self._cds,
9297 raise errors.OpPrereqError("Invalid handshake: %s" % errmsg,
9300 # Load and check source CA
9301 self.source_x509_ca_pem = self.op.source_x509_ca
9302 if not self.source_x509_ca_pem:
9303 raise errors.OpPrereqError("Missing source X509 CA",
9307 (cert, _) = utils.LoadSignedX509Certificate(self.source_x509_ca_pem,
9309 except OpenSSL.crypto.Error, err:
9310 raise errors.OpPrereqError("Unable to load source X509 CA (%s)" %
9311 (err, ), errors.ECODE_INVAL)
9313 (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
9314 if errcode is not None:
9315 raise errors.OpPrereqError("Invalid source X509 CA (%s)" % (msg, ),
9318 self.source_x509_ca = cert
9320 src_instance_name = self.op.source_instance_name
9321 if not src_instance_name:
9322 raise errors.OpPrereqError("Missing source instance name",
9325 self.source_instance_name = \
9326 netutils.GetHostname(name=src_instance_name).name
9329 raise errors.OpPrereqError("Invalid instance creation mode %r" %
9330 self.op.mode, errors.ECODE_INVAL)
9332 def ExpandNames(self):
9333 """ExpandNames for CreateInstance.
9335 Figure out the right locks for instance creation.
9338 self.needed_locks = {}
9340 instance_name = self.op.instance_name
9341 # this is just a preventive check, but someone might still add this
9342 # instance in the meantime, and creation will fail at lock-add time
9343 if instance_name in self.cfg.GetInstanceList():
9344 raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
9345 instance_name, errors.ECODE_EXISTS)
9347 self.add_locks[locking.LEVEL_INSTANCE] = instance_name
9349 if self.op.iallocator:
9350 # TODO: Find a solution to not lock all nodes in the cluster, e.g. by
9351 # specifying a group on instance creation and then selecting nodes from
9353 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
9354 self.needed_locks[locking.LEVEL_NODE_RES] = locking.ALL_SET
9356 self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
9357 nodelist = [self.op.pnode]
9358 if self.op.snode is not None:
9359 self.op.snode = _ExpandNodeName(self.cfg, self.op.snode)
9360 nodelist.append(self.op.snode)
9361 self.needed_locks[locking.LEVEL_NODE] = nodelist
9362 # Lock resources of instance's primary and secondary nodes (copy to
9363 # prevent accidential modification)
9364 self.needed_locks[locking.LEVEL_NODE_RES] = list(nodelist)
9366 # in case of import lock the source node too
9367 if self.op.mode == constants.INSTANCE_IMPORT:
9368 src_node = self.op.src_node
9369 src_path = self.op.src_path
9371 if src_path is None:
9372 self.op.src_path = src_path = self.op.instance_name
9374 if src_node is None:
9375 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
9376 self.op.src_node = None
9377 if os.path.isabs(src_path):
9378 raise errors.OpPrereqError("Importing an instance from a path"
9379 " requires a source node option",
9382 self.op.src_node = src_node = _ExpandNodeName(self.cfg, src_node)
9383 if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
9384 self.needed_locks[locking.LEVEL_NODE].append(src_node)
9385 if not os.path.isabs(src_path):
9386 self.op.src_path = src_path = \
9387 utils.PathJoin(constants.EXPORT_DIR, src_path)
9389 def _RunAllocator(self):
9390 """Run the allocator based on input opcode.
9393 nics = [n.ToDict() for n in self.nics]
9394 ial = IAllocator(self.cfg, self.rpc,
9395 mode=constants.IALLOCATOR_MODE_ALLOC,
9396 name=self.op.instance_name,
9397 disk_template=self.op.disk_template,
9400 vcpus=self.be_full[constants.BE_VCPUS],
9401 memory=self.be_full[constants.BE_MAXMEM],
9402 spindle_use=self.be_full[constants.BE_SPINDLE_USE],
9405 hypervisor=self.op.hypervisor,
9408 ial.Run(self.op.iallocator)
9411 raise errors.OpPrereqError("Can't compute nodes using"
9412 " iallocator '%s': %s" %
9413 (self.op.iallocator, ial.info),
9415 if len(ial.result) != ial.required_nodes:
9416 raise errors.OpPrereqError("iallocator '%s' returned invalid number"
9417 " of nodes (%s), required %s" %
9418 (self.op.iallocator, len(ial.result),
9419 ial.required_nodes), errors.ECODE_FAULT)
9420 self.op.pnode = ial.result[0]
9421 self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
9422 self.op.instance_name, self.op.iallocator,
9423 utils.CommaJoin(ial.result))
9424 if ial.required_nodes == 2:
9425 self.op.snode = ial.result[1]
9427 def BuildHooksEnv(self):
9430 This runs on master, primary and secondary nodes of the instance.
9434 "ADD_MODE": self.op.mode,
9436 if self.op.mode == constants.INSTANCE_IMPORT:
9437 env["SRC_NODE"] = self.op.src_node
9438 env["SRC_PATH"] = self.op.src_path
9439 env["SRC_IMAGES"] = self.src_images
9441 env.update(_BuildInstanceHookEnv(
9442 name=self.op.instance_name,
9443 primary_node=self.op.pnode,
9444 secondary_nodes=self.secondaries,
9445 status=self.op.start,
9446 os_type=self.op.os_type,
9447 minmem=self.be_full[constants.BE_MINMEM],
9448 maxmem=self.be_full[constants.BE_MAXMEM],
9449 vcpus=self.be_full[constants.BE_VCPUS],
9450 nics=_NICListToTuple(self, self.nics),
9451 disk_template=self.op.disk_template,
9452 disks=[(d[constants.IDISK_SIZE], d[constants.IDISK_MODE])
9453 for d in self.disks],
9456 hypervisor_name=self.op.hypervisor,
9462 def BuildHooksNodes(self):
9463 """Build hooks nodes.
9466 nl = [self.cfg.GetMasterNode(), self.op.pnode] + self.secondaries
9469 def _ReadExportInfo(self):
9470 """Reads the export information from disk.
9472 It will override the opcode source node and path with the actual
9473 information, if these two were not specified before.
9475 @return: the export information
9478 assert self.op.mode == constants.INSTANCE_IMPORT
9480 src_node = self.op.src_node
9481 src_path = self.op.src_path
9483 if src_node is None:
9484 locked_nodes = self.owned_locks(locking.LEVEL_NODE)
9485 exp_list = self.rpc.call_export_list(locked_nodes)
9487 for node in exp_list:
9488 if exp_list[node].fail_msg:
9490 if src_path in exp_list[node].payload:
9492 self.op.src_node = src_node = node
9493 self.op.src_path = src_path = utils.PathJoin(constants.EXPORT_DIR,
9497 raise errors.OpPrereqError("No export found for relative path %s" %
9498 src_path, errors.ECODE_INVAL)
9500 _CheckNodeOnline(self, src_node)
9501 result = self.rpc.call_export_info(src_node, src_path)
9502 result.Raise("No export or invalid export found in dir %s" % src_path)
9504 export_info = objects.SerializableConfigParser.Loads(str(result.payload))
9505 if not export_info.has_section(constants.INISECT_EXP):
9506 raise errors.ProgrammerError("Corrupted export config",
9507 errors.ECODE_ENVIRON)
9509 ei_version = export_info.get(constants.INISECT_EXP, "version")
9510 if (int(ei_version) != constants.EXPORT_VERSION):
9511 raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
9512 (ei_version, constants.EXPORT_VERSION),
9513 errors.ECODE_ENVIRON)
9516 def _ReadExportParams(self, einfo):
9517 """Use export parameters as defaults.
9519 In case the opcode doesn't specify (as in override) some instance
9520 parameters, then try to use them from the export information, if
9524 self.op.os_type = einfo.get(constants.INISECT_EXP, "os")
9526 if self.op.disk_template is None:
9527 if einfo.has_option(constants.INISECT_INS, "disk_template"):
9528 self.op.disk_template = einfo.get(constants.INISECT_INS,
9530 if self.op.disk_template not in constants.DISK_TEMPLATES:
9531 raise errors.OpPrereqError("Disk template specified in configuration"
9532 " file is not one of the allowed values:"
9533 " %s" % " ".join(constants.DISK_TEMPLATES))
9535 raise errors.OpPrereqError("No disk template specified and the export"
9536 " is missing the disk_template information",
9539 if not self.op.disks:
9541 # TODO: import the disk iv_name too
9542 for idx in range(constants.MAX_DISKS):
9543 if einfo.has_option(constants.INISECT_INS, "disk%d_size" % idx):
9544 disk_sz = einfo.getint(constants.INISECT_INS, "disk%d_size" % idx)
9545 disks.append({constants.IDISK_SIZE: disk_sz})
9546 self.op.disks = disks
9547 if not disks and self.op.disk_template != constants.DT_DISKLESS:
9548 raise errors.OpPrereqError("No disk info specified and the export"
9549 " is missing the disk information",
9552 if not self.op.nics:
9554 for idx in range(constants.MAX_NICS):
9555 if einfo.has_option(constants.INISECT_INS, "nic%d_mac" % idx):
9557 for name in list(constants.NICS_PARAMETERS) + ["ip", "mac"]:
9558 v = einfo.get(constants.INISECT_INS, "nic%d_%s" % (idx, name))
9565 if not self.op.tags and einfo.has_option(constants.INISECT_INS, "tags"):
9566 self.op.tags = einfo.get(constants.INISECT_INS, "tags").split()
9568 if (self.op.hypervisor is None and
9569 einfo.has_option(constants.INISECT_INS, "hypervisor")):
9570 self.op.hypervisor = einfo.get(constants.INISECT_INS, "hypervisor")
9572 if einfo.has_section(constants.INISECT_HYP):
9573 # use the export parameters but do not override the ones
9574 # specified by the user
9575 for name, value in einfo.items(constants.INISECT_HYP):
9576 if name not in self.op.hvparams:
9577 self.op.hvparams[name] = value
9579 if einfo.has_section(constants.INISECT_BEP):
9580 # use the parameters, without overriding
9581 for name, value in einfo.items(constants.INISECT_BEP):
9582 if name not in self.op.beparams:
9583 self.op.beparams[name] = value
9584 # Compatibility for the old "memory" be param
9585 if name == constants.BE_MEMORY:
9586 if constants.BE_MAXMEM not in self.op.beparams:
9587 self.op.beparams[constants.BE_MAXMEM] = value
9588 if constants.BE_MINMEM not in self.op.beparams:
9589 self.op.beparams[constants.BE_MINMEM] = value
9591 # try to read the parameters old style, from the main section
9592 for name in constants.BES_PARAMETERS:
9593 if (name not in self.op.beparams and
9594 einfo.has_option(constants.INISECT_INS, name)):
9595 self.op.beparams[name] = einfo.get(constants.INISECT_INS, name)
9597 if einfo.has_section(constants.INISECT_OSP):
9598 # use the parameters, without overriding
9599 for name, value in einfo.items(constants.INISECT_OSP):
9600 if name not in self.op.osparams:
9601 self.op.osparams[name] = value
9603 def _RevertToDefaults(self, cluster):
9604 """Revert the instance parameters to the default values.
9608 hv_defs = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type, {})
9609 for name in self.op.hvparams.keys():
9610 if name in hv_defs and hv_defs[name] == self.op.hvparams[name]:
9611 del self.op.hvparams[name]
9613 be_defs = cluster.SimpleFillBE({})
9614 for name in self.op.beparams.keys():
9615 if name in be_defs and be_defs[name] == self.op.beparams[name]:
9616 del self.op.beparams[name]
9618 nic_defs = cluster.SimpleFillNIC({})
9619 for nic in self.op.nics:
9620 for name in constants.NICS_PARAMETERS:
9621 if name in nic and name in nic_defs and nic[name] == nic_defs[name]:
9624 os_defs = cluster.SimpleFillOS(self.op.os_type, {})
9625 for name in self.op.osparams.keys():
9626 if name in os_defs and os_defs[name] == self.op.osparams[name]:
9627 del self.op.osparams[name]
9629 def _CalculateFileStorageDir(self):
9630 """Calculate final instance file storage dir.
9633 # file storage dir calculation/check
9634 self.instance_file_storage_dir = None
9635 if self.op.disk_template in constants.DTS_FILEBASED:
9636 # build the full file storage dir path
9639 if self.op.disk_template == constants.DT_SHARED_FILE:
9640 get_fsd_fn = self.cfg.GetSharedFileStorageDir
9642 get_fsd_fn = self.cfg.GetFileStorageDir
9644 cfg_storagedir = get_fsd_fn()
9645 if not cfg_storagedir:
9646 raise errors.OpPrereqError("Cluster file storage dir not defined")
9647 joinargs.append(cfg_storagedir)
9649 if self.op.file_storage_dir is not None:
9650 joinargs.append(self.op.file_storage_dir)
9652 joinargs.append(self.op.instance_name)
9654 # pylint: disable=W0142
9655 self.instance_file_storage_dir = utils.PathJoin(*joinargs)
9657 def CheckPrereq(self): # pylint: disable=R0914
9658 """Check prerequisites.
9661 self._CalculateFileStorageDir()
9663 if self.op.mode == constants.INSTANCE_IMPORT:
9664 export_info = self._ReadExportInfo()
9665 self._ReadExportParams(export_info)
9666 self._old_instance_name = export_info.get(constants.INISECT_INS, "name")
9668 self._old_instance_name = None
9670 if (not self.cfg.GetVGName() and
9671 self.op.disk_template not in constants.DTS_NOT_LVM):
9672 raise errors.OpPrereqError("Cluster does not support lvm-based"
9673 " instances", errors.ECODE_STATE)
9675 if (self.op.hypervisor is None or
9676 self.op.hypervisor == constants.VALUE_AUTO):
9677 self.op.hypervisor = self.cfg.GetHypervisorType()
9679 cluster = self.cfg.GetClusterInfo()
9680 enabled_hvs = cluster.enabled_hypervisors
9681 if self.op.hypervisor not in enabled_hvs:
9682 raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
9683 " cluster (%s)" % (self.op.hypervisor,
9684 ",".join(enabled_hvs)),
9687 # Check tag validity
9688 for tag in self.op.tags:
9689 objects.TaggableObject.ValidateTag(tag)
9691 # check hypervisor parameter syntax (locally)
9692 utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
9693 filled_hvp = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type,
9695 hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
9696 hv_type.CheckParameterSyntax(filled_hvp)
9697 self.hv_full = filled_hvp
9698 # check that we don't specify global parameters on an instance
9699 _CheckGlobalHvParams(self.op.hvparams)
9701 # fill and remember the beparams dict
9702 default_beparams = cluster.beparams[constants.PP_DEFAULT]
9703 for param, value in self.op.beparams.iteritems():
9704 if value == constants.VALUE_AUTO:
9705 self.op.beparams[param] = default_beparams[param]
9706 objects.UpgradeBeParams(self.op.beparams)
9707 utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
9708 self.be_full = cluster.SimpleFillBE(self.op.beparams)
9710 # build os parameters
9711 self.os_full = cluster.SimpleFillOS(self.op.os_type, self.op.osparams)
9713 # now that hvp/bep are in final format, let's reset to defaults,
9715 if self.op.identify_defaults:
9716 self._RevertToDefaults(cluster)
9720 for idx, nic in enumerate(self.op.nics):
9721 nic_mode_req = nic.get(constants.INIC_MODE, None)
9722 nic_mode = nic_mode_req
9723 if nic_mode is None or nic_mode == constants.VALUE_AUTO:
9724 nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
9726 # in routed mode, for the first nic, the default ip is 'auto'
9727 if nic_mode == constants.NIC_MODE_ROUTED and idx == 0:
9728 default_ip_mode = constants.VALUE_AUTO
9730 default_ip_mode = constants.VALUE_NONE
9732 # ip validity checks
9733 ip = nic.get(constants.INIC_IP, default_ip_mode)
9734 if ip is None or ip.lower() == constants.VALUE_NONE:
9736 elif ip.lower() == constants.VALUE_AUTO:
9737 if not self.op.name_check:
9738 raise errors.OpPrereqError("IP address set to auto but name checks"
9739 " have been skipped",
9741 nic_ip = self.hostname1.ip
9743 if not netutils.IPAddress.IsValid(ip):
9744 raise errors.OpPrereqError("Invalid IP address '%s'" % ip,
9748 # TODO: check the ip address for uniqueness
9749 if nic_mode == constants.NIC_MODE_ROUTED and not nic_ip:
9750 raise errors.OpPrereqError("Routed nic mode requires an ip address",
9753 # MAC address verification
9754 mac = nic.get(constants.INIC_MAC, constants.VALUE_AUTO)
9755 if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
9756 mac = utils.NormalizeAndValidateMac(mac)
9759 self.cfg.ReserveMAC(mac, self.proc.GetECId())
9760 except errors.ReservationError:
9761 raise errors.OpPrereqError("MAC address %s already in use"
9762 " in cluster" % mac,
9763 errors.ECODE_NOTUNIQUE)
9765 # Build nic parameters
9766 link = nic.get(constants.INIC_LINK, None)
9767 if link == constants.VALUE_AUTO:
9768 link = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_LINK]
9771 nicparams[constants.NIC_MODE] = nic_mode
9773 nicparams[constants.NIC_LINK] = link
9775 check_params = cluster.SimpleFillNIC(nicparams)
9776 objects.NIC.CheckParameterSyntax(check_params)
9777 self.nics.append(objects.NIC(mac=mac, ip=nic_ip, nicparams=nicparams))
9779 # disk checks/pre-build
9780 default_vg = self.cfg.GetVGName()
9782 for disk in self.op.disks:
9783 mode = disk.get(constants.IDISK_MODE, constants.DISK_RDWR)
9784 if mode not in constants.DISK_ACCESS_SET:
9785 raise errors.OpPrereqError("Invalid disk access mode '%s'" %
9786 mode, errors.ECODE_INVAL)
9787 size = disk.get(constants.IDISK_SIZE, None)
9789 raise errors.OpPrereqError("Missing disk size", errors.ECODE_INVAL)
9792 except (TypeError, ValueError):
9793 raise errors.OpPrereqError("Invalid disk size '%s'" % size,
9796 data_vg = disk.get(constants.IDISK_VG, default_vg)
9798 constants.IDISK_SIZE: size,
9799 constants.IDISK_MODE: mode,
9800 constants.IDISK_VG: data_vg,
9802 if constants.IDISK_METAVG in disk:
9803 new_disk[constants.IDISK_METAVG] = disk[constants.IDISK_METAVG]
9804 if constants.IDISK_ADOPT in disk:
9805 new_disk[constants.IDISK_ADOPT] = disk[constants.IDISK_ADOPT]
9806 self.disks.append(new_disk)
9808 if self.op.mode == constants.INSTANCE_IMPORT:
9810 for idx in range(len(self.disks)):
9811 option = "disk%d_dump" % idx
9812 if export_info.has_option(constants.INISECT_INS, option):
9813 # FIXME: are the old os-es, disk sizes, etc. useful?
9814 export_name = export_info.get(constants.INISECT_INS, option)
9815 image = utils.PathJoin(self.op.src_path, export_name)
9816 disk_images.append(image)
9818 disk_images.append(False)
9820 self.src_images = disk_images
9822 if self.op.instance_name == self._old_instance_name:
9823 for idx, nic in enumerate(self.nics):
9824 if nic.mac == constants.VALUE_AUTO:
9825 nic_mac_ini = "nic%d_mac" % idx
9826 nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
9828 # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
9830 # ip ping checks (we use the same ip that was resolved in ExpandNames)
9831 if self.op.ip_check:
9832 if netutils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
9833 raise errors.OpPrereqError("IP %s of instance %s already in use" %
9834 (self.check_ip, self.op.instance_name),
9835 errors.ECODE_NOTUNIQUE)
9837 #### mac address generation
9838 # By generating here the mac address both the allocator and the hooks get
9839 # the real final mac address rather than the 'auto' or 'generate' value.
9840 # There is a race condition between the generation and the instance object
9841 # creation, which means that we know the mac is valid now, but we're not
9842 # sure it will be when we actually add the instance. If things go bad
9843 # adding the instance will abort because of a duplicate mac, and the
9844 # creation job will fail.
9845 for nic in self.nics:
9846 if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
9847 nic.mac = self.cfg.GenerateMAC(self.proc.GetECId())
9851 if self.op.iallocator is not None:
9852 self._RunAllocator()
9854 # Release all unneeded node locks
9855 _ReleaseLocks(self, locking.LEVEL_NODE,
9856 keep=filter(None, [self.op.pnode, self.op.snode,
9858 _ReleaseLocks(self, locking.LEVEL_NODE_RES,
9859 keep=filter(None, [self.op.pnode, self.op.snode,
9862 #### node related checks
9864 # check primary node
9865 self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
9866 assert self.pnode is not None, \
9867 "Cannot retrieve locked node %s" % self.op.pnode
9869 raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
9870 pnode.name, errors.ECODE_STATE)
9872 raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
9873 pnode.name, errors.ECODE_STATE)
9874 if not pnode.vm_capable:
9875 raise errors.OpPrereqError("Cannot use non-vm_capable primary node"
9876 " '%s'" % pnode.name, errors.ECODE_STATE)
9878 self.secondaries = []
9880 # mirror node verification
9881 if self.op.disk_template in constants.DTS_INT_MIRROR:
9882 if self.op.snode == pnode.name:
9883 raise errors.OpPrereqError("The secondary node cannot be the"
9884 " primary node", errors.ECODE_INVAL)
9885 _CheckNodeOnline(self, self.op.snode)
9886 _CheckNodeNotDrained(self, self.op.snode)
9887 _CheckNodeVmCapable(self, self.op.snode)
9888 self.secondaries.append(self.op.snode)
9890 snode = self.cfg.GetNodeInfo(self.op.snode)
9891 if pnode.group != snode.group:
9892 self.LogWarning("The primary and secondary nodes are in two"
9893 " different node groups; the disk parameters"
9894 " from the first disk's node group will be"
9897 nodenames = [pnode.name] + self.secondaries
9899 # Verify instance specs
9900 spindle_use = self.be_full.get(constants.BE_SPINDLE_USE, None)
9902 constants.ISPEC_MEM_SIZE: self.be_full.get(constants.BE_MAXMEM, None),
9903 constants.ISPEC_CPU_COUNT: self.be_full.get(constants.BE_VCPUS, None),
9904 constants.ISPEC_DISK_COUNT: len(self.disks),
9905 constants.ISPEC_DISK_SIZE: [disk["size"] for disk in self.disks],
9906 constants.ISPEC_NIC_COUNT: len(self.nics),
9907 constants.ISPEC_SPINDLE_USE: spindle_use,
9910 group_info = self.cfg.GetNodeGroup(pnode.group)
9911 ipolicy = _CalculateGroupIPolicy(cluster, group_info)
9912 res = _ComputeIPolicyInstanceSpecViolation(ipolicy, ispec)
9913 if not self.op.ignore_ipolicy and res:
9914 raise errors.OpPrereqError(("Instance allocation to group %s violates"
9915 " policy: %s") % (pnode.group,
9916 utils.CommaJoin(res)),
9919 if not self.adopt_disks:
9920 if self.op.disk_template == constants.DT_RBD:
9921 # _CheckRADOSFreeSpace() is just a placeholder.
9922 # Any function that checks prerequisites can be placed here.
9923 # Check if there is enough space on the RADOS cluster.
9924 _CheckRADOSFreeSpace()
9926 # Check lv size requirements, if not adopting
9927 req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks)
9928 _CheckNodesFreeDiskPerVG(self, nodenames, req_sizes)
9930 elif self.op.disk_template == constants.DT_PLAIN: # Check the adoption data
9931 all_lvs = set(["%s/%s" % (disk[constants.IDISK_VG],
9932 disk[constants.IDISK_ADOPT])
9933 for disk in self.disks])
9934 if len(all_lvs) != len(self.disks):
9935 raise errors.OpPrereqError("Duplicate volume names given for adoption",
9937 for lv_name in all_lvs:
9939 # FIXME: lv_name here is "vg/lv" need to ensure that other calls
9940 # to ReserveLV uses the same syntax
9941 self.cfg.ReserveLV(lv_name, self.proc.GetECId())
9942 except errors.ReservationError:
9943 raise errors.OpPrereqError("LV named %s used by another instance" %
9944 lv_name, errors.ECODE_NOTUNIQUE)
9946 vg_names = self.rpc.call_vg_list([pnode.name])[pnode.name]
9947 vg_names.Raise("Cannot get VG information from node %s" % pnode.name)
9949 node_lvs = self.rpc.call_lv_list([pnode.name],
9950 vg_names.payload.keys())[pnode.name]
9951 node_lvs.Raise("Cannot get LV information from node %s" % pnode.name)
9952 node_lvs = node_lvs.payload
9954 delta = all_lvs.difference(node_lvs.keys())
9956 raise errors.OpPrereqError("Missing logical volume(s): %s" %
9957 utils.CommaJoin(delta),
9959 online_lvs = [lv for lv in all_lvs if node_lvs[lv][2]]
9961 raise errors.OpPrereqError("Online logical volumes found, cannot"
9962 " adopt: %s" % utils.CommaJoin(online_lvs),
9964 # update the size of disk based on what is found
9965 for dsk in self.disks:
9966 dsk[constants.IDISK_SIZE] = \
9967 int(float(node_lvs["%s/%s" % (dsk[constants.IDISK_VG],
9968 dsk[constants.IDISK_ADOPT])][0]))
9970 elif self.op.disk_template == constants.DT_BLOCK:
9971 # Normalize and de-duplicate device paths
9972 all_disks = set([os.path.abspath(disk[constants.IDISK_ADOPT])
9973 for disk in self.disks])
9974 if len(all_disks) != len(self.disks):
9975 raise errors.OpPrereqError("Duplicate disk names given for adoption",
9977 baddisks = [d for d in all_disks
9978 if not d.startswith(constants.ADOPTABLE_BLOCKDEV_ROOT)]
9980 raise errors.OpPrereqError("Device node(s) %s lie outside %s and"
9981 " cannot be adopted" %
9982 (", ".join(baddisks),
9983 constants.ADOPTABLE_BLOCKDEV_ROOT),
9986 node_disks = self.rpc.call_bdev_sizes([pnode.name],
9987 list(all_disks))[pnode.name]
9988 node_disks.Raise("Cannot get block device information from node %s" %
9990 node_disks = node_disks.payload
9991 delta = all_disks.difference(node_disks.keys())
9993 raise errors.OpPrereqError("Missing block device(s): %s" %
9994 utils.CommaJoin(delta),
9996 for dsk in self.disks:
9997 dsk[constants.IDISK_SIZE] = \
9998 int(float(node_disks[dsk[constants.IDISK_ADOPT]]))
10000 _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
10002 _CheckNodeHasOS(self, pnode.name, self.op.os_type, self.op.force_variant)
10003 # check OS parameters (remotely)
10004 _CheckOSParams(self, True, nodenames, self.op.os_type, self.os_full)
10006 _CheckNicsBridgesExist(self, self.nics, self.pnode.name)
10008 # memory check on primary node
10009 #TODO(dynmem): use MINMEM for checking
10011 _CheckNodeFreeMemory(self, self.pnode.name,
10012 "creating instance %s" % self.op.instance_name,
10013 self.be_full[constants.BE_MAXMEM],
10014 self.op.hypervisor)
10016 self.dry_run_result = list(nodenames)
10018 def Exec(self, feedback_fn):
10019 """Create and add the instance to the cluster.
10022 instance = self.op.instance_name
10023 pnode_name = self.pnode.name
10025 assert not (self.owned_locks(locking.LEVEL_NODE_RES) -
10026 self.owned_locks(locking.LEVEL_NODE)), \
10027 "Node locks differ from node resource locks"
10029 ht_kind = self.op.hypervisor
10030 if ht_kind in constants.HTS_REQ_PORT:
10031 network_port = self.cfg.AllocatePort()
10033 network_port = None
10035 # This is ugly but we got a chicken-egg problem here
10036 # We can only take the group disk parameters, as the instance
10037 # has no disks yet (we are generating them right here).
10038 node = self.cfg.GetNodeInfo(pnode_name)
10039 nodegroup = self.cfg.GetNodeGroup(node.group)
10040 disks = _GenerateDiskTemplate(self,
10041 self.op.disk_template,
10042 instance, pnode_name,
10045 self.instance_file_storage_dir,
10046 self.op.file_driver,
10049 self.cfg.GetGroupDiskParams(nodegroup))
10051 iobj = objects.Instance(name=instance, os=self.op.os_type,
10052 primary_node=pnode_name,
10053 nics=self.nics, disks=disks,
10054 disk_template=self.op.disk_template,
10055 admin_state=constants.ADMINST_DOWN,
10056 network_port=network_port,
10057 beparams=self.op.beparams,
10058 hvparams=self.op.hvparams,
10059 hypervisor=self.op.hypervisor,
10060 osparams=self.op.osparams,
10064 for tag in self.op.tags:
10067 if self.adopt_disks:
10068 if self.op.disk_template == constants.DT_PLAIN:
10069 # rename LVs to the newly-generated names; we need to construct
10070 # 'fake' LV disks with the old data, plus the new unique_id
10071 tmp_disks = [objects.Disk.FromDict(v.ToDict()) for v in disks]
10073 for t_dsk, a_dsk in zip(tmp_disks, self.disks):
10074 rename_to.append(t_dsk.logical_id)
10075 t_dsk.logical_id = (t_dsk.logical_id[0], a_dsk[constants.IDISK_ADOPT])
10076 self.cfg.SetDiskID(t_dsk, pnode_name)
10077 result = self.rpc.call_blockdev_rename(pnode_name,
10078 zip(tmp_disks, rename_to))
10079 result.Raise("Failed to rename adoped LVs")
10081 feedback_fn("* creating instance disks...")
10083 _CreateDisks(self, iobj)
10084 except errors.OpExecError:
10085 self.LogWarning("Device creation failed, reverting...")
10087 _RemoveDisks(self, iobj)
10089 self.cfg.ReleaseDRBDMinors(instance)
10092 feedback_fn("adding instance %s to cluster config" % instance)
10094 self.cfg.AddInstance(iobj, self.proc.GetECId())
10096 # Declare that we don't want to remove the instance lock anymore, as we've
10097 # added the instance to the config
10098 del self.remove_locks[locking.LEVEL_INSTANCE]
10100 if self.op.mode == constants.INSTANCE_IMPORT:
10101 # Release unused nodes
10102 _ReleaseLocks(self, locking.LEVEL_NODE, keep=[self.op.src_node])
10104 # Release all nodes
10105 _ReleaseLocks(self, locking.LEVEL_NODE)
10108 if not self.adopt_disks and self.cfg.GetClusterInfo().prealloc_wipe_disks:
10109 feedback_fn("* wiping instance disks...")
10111 _WipeDisks(self, iobj)
10112 except errors.OpExecError, err:
10113 logging.exception("Wiping disks failed")
10114 self.LogWarning("Wiping instance disks failed (%s)", err)
10118 # Something is already wrong with the disks, don't do anything else
10120 elif self.op.wait_for_sync:
10121 disk_abort = not _WaitForSync(self, iobj)
10122 elif iobj.disk_template in constants.DTS_INT_MIRROR:
10123 # make sure the disks are not degraded (still sync-ing is ok)
10124 feedback_fn("* checking mirrors status")
10125 disk_abort = not _WaitForSync(self, iobj, oneshot=True)
10130 _RemoveDisks(self, iobj)
10131 self.cfg.RemoveInstance(iobj.name)
10132 # Make sure the instance lock gets removed
10133 self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
10134 raise errors.OpExecError("There are some degraded disks for"
10137 # Release all node resource locks
10138 _ReleaseLocks(self, locking.LEVEL_NODE_RES)
10140 if iobj.disk_template != constants.DT_DISKLESS and not self.adopt_disks:
10141 # we need to set the disks ID to the primary node, since the
10142 # preceding code might or might have not done it, depending on
10143 # disk template and other options
10144 for disk in iobj.disks:
10145 self.cfg.SetDiskID(disk, pnode_name)
10146 if self.op.mode == constants.INSTANCE_CREATE:
10147 if not self.op.no_install:
10148 pause_sync = (iobj.disk_template in constants.DTS_INT_MIRROR and
10149 not self.op.wait_for_sync)
10151 feedback_fn("* pausing disk sync to install instance OS")
10152 result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
10155 for idx, success in enumerate(result.payload):
10157 logging.warn("pause-sync of instance %s for disk %d failed",
10160 feedback_fn("* running the instance OS create scripts...")
10161 # FIXME: pass debug option from opcode to backend
10163 self.rpc.call_instance_os_add(pnode_name, (iobj, None), False,
10164 self.op.debug_level)
10166 feedback_fn("* resuming disk sync")
10167 result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
10170 for idx, success in enumerate(result.payload):
10172 logging.warn("resume-sync of instance %s for disk %d failed",
10175 os_add_result.Raise("Could not add os for instance %s"
10176 " on node %s" % (instance, pnode_name))
10179 if self.op.mode == constants.INSTANCE_IMPORT:
10180 feedback_fn("* running the instance OS import scripts...")
10184 for idx, image in enumerate(self.src_images):
10188 # FIXME: pass debug option from opcode to backend
10189 dt = masterd.instance.DiskTransfer("disk/%s" % idx,
10190 constants.IEIO_FILE, (image, ),
10191 constants.IEIO_SCRIPT,
10192 (iobj.disks[idx], idx),
10194 transfers.append(dt)
10197 masterd.instance.TransferInstanceData(self, feedback_fn,
10198 self.op.src_node, pnode_name,
10199 self.pnode.secondary_ip,
10201 if not compat.all(import_result):
10202 self.LogWarning("Some disks for instance %s on node %s were not"
10203 " imported successfully" % (instance, pnode_name))
10205 rename_from = self._old_instance_name
10207 elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
10208 feedback_fn("* preparing remote import...")
10209 # The source cluster will stop the instance before attempting to make
10210 # a connection. In some cases stopping an instance can take a long
10211 # time, hence the shutdown timeout is added to the connection
10213 connect_timeout = (constants.RIE_CONNECT_TIMEOUT +
10214 self.op.source_shutdown_timeout)
10215 timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
10217 assert iobj.primary_node == self.pnode.name
10219 masterd.instance.RemoteImport(self, feedback_fn, iobj, self.pnode,
10220 self.source_x509_ca,
10221 self._cds, timeouts)
10222 if not compat.all(disk_results):
10223 # TODO: Should the instance still be started, even if some disks
10224 # failed to import (valid for local imports, too)?
10225 self.LogWarning("Some disks for instance %s on node %s were not"
10226 " imported successfully" % (instance, pnode_name))
10228 rename_from = self.source_instance_name
10231 # also checked in the prereq part
10232 raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
10235 # Run rename script on newly imported instance
10236 assert iobj.name == instance
10237 feedback_fn("Running rename script for %s" % instance)
10238 result = self.rpc.call_instance_run_rename(pnode_name, iobj,
10240 self.op.debug_level)
10241 if result.fail_msg:
10242 self.LogWarning("Failed to run rename script for %s on node"
10243 " %s: %s" % (instance, pnode_name, result.fail_msg))
10245 assert not self.owned_locks(locking.LEVEL_NODE_RES)
10248 iobj.admin_state = constants.ADMINST_UP
10249 self.cfg.Update(iobj, feedback_fn)
10250 logging.info("Starting instance %s on node %s", instance, pnode_name)
10251 feedback_fn("* starting instance...")
10252 result = self.rpc.call_instance_start(pnode_name, (iobj, None, None),
10254 result.Raise("Could not start instance")
10256 return list(iobj.all_nodes)
10259 def _CheckRADOSFreeSpace():
10260 """Compute disk size requirements inside the RADOS cluster.
10263 # For the RADOS cluster we assume there is always enough space.
10267 class LUInstanceConsole(NoHooksLU):
10268 """Connect to an instance's console.
10270 This is somewhat special in that it returns the command line that
10271 you need to run on the master node in order to connect to the
10277 def ExpandNames(self):
10278 self.share_locks = _ShareAll()
10279 self._ExpandAndLockInstance()
10281 def CheckPrereq(self):
10282 """Check prerequisites.
10284 This checks that the instance is in the cluster.
10287 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
10288 assert self.instance is not None, \
10289 "Cannot retrieve locked instance %s" % self.op.instance_name
10290 _CheckNodeOnline(self, self.instance.primary_node)
10292 def Exec(self, feedback_fn):
10293 """Connect to the console of an instance
10296 instance = self.instance
10297 node = instance.primary_node
10299 node_insts = self.rpc.call_instance_list([node],
10300 [instance.hypervisor])[node]
10301 node_insts.Raise("Can't get node information from %s" % node)
10303 if instance.name not in node_insts.payload:
10304 if instance.admin_state == constants.ADMINST_UP:
10305 state = constants.INSTST_ERRORDOWN
10306 elif instance.admin_state == constants.ADMINST_DOWN:
10307 state = constants.INSTST_ADMINDOWN
10309 state = constants.INSTST_ADMINOFFLINE
10310 raise errors.OpExecError("Instance %s is not running (state %s)" %
10311 (instance.name, state))
10313 logging.debug("Connecting to console of %s on %s", instance.name, node)
10315 return _GetInstanceConsole(self.cfg.GetClusterInfo(), instance)
10318 def _GetInstanceConsole(cluster, instance):
10319 """Returns console information for an instance.
10321 @type cluster: L{objects.Cluster}
10322 @type instance: L{objects.Instance}
10326 hyper = hypervisor.GetHypervisor(instance.hypervisor)
10327 # beparams and hvparams are passed separately, to avoid editing the
10328 # instance and then saving the defaults in the instance itself.
10329 hvparams = cluster.FillHV(instance)
10330 beparams = cluster.FillBE(instance)
10331 console = hyper.GetInstanceConsole(instance, hvparams, beparams)
10333 assert console.instance == instance.name
10334 assert console.Validate()
10336 return console.ToDict()
10339 class LUInstanceReplaceDisks(LogicalUnit):
10340 """Replace the disks of an instance.
10343 HPATH = "mirrors-replace"
10344 HTYPE = constants.HTYPE_INSTANCE
10347 def CheckArguments(self):
10348 TLReplaceDisks.CheckArguments(self.op.mode, self.op.remote_node,
10349 self.op.iallocator)
10351 def ExpandNames(self):
10352 self._ExpandAndLockInstance()
10354 assert locking.LEVEL_NODE not in self.needed_locks
10355 assert locking.LEVEL_NODE_RES not in self.needed_locks
10356 assert locking.LEVEL_NODEGROUP not in self.needed_locks
10358 assert self.op.iallocator is None or self.op.remote_node is None, \
10359 "Conflicting options"
10361 if self.op.remote_node is not None:
10362 self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
10364 # Warning: do not remove the locking of the new secondary here
10365 # unless DRBD8.AddChildren is changed to work in parallel;
10366 # currently it doesn't since parallel invocations of
10367 # FindUnusedMinor will conflict
10368 self.needed_locks[locking.LEVEL_NODE] = [self.op.remote_node]
10369 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
10371 self.needed_locks[locking.LEVEL_NODE] = []
10372 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
10374 if self.op.iallocator is not None:
10375 # iallocator will select a new node in the same group
10376 self.needed_locks[locking.LEVEL_NODEGROUP] = []
10378 self.needed_locks[locking.LEVEL_NODE_RES] = []
10380 self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
10381 self.op.iallocator, self.op.remote_node,
10382 self.op.disks, False, self.op.early_release,
10383 self.op.ignore_ipolicy)
10385 self.tasklets = [self.replacer]
10387 def DeclareLocks(self, level):
10388 if level == locking.LEVEL_NODEGROUP:
10389 assert self.op.remote_node is None
10390 assert self.op.iallocator is not None
10391 assert not self.needed_locks[locking.LEVEL_NODEGROUP]
10393 self.share_locks[locking.LEVEL_NODEGROUP] = 1
10394 # Lock all groups used by instance optimistically; this requires going
10395 # via the node before it's locked, requiring verification later on
10396 self.needed_locks[locking.LEVEL_NODEGROUP] = \
10397 self.cfg.GetInstanceNodeGroups(self.op.instance_name)
10399 elif level == locking.LEVEL_NODE:
10400 if self.op.iallocator is not None:
10401 assert self.op.remote_node is None
10402 assert not self.needed_locks[locking.LEVEL_NODE]
10404 # Lock member nodes of all locked groups
10405 self.needed_locks[locking.LEVEL_NODE] = [node_name
10406 for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
10407 for node_name in self.cfg.GetNodeGroup(group_uuid).members]
10409 self._LockInstancesNodes()
10410 elif level == locking.LEVEL_NODE_RES:
10412 self.needed_locks[locking.LEVEL_NODE_RES] = \
10413 self.needed_locks[locking.LEVEL_NODE]
10415 def BuildHooksEnv(self):
10416 """Build hooks env.
10418 This runs on the master, the primary and all the secondaries.
10421 instance = self.replacer.instance
10423 "MODE": self.op.mode,
10424 "NEW_SECONDARY": self.op.remote_node,
10425 "OLD_SECONDARY": instance.secondary_nodes[0],
10427 env.update(_BuildInstanceHookEnvByObject(self, instance))
10430 def BuildHooksNodes(self):
10431 """Build hooks nodes.
10434 instance = self.replacer.instance
10436 self.cfg.GetMasterNode(),
10437 instance.primary_node,
10439 if self.op.remote_node is not None:
10440 nl.append(self.op.remote_node)
10443 def CheckPrereq(self):
10444 """Check prerequisites.
10447 assert (self.glm.is_owned(locking.LEVEL_NODEGROUP) or
10448 self.op.iallocator is None)
10450 # Verify if node group locks are still correct
10451 owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
10453 _CheckInstanceNodeGroups(self.cfg, self.op.instance_name, owned_groups)
10455 return LogicalUnit.CheckPrereq(self)
10458 class TLReplaceDisks(Tasklet):
10459 """Replaces disks for an instance.
10461 Note: Locking is not within the scope of this class.
10464 def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
10465 disks, delay_iallocator, early_release, ignore_ipolicy):
10466 """Initializes this class.
10469 Tasklet.__init__(self, lu)
10472 self.instance_name = instance_name
10474 self.iallocator_name = iallocator_name
10475 self.remote_node = remote_node
10477 self.delay_iallocator = delay_iallocator
10478 self.early_release = early_release
10479 self.ignore_ipolicy = ignore_ipolicy
10482 self.instance = None
10483 self.new_node = None
10484 self.target_node = None
10485 self.other_node = None
10486 self.remote_node_info = None
10487 self.node_secondary_ip = None
10490 def CheckArguments(mode, remote_node, iallocator):
10491 """Helper function for users of this class.
10494 # check for valid parameter combination
10495 if mode == constants.REPLACE_DISK_CHG:
10496 if remote_node is None and iallocator is None:
10497 raise errors.OpPrereqError("When changing the secondary either an"
10498 " iallocator script must be used or the"
10499 " new node given", errors.ECODE_INVAL)
10501 if remote_node is not None and iallocator is not None:
10502 raise errors.OpPrereqError("Give either the iallocator or the new"
10503 " secondary, not both", errors.ECODE_INVAL)
10505 elif remote_node is not None or iallocator is not None:
10506 # Not replacing the secondary
10507 raise errors.OpPrereqError("The iallocator and new node options can"
10508 " only be used when changing the"
10509 " secondary node", errors.ECODE_INVAL)
10512 def _RunAllocator(lu, iallocator_name, instance_name, relocate_from):
10513 """Compute a new secondary node using an IAllocator.
10516 ial = IAllocator(lu.cfg, lu.rpc,
10517 mode=constants.IALLOCATOR_MODE_RELOC,
10518 name=instance_name,
10519 relocate_from=list(relocate_from))
10521 ial.Run(iallocator_name)
10523 if not ial.success:
10524 raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
10525 " %s" % (iallocator_name, ial.info),
10526 errors.ECODE_NORES)
10528 if len(ial.result) != ial.required_nodes:
10529 raise errors.OpPrereqError("iallocator '%s' returned invalid number"
10530 " of nodes (%s), required %s" %
10532 len(ial.result), ial.required_nodes),
10533 errors.ECODE_FAULT)
10535 remote_node_name = ial.result[0]
10537 lu.LogInfo("Selected new secondary for instance '%s': %s",
10538 instance_name, remote_node_name)
10540 return remote_node_name
10542 def _FindFaultyDisks(self, node_name):
10543 """Wrapper for L{_FindFaultyInstanceDisks}.
10546 return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
10549 def _CheckDisksActivated(self, instance):
10550 """Checks if the instance disks are activated.
10552 @param instance: The instance to check disks
10553 @return: True if they are activated, False otherwise
10556 nodes = instance.all_nodes
10558 for idx, dev in enumerate(instance.disks):
10560 self.lu.LogInfo("Checking disk/%d on %s", idx, node)
10561 self.cfg.SetDiskID(dev, node)
10563 result = _BlockdevFind(self, node, dev, instance)
10567 elif result.fail_msg or not result.payload:
10572 def CheckPrereq(self):
10573 """Check prerequisites.
10575 This checks that the instance is in the cluster.
10578 self.instance = instance = self.cfg.GetInstanceInfo(self.instance_name)
10579 assert instance is not None, \
10580 "Cannot retrieve locked instance %s" % self.instance_name
10582 if instance.disk_template != constants.DT_DRBD8:
10583 raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
10584 " instances", errors.ECODE_INVAL)
10586 if len(instance.secondary_nodes) != 1:
10587 raise errors.OpPrereqError("The instance has a strange layout,"
10588 " expected one secondary but found %d" %
10589 len(instance.secondary_nodes),
10590 errors.ECODE_FAULT)
10592 if not self.delay_iallocator:
10593 self._CheckPrereq2()
10595 def _CheckPrereq2(self):
10596 """Check prerequisites, second part.
10598 This function should always be part of CheckPrereq. It was separated and is
10599 now called from Exec because during node evacuation iallocator was only
10600 called with an unmodified cluster model, not taking planned changes into
10604 instance = self.instance
10605 secondary_node = instance.secondary_nodes[0]
10607 if self.iallocator_name is None:
10608 remote_node = self.remote_node
10610 remote_node = self._RunAllocator(self.lu, self.iallocator_name,
10611 instance.name, instance.secondary_nodes)
10613 if remote_node is None:
10614 self.remote_node_info = None
10616 assert remote_node in self.lu.owned_locks(locking.LEVEL_NODE), \
10617 "Remote node '%s' is not locked" % remote_node
10619 self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
10620 assert self.remote_node_info is not None, \
10621 "Cannot retrieve locked node %s" % remote_node
10623 if remote_node == self.instance.primary_node:
10624 raise errors.OpPrereqError("The specified node is the primary node of"
10625 " the instance", errors.ECODE_INVAL)
10627 if remote_node == secondary_node:
10628 raise errors.OpPrereqError("The specified node is already the"
10629 " secondary node of the instance",
10630 errors.ECODE_INVAL)
10632 if self.disks and self.mode in (constants.REPLACE_DISK_AUTO,
10633 constants.REPLACE_DISK_CHG):
10634 raise errors.OpPrereqError("Cannot specify disks to be replaced",
10635 errors.ECODE_INVAL)
10637 if self.mode == constants.REPLACE_DISK_AUTO:
10638 if not self._CheckDisksActivated(instance):
10639 raise errors.OpPrereqError("Please run activate-disks on instance %s"
10640 " first" % self.instance_name,
10641 errors.ECODE_STATE)
10642 faulty_primary = self._FindFaultyDisks(instance.primary_node)
10643 faulty_secondary = self._FindFaultyDisks(secondary_node)
10645 if faulty_primary and faulty_secondary:
10646 raise errors.OpPrereqError("Instance %s has faulty disks on more than"
10647 " one node and can not be repaired"
10648 " automatically" % self.instance_name,
10649 errors.ECODE_STATE)
10652 self.disks = faulty_primary
10653 self.target_node = instance.primary_node
10654 self.other_node = secondary_node
10655 check_nodes = [self.target_node, self.other_node]
10656 elif faulty_secondary:
10657 self.disks = faulty_secondary
10658 self.target_node = secondary_node
10659 self.other_node = instance.primary_node
10660 check_nodes = [self.target_node, self.other_node]
10666 # Non-automatic modes
10667 if self.mode == constants.REPLACE_DISK_PRI:
10668 self.target_node = instance.primary_node
10669 self.other_node = secondary_node
10670 check_nodes = [self.target_node, self.other_node]
10672 elif self.mode == constants.REPLACE_DISK_SEC:
10673 self.target_node = secondary_node
10674 self.other_node = instance.primary_node
10675 check_nodes = [self.target_node, self.other_node]
10677 elif self.mode == constants.REPLACE_DISK_CHG:
10678 self.new_node = remote_node
10679 self.other_node = instance.primary_node
10680 self.target_node = secondary_node
10681 check_nodes = [self.new_node, self.other_node]
10683 _CheckNodeNotDrained(self.lu, remote_node)
10684 _CheckNodeVmCapable(self.lu, remote_node)
10686 old_node_info = self.cfg.GetNodeInfo(secondary_node)
10687 assert old_node_info is not None
10688 if old_node_info.offline and not self.early_release:
10689 # doesn't make sense to delay the release
10690 self.early_release = True
10691 self.lu.LogInfo("Old secondary %s is offline, automatically enabling"
10692 " early-release mode", secondary_node)
10695 raise errors.ProgrammerError("Unhandled disk replace mode (%s)" %
10698 # If not specified all disks should be replaced
10700 self.disks = range(len(self.instance.disks))
10702 # TODO: This is ugly, but right now we can't distinguish between internal
10703 # submitted opcode and external one. We should fix that.
10704 if self.remote_node_info:
10705 # We change the node, lets verify it still meets instance policy
10706 new_group_info = self.cfg.GetNodeGroup(self.remote_node_info.group)
10707 ipolicy = _CalculateGroupIPolicy(self.cfg.GetClusterInfo(),
10709 _CheckTargetNodeIPolicy(self, ipolicy, instance, self.remote_node_info,
10710 ignore=self.ignore_ipolicy)
10712 for node in check_nodes:
10713 _CheckNodeOnline(self.lu, node)
10715 touched_nodes = frozenset(node_name for node_name in [self.new_node,
10718 if node_name is not None)
10720 # Release unneeded node and node resource locks
10721 _ReleaseLocks(self.lu, locking.LEVEL_NODE, keep=touched_nodes)
10722 _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES, keep=touched_nodes)
10724 # Release any owned node group
10725 if self.lu.glm.is_owned(locking.LEVEL_NODEGROUP):
10726 _ReleaseLocks(self.lu, locking.LEVEL_NODEGROUP)
10728 # Check whether disks are valid
10729 for disk_idx in self.disks:
10730 instance.FindDisk(disk_idx)
10732 # Get secondary node IP addresses
10733 self.node_secondary_ip = dict((name, node.secondary_ip) for (name, node)
10734 in self.cfg.GetMultiNodeInfo(touched_nodes))
10736 def Exec(self, feedback_fn):
10737 """Execute disk replacement.
10739 This dispatches the disk replacement to the appropriate handler.
10742 if self.delay_iallocator:
10743 self._CheckPrereq2()
10746 # Verify owned locks before starting operation
10747 owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE)
10748 assert set(owned_nodes) == set(self.node_secondary_ip), \
10749 ("Incorrect node locks, owning %s, expected %s" %
10750 (owned_nodes, self.node_secondary_ip.keys()))
10751 assert (self.lu.owned_locks(locking.LEVEL_NODE) ==
10752 self.lu.owned_locks(locking.LEVEL_NODE_RES))
10754 owned_instances = self.lu.owned_locks(locking.LEVEL_INSTANCE)
10755 assert list(owned_instances) == [self.instance_name], \
10756 "Instance '%s' not locked" % self.instance_name
10758 assert not self.lu.glm.is_owned(locking.LEVEL_NODEGROUP), \
10759 "Should not own any node group lock at this point"
10762 feedback_fn("No disks need replacement")
10765 feedback_fn("Replacing disk(s) %s for %s" %
10766 (utils.CommaJoin(self.disks), self.instance.name))
10768 activate_disks = (self.instance.admin_state != constants.ADMINST_UP)
10770 # Activate the instance disks if we're replacing them on a down instance
10772 _StartInstanceDisks(self.lu, self.instance, True)
10775 # Should we replace the secondary node?
10776 if self.new_node is not None:
10777 fn = self._ExecDrbd8Secondary
10779 fn = self._ExecDrbd8DiskOnly
10781 result = fn(feedback_fn)
10783 # Deactivate the instance disks if we're replacing them on a
10786 _SafeShutdownInstanceDisks(self.lu, self.instance)
10788 assert not self.lu.owned_locks(locking.LEVEL_NODE)
10791 # Verify owned locks
10792 owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE_RES)
10793 nodes = frozenset(self.node_secondary_ip)
10794 assert ((self.early_release and not owned_nodes) or
10795 (not self.early_release and not (set(owned_nodes) - nodes))), \
10796 ("Not owning the correct locks, early_release=%s, owned=%r,"
10797 " nodes=%r" % (self.early_release, owned_nodes, nodes))
10801 def _CheckVolumeGroup(self, nodes):
10802 self.lu.LogInfo("Checking volume groups")
10804 vgname = self.cfg.GetVGName()
10806 # Make sure volume group exists on all involved nodes
10807 results = self.rpc.call_vg_list(nodes)
10809 raise errors.OpExecError("Can't list volume groups on the nodes")
10812 res = results[node]
10813 res.Raise("Error checking node %s" % node)
10814 if vgname not in res.payload:
10815 raise errors.OpExecError("Volume group '%s' not found on node %s" %
10818 def _CheckDisksExistence(self, nodes):
10819 # Check disk existence
10820 for idx, dev in enumerate(self.instance.disks):
10821 if idx not in self.disks:
10825 self.lu.LogInfo("Checking disk/%d on %s" % (idx, node))
10826 self.cfg.SetDiskID(dev, node)
10828 result = _BlockdevFind(self, node, dev, self.instance)
10830 msg = result.fail_msg
10831 if msg or not result.payload:
10833 msg = "disk not found"
10834 raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
10837 def _CheckDisksConsistency(self, node_name, on_primary, ldisk):
10838 for idx, dev in enumerate(self.instance.disks):
10839 if idx not in self.disks:
10842 self.lu.LogInfo("Checking disk/%d consistency on node %s" %
10845 if not _CheckDiskConsistency(self.lu, self.instance, dev, node_name,
10846 on_primary, ldisk=ldisk):
10847 raise errors.OpExecError("Node %s has degraded storage, unsafe to"
10848 " replace disks for instance %s" %
10849 (node_name, self.instance.name))
10851 def _CreateNewStorage(self, node_name):
10852 """Create new storage on the primary or secondary node.
10854 This is only used for same-node replaces, not for changing the
10855 secondary node, hence we don't want to modify the existing disk.
10860 disks = _AnnotateDiskParams(self.instance, self.instance.disks, self.cfg)
10861 for idx, dev in enumerate(disks):
10862 if idx not in self.disks:
10865 self.lu.LogInfo("Adding storage on %s for disk/%d" % (node_name, idx))
10867 self.cfg.SetDiskID(dev, node_name)
10869 lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
10870 names = _GenerateUniqueNames(self.lu, lv_names)
10872 (data_disk, meta_disk) = dev.children
10873 vg_data = data_disk.logical_id[0]
10874 lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
10875 logical_id=(vg_data, names[0]),
10876 params=data_disk.params)
10877 vg_meta = meta_disk.logical_id[0]
10878 lv_meta = objects.Disk(dev_type=constants.LD_LV, size=DRBD_META_SIZE,
10879 logical_id=(vg_meta, names[1]),
10880 params=meta_disk.params)
10882 new_lvs = [lv_data, lv_meta]
10883 old_lvs = [child.Copy() for child in dev.children]
10884 iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
10886 # we pass force_create=True to force the LVM creation
10887 for new_lv in new_lvs:
10888 _CreateBlockDevInner(self.lu, node_name, self.instance, new_lv, True,
10889 _GetInstanceInfoText(self.instance), False)
10893 def _CheckDevices(self, node_name, iv_names):
10894 for name, (dev, _, _) in iv_names.iteritems():
10895 self.cfg.SetDiskID(dev, node_name)
10897 result = _BlockdevFind(self, node_name, dev, self.instance)
10899 msg = result.fail_msg
10900 if msg or not result.payload:
10902 msg = "disk not found"
10903 raise errors.OpExecError("Can't find DRBD device %s: %s" %
10906 if result.payload.is_degraded:
10907 raise errors.OpExecError("DRBD device %s is degraded!" % name)
10909 def _RemoveOldStorage(self, node_name, iv_names):
10910 for name, (_, old_lvs, _) in iv_names.iteritems():
10911 self.lu.LogInfo("Remove logical volumes for %s" % name)
10914 self.cfg.SetDiskID(lv, node_name)
10916 msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
10918 self.lu.LogWarning("Can't remove old LV: %s" % msg,
10919 hint="remove unused LVs manually")
10921 def _ExecDrbd8DiskOnly(self, feedback_fn): # pylint: disable=W0613
10922 """Replace a disk on the primary or secondary for DRBD 8.
10924 The algorithm for replace is quite complicated:
10926 1. for each disk to be replaced:
10928 1. create new LVs on the target node with unique names
10929 1. detach old LVs from the drbd device
10930 1. rename old LVs to name_replaced.<time_t>
10931 1. rename new LVs to old LVs
10932 1. attach the new LVs (with the old names now) to the drbd device
10934 1. wait for sync across all devices
10936 1. for each modified disk:
10938 1. remove old LVs (which have the name name_replaces.<time_t>)
10940 Failures are not very well handled.
10945 # Step: check device activation
10946 self.lu.LogStep(1, steps_total, "Check device existence")
10947 self._CheckDisksExistence([self.other_node, self.target_node])
10948 self._CheckVolumeGroup([self.target_node, self.other_node])
10950 # Step: check other node consistency
10951 self.lu.LogStep(2, steps_total, "Check peer consistency")
10952 self._CheckDisksConsistency(self.other_node,
10953 self.other_node == self.instance.primary_node,
10956 # Step: create new storage
10957 self.lu.LogStep(3, steps_total, "Allocate new storage")
10958 iv_names = self._CreateNewStorage(self.target_node)
10960 # Step: for each lv, detach+rename*2+attach
10961 self.lu.LogStep(4, steps_total, "Changing drbd configuration")
10962 for dev, old_lvs, new_lvs in iv_names.itervalues():
10963 self.lu.LogInfo("Detaching %s drbd from local storage" % dev.iv_name)
10965 result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
10967 result.Raise("Can't detach drbd from local storage on node"
10968 " %s for device %s" % (self.target_node, dev.iv_name))
10970 #cfg.Update(instance)
10972 # ok, we created the new LVs, so now we know we have the needed
10973 # storage; as such, we proceed on the target node to rename
10974 # old_lv to _old, and new_lv to old_lv; note that we rename LVs
10975 # using the assumption that logical_id == physical_id (which in
10976 # turn is the unique_id on that node)
10978 # FIXME(iustin): use a better name for the replaced LVs
10979 temp_suffix = int(time.time())
10980 ren_fn = lambda d, suff: (d.physical_id[0],
10981 d.physical_id[1] + "_replaced-%s" % suff)
10983 # Build the rename list based on what LVs exist on the node
10984 rename_old_to_new = []
10985 for to_ren in old_lvs:
10986 result = self.rpc.call_blockdev_find(self.target_node, to_ren)
10987 if not result.fail_msg and result.payload:
10989 rename_old_to_new.append((to_ren, ren_fn(to_ren, temp_suffix)))
10991 self.lu.LogInfo("Renaming the old LVs on the target node")
10992 result = self.rpc.call_blockdev_rename(self.target_node,
10994 result.Raise("Can't rename old LVs on node %s" % self.target_node)
10996 # Now we rename the new LVs to the old LVs
10997 self.lu.LogInfo("Renaming the new LVs on the target node")
10998 rename_new_to_old = [(new, old.physical_id)
10999 for old, new in zip(old_lvs, new_lvs)]
11000 result = self.rpc.call_blockdev_rename(self.target_node,
11002 result.Raise("Can't rename new LVs on node %s" % self.target_node)
11004 # Intermediate steps of in memory modifications
11005 for old, new in zip(old_lvs, new_lvs):
11006 new.logical_id = old.logical_id
11007 self.cfg.SetDiskID(new, self.target_node)
11009 # We need to modify old_lvs so that removal later removes the
11010 # right LVs, not the newly added ones; note that old_lvs is a
11012 for disk in old_lvs:
11013 disk.logical_id = ren_fn(disk, temp_suffix)
11014 self.cfg.SetDiskID(disk, self.target_node)
11016 # Now that the new lvs have the old name, we can add them to the device
11017 self.lu.LogInfo("Adding new mirror component on %s" % self.target_node)
11018 result = self.rpc.call_blockdev_addchildren(self.target_node,
11019 (dev, self.instance), new_lvs)
11020 msg = result.fail_msg
11022 for new_lv in new_lvs:
11023 msg2 = self.rpc.call_blockdev_remove(self.target_node,
11026 self.lu.LogWarning("Can't rollback device %s: %s", dev, msg2,
11027 hint=("cleanup manually the unused logical"
11029 raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
11031 cstep = itertools.count(5)
11033 if self.early_release:
11034 self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
11035 self._RemoveOldStorage(self.target_node, iv_names)
11036 # TODO: Check if releasing locks early still makes sense
11037 _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES)
11039 # Release all resource locks except those used by the instance
11040 _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES,
11041 keep=self.node_secondary_ip.keys())
11043 # Release all node locks while waiting for sync
11044 _ReleaseLocks(self.lu, locking.LEVEL_NODE)
11046 # TODO: Can the instance lock be downgraded here? Take the optional disk
11047 # shutdown in the caller into consideration.
11050 # This can fail as the old devices are degraded and _WaitForSync
11051 # does a combined result over all disks, so we don't check its return value
11052 self.lu.LogStep(cstep.next(), steps_total, "Sync devices")
11053 _WaitForSync(self.lu, self.instance)
11055 # Check all devices manually
11056 self._CheckDevices(self.instance.primary_node, iv_names)
11058 # Step: remove old storage
11059 if not self.early_release:
11060 self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
11061 self._RemoveOldStorage(self.target_node, iv_names)
11063 def _ExecDrbd8Secondary(self, feedback_fn):
11064 """Replace the secondary node for DRBD 8.
11066 The algorithm for replace is quite complicated:
11067 - for all disks of the instance:
11068 - create new LVs on the new node with same names
11069 - shutdown the drbd device on the old secondary
11070 - disconnect the drbd network on the primary
11071 - create the drbd device on the new secondary
11072 - network attach the drbd on the primary, using an artifice:
11073 the drbd code for Attach() will connect to the network if it
11074 finds a device which is connected to the good local disks but
11075 not network enabled
11076 - wait for sync across all devices
11077 - remove all disks from the old secondary
11079 Failures are not very well handled.
11084 pnode = self.instance.primary_node
11086 # Step: check device activation
11087 self.lu.LogStep(1, steps_total, "Check device existence")
11088 self._CheckDisksExistence([self.instance.primary_node])
11089 self._CheckVolumeGroup([self.instance.primary_node])
11091 # Step: check other node consistency
11092 self.lu.LogStep(2, steps_total, "Check peer consistency")
11093 self._CheckDisksConsistency(self.instance.primary_node, True, True)
11095 # Step: create new storage
11096 self.lu.LogStep(3, steps_total, "Allocate new storage")
11097 disks = _AnnotateDiskParams(self.instance, self.instance.disks, self.cfg)
11098 for idx, dev in enumerate(disks):
11099 self.lu.LogInfo("Adding new local storage on %s for disk/%d" %
11100 (self.new_node, idx))
11101 # we pass force_create=True to force LVM creation
11102 for new_lv in dev.children:
11103 _CreateBlockDevInner(self.lu, self.new_node, self.instance, new_lv,
11104 True, _GetInstanceInfoText(self.instance), False)
11106 # Step 4: dbrd minors and drbd setups changes
11107 # after this, we must manually remove the drbd minors on both the
11108 # error and the success paths
11109 self.lu.LogStep(4, steps_total, "Changing drbd configuration")
11110 minors = self.cfg.AllocateDRBDMinor([self.new_node
11111 for dev in self.instance.disks],
11112 self.instance.name)
11113 logging.debug("Allocated minors %r", minors)
11116 for idx, (dev, new_minor) in enumerate(zip(self.instance.disks, minors)):
11117 self.lu.LogInfo("activating a new drbd on %s for disk/%d" %
11118 (self.new_node, idx))
11119 # create new devices on new_node; note that we create two IDs:
11120 # one without port, so the drbd will be activated without
11121 # networking information on the new node at this stage, and one
11122 # with network, for the latter activation in step 4
11123 (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
11124 if self.instance.primary_node == o_node1:
11127 assert self.instance.primary_node == o_node2, "Three-node instance?"
11130 new_alone_id = (self.instance.primary_node, self.new_node, None,
11131 p_minor, new_minor, o_secret)
11132 new_net_id = (self.instance.primary_node, self.new_node, o_port,
11133 p_minor, new_minor, o_secret)
11135 iv_names[idx] = (dev, dev.children, new_net_id)
11136 logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
11138 new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
11139 logical_id=new_alone_id,
11140 children=dev.children,
11143 (anno_new_drbd,) = _AnnotateDiskParams(self.instance, [new_drbd],
11146 _CreateSingleBlockDev(self.lu, self.new_node, self.instance,
11148 _GetInstanceInfoText(self.instance), False)
11149 except errors.GenericError:
11150 self.cfg.ReleaseDRBDMinors(self.instance.name)
11153 # We have new devices, shutdown the drbd on the old secondary
11154 for idx, dev in enumerate(self.instance.disks):
11155 self.lu.LogInfo("Shutting down drbd for disk/%d on old node" % idx)
11156 self.cfg.SetDiskID(dev, self.target_node)
11157 msg = self.rpc.call_blockdev_shutdown(self.target_node, dev).fail_msg
11159 self.lu.LogWarning("Failed to shutdown drbd for disk/%d on old"
11160 "node: %s" % (idx, msg),
11161 hint=("Please cleanup this device manually as"
11162 " soon as possible"))
11164 self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
11165 result = self.rpc.call_drbd_disconnect_net([pnode], self.node_secondary_ip,
11166 self.instance.disks)[pnode]
11168 msg = result.fail_msg
11170 # detaches didn't succeed (unlikely)
11171 self.cfg.ReleaseDRBDMinors(self.instance.name)
11172 raise errors.OpExecError("Can't detach the disks from the network on"
11173 " old node: %s" % (msg,))
11175 # if we managed to detach at least one, we update all the disks of
11176 # the instance to point to the new secondary
11177 self.lu.LogInfo("Updating instance configuration")
11178 for dev, _, new_logical_id in iv_names.itervalues():
11179 dev.logical_id = new_logical_id
11180 self.cfg.SetDiskID(dev, self.instance.primary_node)
11182 self.cfg.Update(self.instance, feedback_fn)
11184 # Release all node locks (the configuration has been updated)
11185 _ReleaseLocks(self.lu, locking.LEVEL_NODE)
11187 # and now perform the drbd attach
11188 self.lu.LogInfo("Attaching primary drbds to new secondary"
11189 " (standalone => connected)")
11190 result = self.rpc.call_drbd_attach_net([self.instance.primary_node,
11192 self.node_secondary_ip,
11193 (self.instance.disks, self.instance),
11194 self.instance.name,
11196 for to_node, to_result in result.items():
11197 msg = to_result.fail_msg
11199 self.lu.LogWarning("Can't attach drbd disks on node %s: %s",
11201 hint=("please do a gnt-instance info to see the"
11202 " status of disks"))
11204 cstep = itertools.count(5)
11206 if self.early_release:
11207 self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
11208 self._RemoveOldStorage(self.target_node, iv_names)
11209 # TODO: Check if releasing locks early still makes sense
11210 _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES)
11212 # Release all resource locks except those used by the instance
11213 _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES,
11214 keep=self.node_secondary_ip.keys())
11216 # TODO: Can the instance lock be downgraded here? Take the optional disk
11217 # shutdown in the caller into consideration.
11220 # This can fail as the old devices are degraded and _WaitForSync
11221 # does a combined result over all disks, so we don't check its return value
11222 self.lu.LogStep(cstep.next(), steps_total, "Sync devices")
11223 _WaitForSync(self.lu, self.instance)
11225 # Check all devices manually
11226 self._CheckDevices(self.instance.primary_node, iv_names)
11228 # Step: remove old storage
11229 if not self.early_release:
11230 self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
11231 self._RemoveOldStorage(self.target_node, iv_names)
11234 class LURepairNodeStorage(NoHooksLU):
11235 """Repairs the volume group on a node.
11240 def CheckArguments(self):
11241 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
11243 storage_type = self.op.storage_type
11245 if (constants.SO_FIX_CONSISTENCY not in
11246 constants.VALID_STORAGE_OPERATIONS.get(storage_type, [])):
11247 raise errors.OpPrereqError("Storage units of type '%s' can not be"
11248 " repaired" % storage_type,
11249 errors.ECODE_INVAL)
11251 def ExpandNames(self):
11252 self.needed_locks = {
11253 locking.LEVEL_NODE: [self.op.node_name],
11256 def _CheckFaultyDisks(self, instance, node_name):
11257 """Ensure faulty disks abort the opcode or at least warn."""
11259 if _FindFaultyInstanceDisks(self.cfg, self.rpc, instance,
11261 raise errors.OpPrereqError("Instance '%s' has faulty disks on"
11262 " node '%s'" % (instance.name, node_name),
11263 errors.ECODE_STATE)
11264 except errors.OpPrereqError, err:
11265 if self.op.ignore_consistency:
11266 self.proc.LogWarning(str(err.args[0]))
11270 def CheckPrereq(self):
11271 """Check prerequisites.
11274 # Check whether any instance on this node has faulty disks
11275 for inst in _GetNodeInstances(self.cfg, self.op.node_name):
11276 if inst.admin_state != constants.ADMINST_UP:
11278 check_nodes = set(inst.all_nodes)
11279 check_nodes.discard(self.op.node_name)
11280 for inst_node_name in check_nodes:
11281 self._CheckFaultyDisks(inst, inst_node_name)
11283 def Exec(self, feedback_fn):
11284 feedback_fn("Repairing storage unit '%s' on %s ..." %
11285 (self.op.name, self.op.node_name))
11287 st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
11288 result = self.rpc.call_storage_execute(self.op.node_name,
11289 self.op.storage_type, st_args,
11291 constants.SO_FIX_CONSISTENCY)
11292 result.Raise("Failed to repair storage unit '%s' on %s" %
11293 (self.op.name, self.op.node_name))
11296 class LUNodeEvacuate(NoHooksLU):
11297 """Evacuates instances off a list of nodes.
11302 _MODE2IALLOCATOR = {
11303 constants.NODE_EVAC_PRI: constants.IALLOCATOR_NEVAC_PRI,
11304 constants.NODE_EVAC_SEC: constants.IALLOCATOR_NEVAC_SEC,
11305 constants.NODE_EVAC_ALL: constants.IALLOCATOR_NEVAC_ALL,
11307 assert frozenset(_MODE2IALLOCATOR.keys()) == constants.NODE_EVAC_MODES
11308 assert (frozenset(_MODE2IALLOCATOR.values()) ==
11309 constants.IALLOCATOR_NEVAC_MODES)
11311 def CheckArguments(self):
11312 _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
11314 def ExpandNames(self):
11315 self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
11317 if self.op.remote_node is not None:
11318 self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
11319 assert self.op.remote_node
11321 if self.op.remote_node == self.op.node_name:
11322 raise errors.OpPrereqError("Can not use evacuated node as a new"
11323 " secondary node", errors.ECODE_INVAL)
11325 if self.op.mode != constants.NODE_EVAC_SEC:
11326 raise errors.OpPrereqError("Without the use of an iallocator only"
11327 " secondary instances can be evacuated",
11328 errors.ECODE_INVAL)
11331 self.share_locks = _ShareAll()
11332 self.needed_locks = {
11333 locking.LEVEL_INSTANCE: [],
11334 locking.LEVEL_NODEGROUP: [],
11335 locking.LEVEL_NODE: [],
11338 # Determine nodes (via group) optimistically, needs verification once locks
11339 # have been acquired
11340 self.lock_nodes = self._DetermineNodes()
11342 def _DetermineNodes(self):
11343 """Gets the list of nodes to operate on.
11346 if self.op.remote_node is None:
11347 # Iallocator will choose any node(s) in the same group
11348 group_nodes = self.cfg.GetNodeGroupMembersByNodes([self.op.node_name])
11350 group_nodes = frozenset([self.op.remote_node])
11352 # Determine nodes to be locked
11353 return set([self.op.node_name]) | group_nodes
11355 def _DetermineInstances(self):
11356 """Builds list of instances to operate on.
11359 assert self.op.mode in constants.NODE_EVAC_MODES
11361 if self.op.mode == constants.NODE_EVAC_PRI:
11362 # Primary instances only
11363 inst_fn = _GetNodePrimaryInstances
11364 assert self.op.remote_node is None, \
11365 "Evacuating primary instances requires iallocator"
11366 elif self.op.mode == constants.NODE_EVAC_SEC:
11367 # Secondary instances only
11368 inst_fn = _GetNodeSecondaryInstances
11371 assert self.op.mode == constants.NODE_EVAC_ALL
11372 inst_fn = _GetNodeInstances
11373 # TODO: In 2.6, change the iallocator interface to take an evacuation mode
11375 raise errors.OpPrereqError("Due to an issue with the iallocator"
11376 " interface it is not possible to evacuate"
11377 " all instances at once; specify explicitly"
11378 " whether to evacuate primary or secondary"
11380 errors.ECODE_INVAL)
11382 return inst_fn(self.cfg, self.op.node_name)
11384 def DeclareLocks(self, level):
11385 if level == locking.LEVEL_INSTANCE:
11386 # Lock instances optimistically, needs verification once node and group
11387 # locks have been acquired
11388 self.needed_locks[locking.LEVEL_INSTANCE] = \
11389 set(i.name for i in self._DetermineInstances())
11391 elif level == locking.LEVEL_NODEGROUP:
11392 # Lock node groups for all potential target nodes optimistically, needs
11393 # verification once nodes have been acquired
11394 self.needed_locks[locking.LEVEL_NODEGROUP] = \
11395 self.cfg.GetNodeGroupsFromNodes(self.lock_nodes)
11397 elif level == locking.LEVEL_NODE:
11398 self.needed_locks[locking.LEVEL_NODE] = self.lock_nodes
11400 def CheckPrereq(self):
11402 owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
11403 owned_nodes = self.owned_locks(locking.LEVEL_NODE)
11404 owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
11406 need_nodes = self._DetermineNodes()
11408 if not owned_nodes.issuperset(need_nodes):
11409 raise errors.OpPrereqError("Nodes in same group as '%s' changed since"
11410 " locks were acquired, current nodes are"
11411 " are '%s', used to be '%s'; retry the"
11413 (self.op.node_name,
11414 utils.CommaJoin(need_nodes),
11415 utils.CommaJoin(owned_nodes)),
11416 errors.ECODE_STATE)
11418 wanted_groups = self.cfg.GetNodeGroupsFromNodes(owned_nodes)
11419 if owned_groups != wanted_groups:
11420 raise errors.OpExecError("Node groups changed since locks were acquired,"
11421 " current groups are '%s', used to be '%s';"
11422 " retry the operation" %
11423 (utils.CommaJoin(wanted_groups),
11424 utils.CommaJoin(owned_groups)))
11426 # Determine affected instances
11427 self.instances = self._DetermineInstances()
11428 self.instance_names = [i.name for i in self.instances]
11430 if set(self.instance_names) != owned_instances:
11431 raise errors.OpExecError("Instances on node '%s' changed since locks"
11432 " were acquired, current instances are '%s',"
11433 " used to be '%s'; retry the operation" %
11434 (self.op.node_name,
11435 utils.CommaJoin(self.instance_names),
11436 utils.CommaJoin(owned_instances)))
11438 if self.instance_names:
11439 self.LogInfo("Evacuating instances from node '%s': %s",
11441 utils.CommaJoin(utils.NiceSort(self.instance_names)))
11443 self.LogInfo("No instances to evacuate from node '%s'",
11446 if self.op.remote_node is not None:
11447 for i in self.instances:
11448 if i.primary_node == self.op.remote_node:
11449 raise errors.OpPrereqError("Node %s is the primary node of"
11450 " instance %s, cannot use it as"
11452 (self.op.remote_node, i.name),
11453 errors.ECODE_INVAL)
11455 def Exec(self, feedback_fn):
11456 assert (self.op.iallocator is not None) ^ (self.op.remote_node is not None)
11458 if not self.instance_names:
11459 # No instances to evacuate
11462 elif self.op.iallocator is not None:
11463 # TODO: Implement relocation to other group
11464 ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_NODE_EVAC,
11465 evac_mode=self._MODE2IALLOCATOR[self.op.mode],
11466 instances=list(self.instance_names))
11468 ial.Run(self.op.iallocator)
11470 if not ial.success:
11471 raise errors.OpPrereqError("Can't compute node evacuation using"
11472 " iallocator '%s': %s" %
11473 (self.op.iallocator, ial.info),
11474 errors.ECODE_NORES)
11476 jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, True)
11478 elif self.op.remote_node is not None:
11479 assert self.op.mode == constants.NODE_EVAC_SEC
11481 [opcodes.OpInstanceReplaceDisks(instance_name=instance_name,
11482 remote_node=self.op.remote_node,
11484 mode=constants.REPLACE_DISK_CHG,
11485 early_release=self.op.early_release)]
11486 for instance_name in self.instance_names
11490 raise errors.ProgrammerError("No iallocator or remote node")
11492 return ResultWithJobs(jobs)
11495 def _SetOpEarlyRelease(early_release, op):
11496 """Sets C{early_release} flag on opcodes if available.
11500 op.early_release = early_release
11501 except AttributeError:
11502 assert not isinstance(op, opcodes.OpInstanceReplaceDisks)
11507 def _NodeEvacDest(use_nodes, group, nodes):
11508 """Returns group or nodes depending on caller's choice.
11512 return utils.CommaJoin(nodes)
11517 def _LoadNodeEvacResult(lu, alloc_result, early_release, use_nodes):
11518 """Unpacks the result of change-group and node-evacuate iallocator requests.
11520 Iallocator modes L{constants.IALLOCATOR_MODE_NODE_EVAC} and
11521 L{constants.IALLOCATOR_MODE_CHG_GROUP}.
11523 @type lu: L{LogicalUnit}
11524 @param lu: Logical unit instance
11525 @type alloc_result: tuple/list
11526 @param alloc_result: Result from iallocator
11527 @type early_release: bool
11528 @param early_release: Whether to release locks early if possible
11529 @type use_nodes: bool
11530 @param use_nodes: Whether to display node names instead of groups
11533 (moved, failed, jobs) = alloc_result
11536 failreason = utils.CommaJoin("%s (%s)" % (name, reason)
11537 for (name, reason) in failed)
11538 lu.LogWarning("Unable to evacuate instances %s", failreason)
11539 raise errors.OpExecError("Unable to evacuate instances %s" % failreason)
11542 lu.LogInfo("Instances to be moved: %s",
11543 utils.CommaJoin("%s (to %s)" %
11544 (name, _NodeEvacDest(use_nodes, group, nodes))
11545 for (name, group, nodes) in moved))
11547 return [map(compat.partial(_SetOpEarlyRelease, early_release),
11548 map(opcodes.OpCode.LoadOpCode, ops))
11552 class LUInstanceGrowDisk(LogicalUnit):
11553 """Grow a disk of an instance.
11556 HPATH = "disk-grow"
11557 HTYPE = constants.HTYPE_INSTANCE
11560 def ExpandNames(self):
11561 self._ExpandAndLockInstance()
11562 self.needed_locks[locking.LEVEL_NODE] = []
11563 self.needed_locks[locking.LEVEL_NODE_RES] = []
11564 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
11565 self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
11567 def DeclareLocks(self, level):
11568 if level == locking.LEVEL_NODE:
11569 self._LockInstancesNodes()
11570 elif level == locking.LEVEL_NODE_RES:
11572 self.needed_locks[locking.LEVEL_NODE_RES] = \
11573 self.needed_locks[locking.LEVEL_NODE][:]
11575 def BuildHooksEnv(self):
11576 """Build hooks env.
11578 This runs on the master, the primary and all the secondaries.
11582 "DISK": self.op.disk,
11583 "AMOUNT": self.op.amount,
11584 "ABSOLUTE": self.op.absolute,
11586 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
11589 def BuildHooksNodes(self):
11590 """Build hooks nodes.
11593 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
11596 def CheckPrereq(self):
11597 """Check prerequisites.
11599 This checks that the instance is in the cluster.
11602 instance = self.cfg.GetInstanceInfo(self.op.instance_name)
11603 assert instance is not None, \
11604 "Cannot retrieve locked instance %s" % self.op.instance_name
11605 nodenames = list(instance.all_nodes)
11606 for node in nodenames:
11607 _CheckNodeOnline(self, node)
11609 self.instance = instance
11611 if instance.disk_template not in constants.DTS_GROWABLE:
11612 raise errors.OpPrereqError("Instance's disk layout does not support"
11613 " growing", errors.ECODE_INVAL)
11615 self.disk = instance.FindDisk(self.op.disk)
11617 if self.op.absolute:
11618 self.target = self.op.amount
11619 self.delta = self.target - self.disk.size
11621 raise errors.OpPrereqError("Requested size (%s) is smaller than "
11622 "current disk size (%s)" %
11623 (utils.FormatUnit(self.target, "h"),
11624 utils.FormatUnit(self.disk.size, "h")),
11625 errors.ECODE_STATE)
11627 self.delta = self.op.amount
11628 self.target = self.disk.size + self.delta
11630 raise errors.OpPrereqError("Requested increment (%s) is negative" %
11631 utils.FormatUnit(self.delta, "h"),
11632 errors.ECODE_INVAL)
11634 if instance.disk_template not in (constants.DT_FILE,
11635 constants.DT_SHARED_FILE,
11637 # TODO: check the free disk space for file, when that feature will be
11639 _CheckNodesFreeDiskPerVG(self, nodenames,
11640 self.disk.ComputeGrowth(self.delta))
11642 def Exec(self, feedback_fn):
11643 """Execute disk grow.
11646 instance = self.instance
11649 assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
11650 assert (self.owned_locks(locking.LEVEL_NODE) ==
11651 self.owned_locks(locking.LEVEL_NODE_RES))
11653 disks_ok, _ = _AssembleInstanceDisks(self, self.instance, disks=[disk])
11655 raise errors.OpExecError("Cannot activate block device to grow")
11657 feedback_fn("Growing disk %s of instance '%s' by %s to %s" %
11658 (self.op.disk, instance.name,
11659 utils.FormatUnit(self.delta, "h"),
11660 utils.FormatUnit(self.target, "h")))
11662 # First run all grow ops in dry-run mode
11663 for node in instance.all_nodes:
11664 self.cfg.SetDiskID(disk, node)
11665 result = self.rpc.call_blockdev_grow(node, (disk, instance), self.delta,
11667 result.Raise("Grow request failed to node %s" % node)
11669 # We know that (as far as we can test) operations across different
11670 # nodes will succeed, time to run it for real
11671 for node in instance.all_nodes:
11672 self.cfg.SetDiskID(disk, node)
11673 result = self.rpc.call_blockdev_grow(node, (disk, instance), self.delta,
11675 result.Raise("Grow request failed to node %s" % node)
11677 # TODO: Rewrite code to work properly
11678 # DRBD goes into sync mode for a short amount of time after executing the
11679 # "resize" command. DRBD 8.x below version 8.0.13 contains a bug whereby
11680 # calling "resize" in sync mode fails. Sleeping for a short amount of
11681 # time is a work-around.
11684 disk.RecordGrow(self.delta)
11685 self.cfg.Update(instance, feedback_fn)
11687 # Changes have been recorded, release node lock
11688 _ReleaseLocks(self, locking.LEVEL_NODE)
11690 # Downgrade lock while waiting for sync
11691 self.glm.downgrade(locking.LEVEL_INSTANCE)
11693 if self.op.wait_for_sync:
11694 disk_abort = not _WaitForSync(self, instance, disks=[disk])
11696 self.proc.LogWarning("Disk sync-ing has not returned a good"
11697 " status; please check the instance")
11698 if instance.admin_state != constants.ADMINST_UP:
11699 _SafeShutdownInstanceDisks(self, instance, disks=[disk])
11700 elif instance.admin_state != constants.ADMINST_UP:
11701 self.proc.LogWarning("Not shutting down the disk even if the instance is"
11702 " not supposed to be running because no wait for"
11703 " sync mode was requested")
11705 assert self.owned_locks(locking.LEVEL_NODE_RES)
11706 assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
11709 class LUInstanceQueryData(NoHooksLU):
11710 """Query runtime instance data.
11715 def ExpandNames(self):
11716 self.needed_locks = {}
11718 # Use locking if requested or when non-static information is wanted
11719 if not (self.op.static or self.op.use_locking):
11720 self.LogWarning("Non-static data requested, locks need to be acquired")
11721 self.op.use_locking = True
11723 if self.op.instances or not self.op.use_locking:
11724 # Expand instance names right here
11725 self.wanted_names = _GetWantedInstances(self, self.op.instances)
11727 # Will use acquired locks
11728 self.wanted_names = None
11730 if self.op.use_locking:
11731 self.share_locks = _ShareAll()
11733 if self.wanted_names is None:
11734 self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
11736 self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
11738 self.needed_locks[locking.LEVEL_NODEGROUP] = []
11739 self.needed_locks[locking.LEVEL_NODE] = []
11740 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
11742 def DeclareLocks(self, level):
11743 if self.op.use_locking:
11744 if level == locking.LEVEL_NODEGROUP:
11745 owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
11747 # Lock all groups used by instances optimistically; this requires going
11748 # via the node before it's locked, requiring verification later on
11749 self.needed_locks[locking.LEVEL_NODEGROUP] = \
11750 frozenset(group_uuid
11751 for instance_name in owned_instances
11753 self.cfg.GetInstanceNodeGroups(instance_name))
11755 elif level == locking.LEVEL_NODE:
11756 self._LockInstancesNodes()
11758 def CheckPrereq(self):
11759 """Check prerequisites.
11761 This only checks the optional instance list against the existing names.
11764 owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
11765 owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
11766 owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
11768 if self.wanted_names is None:
11769 assert self.op.use_locking, "Locking was not used"
11770 self.wanted_names = owned_instances
11772 instances = dict(self.cfg.GetMultiInstanceInfo(self.wanted_names))
11774 if self.op.use_locking:
11775 _CheckInstancesNodeGroups(self.cfg, instances, owned_groups, owned_nodes,
11778 assert not (owned_instances or owned_groups or owned_nodes)
11780 self.wanted_instances = instances.values()
11782 def _ComputeBlockdevStatus(self, node, instance, dev):
11783 """Returns the status of a block device
11786 if self.op.static or not node:
11789 self.cfg.SetDiskID(dev, node)
11791 result = self.rpc.call_blockdev_find(node, dev)
11795 result.Raise("Can't compute disk status for %s" % instance.name)
11797 status = result.payload
11801 return (status.dev_path, status.major, status.minor,
11802 status.sync_percent, status.estimated_time,
11803 status.is_degraded, status.ldisk_status)
11805 def _ComputeDiskStatus(self, instance, snode, dev):
11806 """Compute block device status.
11809 (anno_dev,) = _AnnotateDiskParams(instance, [dev], self.cfg)
11811 return self._ComputeDiskStatusInner(instance, snode, anno_dev)
11813 def _ComputeDiskStatusInner(self, instance, snode, dev):
11814 """Compute block device status.
11816 @attention: The device has to be annotated already.
11819 if dev.dev_type in constants.LDS_DRBD:
11820 # we change the snode then (otherwise we use the one passed in)
11821 if dev.logical_id[0] == instance.primary_node:
11822 snode = dev.logical_id[1]
11824 snode = dev.logical_id[0]
11826 dev_pstatus = self._ComputeBlockdevStatus(instance.primary_node,
11828 dev_sstatus = self._ComputeBlockdevStatus(snode, instance, dev)
11831 dev_children = map(compat.partial(self._ComputeDiskStatusInner,
11838 "iv_name": dev.iv_name,
11839 "dev_type": dev.dev_type,
11840 "logical_id": dev.logical_id,
11841 "physical_id": dev.physical_id,
11842 "pstatus": dev_pstatus,
11843 "sstatus": dev_sstatus,
11844 "children": dev_children,
11849 def Exec(self, feedback_fn):
11850 """Gather and return data"""
11853 cluster = self.cfg.GetClusterInfo()
11855 node_names = itertools.chain(*(i.all_nodes for i in self.wanted_instances))
11856 nodes = dict(self.cfg.GetMultiNodeInfo(node_names))
11858 groups = dict(self.cfg.GetMultiNodeGroupInfo(node.group
11859 for node in nodes.values()))
11861 group2name_fn = lambda uuid: groups[uuid].name
11863 for instance in self.wanted_instances:
11864 pnode = nodes[instance.primary_node]
11866 if self.op.static or pnode.offline:
11867 remote_state = None
11869 self.LogWarning("Primary node %s is marked offline, returning static"
11870 " information only for instance %s" %
11871 (pnode.name, instance.name))
11873 remote_info = self.rpc.call_instance_info(instance.primary_node,
11875 instance.hypervisor)
11876 remote_info.Raise("Error checking node %s" % instance.primary_node)
11877 remote_info = remote_info.payload
11878 if remote_info and "state" in remote_info:
11879 remote_state = "up"
11881 if instance.admin_state == constants.ADMINST_UP:
11882 remote_state = "down"
11884 remote_state = instance.admin_state
11886 disks = map(compat.partial(self._ComputeDiskStatus, instance, None),
11889 snodes_group_uuids = [nodes[snode_name].group
11890 for snode_name in instance.secondary_nodes]
11892 result[instance.name] = {
11893 "name": instance.name,
11894 "config_state": instance.admin_state,
11895 "run_state": remote_state,
11896 "pnode": instance.primary_node,
11897 "pnode_group_uuid": pnode.group,
11898 "pnode_group_name": group2name_fn(pnode.group),
11899 "snodes": instance.secondary_nodes,
11900 "snodes_group_uuids": snodes_group_uuids,
11901 "snodes_group_names": map(group2name_fn, snodes_group_uuids),
11903 # this happens to be the same format used for hooks
11904 "nics": _NICListToTuple(self, instance.nics),
11905 "disk_template": instance.disk_template,
11907 "hypervisor": instance.hypervisor,
11908 "network_port": instance.network_port,
11909 "hv_instance": instance.hvparams,
11910 "hv_actual": cluster.FillHV(instance, skip_globals=True),
11911 "be_instance": instance.beparams,
11912 "be_actual": cluster.FillBE(instance),
11913 "os_instance": instance.osparams,
11914 "os_actual": cluster.SimpleFillOS(instance.os, instance.osparams),
11915 "serial_no": instance.serial_no,
11916 "mtime": instance.mtime,
11917 "ctime": instance.ctime,
11918 "uuid": instance.uuid,
11924 def PrepareContainerMods(mods, private_fn):
11925 """Prepares a list of container modifications by adding a private data field.
11927 @type mods: list of tuples; (operation, index, parameters)
11928 @param mods: List of modifications
11929 @type private_fn: callable or None
11930 @param private_fn: Callable for constructing a private data field for a
11935 if private_fn is None:
11940 return [(op, idx, params, fn()) for (op, idx, params) in mods]
11943 #: Type description for changes as returned by L{ApplyContainerMods}'s
11945 _TApplyContModsCbChanges = \
11946 ht.TMaybeListOf(ht.TAnd(ht.TIsLength(2), ht.TItems([
11947 ht.TNonEmptyString,
11952 def ApplyContainerMods(kind, container, chgdesc, mods,
11953 create_fn, modify_fn, remove_fn):
11954 """Applies descriptions in C{mods} to C{container}.
11957 @param kind: One-word item description
11958 @type container: list
11959 @param container: Container to modify
11960 @type chgdesc: None or list
11961 @param chgdesc: List of applied changes
11963 @param mods: Modifications as returned by L{PrepareContainerMods}
11964 @type create_fn: callable
11965 @param create_fn: Callback for creating a new item (L{constants.DDM_ADD});
11966 receives absolute item index, parameters and private data object as added
11967 by L{PrepareContainerMods}, returns tuple containing new item and changes
11969 @type modify_fn: callable
11970 @param modify_fn: Callback for modifying an existing item
11971 (L{constants.DDM_MODIFY}); receives absolute item index, item, parameters
11972 and private data object as added by L{PrepareContainerMods}, returns
11974 @type remove_fn: callable
11975 @param remove_fn: Callback on removing item; receives absolute item index,
11976 item and private data object as added by L{PrepareContainerMods}
11979 for (op, idx, params, private) in mods:
11982 absidx = len(container) - 1
11984 raise IndexError("Not accepting negative indices other than -1")
11985 elif idx > len(container):
11986 raise IndexError("Got %s index %s, but there are only %s" %
11987 (kind, idx, len(container)))
11993 if op == constants.DDM_ADD:
11994 # Calculate where item will be added
11996 addidx = len(container)
12000 if create_fn is None:
12003 (item, changes) = create_fn(addidx, params, private)
12006 container.append(item)
12009 assert idx <= len(container)
12010 # list.insert does so before the specified index
12011 container.insert(idx, item)
12013 # Retrieve existing item
12015 item = container[absidx]
12017 raise IndexError("Invalid %s index %s" % (kind, idx))
12019 if op == constants.DDM_REMOVE:
12022 if remove_fn is not None:
12023 remove_fn(absidx, item, private)
12025 changes = [("%s/%s" % (kind, absidx), "remove")]
12027 assert container[absidx] == item
12028 del container[absidx]
12029 elif op == constants.DDM_MODIFY:
12030 if modify_fn is not None:
12031 changes = modify_fn(absidx, item, params, private)
12033 raise errors.ProgrammerError("Unhandled operation '%s'" % op)
12035 assert _TApplyContModsCbChanges(changes)
12037 if not (chgdesc is None or changes is None):
12038 chgdesc.extend(changes)
12041 def _UpdateIvNames(base_index, disks):
12042 """Updates the C{iv_name} attribute of disks.
12044 @type disks: list of L{objects.Disk}
12047 for (idx, disk) in enumerate(disks):
12048 disk.iv_name = "disk/%s" % (base_index + idx, )
12051 class _InstNicModPrivate:
12052 """Data structure for network interface modifications.
12054 Used by L{LUInstanceSetParams}.
12057 def __init__(self):
12062 class LUInstanceSetParams(LogicalUnit):
12063 """Modifies an instances's parameters.
12066 HPATH = "instance-modify"
12067 HTYPE = constants.HTYPE_INSTANCE
12071 def _UpgradeDiskNicMods(kind, mods, verify_fn):
12072 assert ht.TList(mods)
12073 assert not mods or len(mods[0]) in (2, 3)
12075 if mods and len(mods[0]) == 2:
12079 for op, params in mods:
12080 if op in (constants.DDM_ADD, constants.DDM_REMOVE):
12081 result.append((op, -1, params))
12085 raise errors.OpPrereqError("Only one %s add or remove operation is"
12086 " supported at a time" % kind,
12087 errors.ECODE_INVAL)
12089 result.append((constants.DDM_MODIFY, op, params))
12091 assert verify_fn(result)
12098 def _CheckMods(kind, mods, key_types, item_fn):
12099 """Ensures requested disk/NIC modifications are valid.
12102 for (op, _, params) in mods:
12103 assert ht.TDict(params)
12105 utils.ForceDictType(params, key_types)
12107 if op == constants.DDM_REMOVE:
12109 raise errors.OpPrereqError("No settings should be passed when"
12110 " removing a %s" % kind,
12111 errors.ECODE_INVAL)
12112 elif op in (constants.DDM_ADD, constants.DDM_MODIFY):
12113 item_fn(op, params)
12115 raise errors.ProgrammerError("Unhandled operation '%s'" % op)
12118 def _VerifyDiskModification(op, params):
12119 """Verifies a disk modification.
12122 if op == constants.DDM_ADD:
12123 mode = params.setdefault(constants.IDISK_MODE, constants.DISK_RDWR)
12124 if mode not in constants.DISK_ACCESS_SET:
12125 raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode,
12126 errors.ECODE_INVAL)
12128 size = params.get(constants.IDISK_SIZE, None)
12130 raise errors.OpPrereqError("Required disk parameter '%s' missing" %
12131 constants.IDISK_SIZE, errors.ECODE_INVAL)
12135 except (TypeError, ValueError), err:
12136 raise errors.OpPrereqError("Invalid disk size parameter: %s" % err,
12137 errors.ECODE_INVAL)
12139 params[constants.IDISK_SIZE] = size
12141 elif op == constants.DDM_MODIFY and constants.IDISK_SIZE in params:
12142 raise errors.OpPrereqError("Disk size change not possible, use"
12143 " grow-disk", errors.ECODE_INVAL)
12146 def _VerifyNicModification(op, params):
12147 """Verifies a network interface modification.
12150 if op in (constants.DDM_ADD, constants.DDM_MODIFY):
12151 ip = params.get(constants.INIC_IP, None)
12154 elif ip.lower() == constants.VALUE_NONE:
12155 params[constants.INIC_IP] = None
12156 elif not netutils.IPAddress.IsValid(ip):
12157 raise errors.OpPrereqError("Invalid IP address '%s'" % ip,
12158 errors.ECODE_INVAL)
12160 bridge = params.get("bridge", None)
12161 link = params.get(constants.INIC_LINK, None)
12162 if bridge and link:
12163 raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
12164 " at the same time", errors.ECODE_INVAL)
12165 elif bridge and bridge.lower() == constants.VALUE_NONE:
12166 params["bridge"] = None
12167 elif link and link.lower() == constants.VALUE_NONE:
12168 params[constants.INIC_LINK] = None
12170 if op == constants.DDM_ADD:
12171 macaddr = params.get(constants.INIC_MAC, None)
12172 if macaddr is None:
12173 params[constants.INIC_MAC] = constants.VALUE_AUTO
12175 if constants.INIC_MAC in params:
12176 macaddr = params[constants.INIC_MAC]
12177 if macaddr not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
12178 macaddr = utils.NormalizeAndValidateMac(macaddr)
12180 if op == constants.DDM_MODIFY and macaddr == constants.VALUE_AUTO:
12181 raise errors.OpPrereqError("'auto' is not a valid MAC address when"
12182 " modifying an existing NIC",
12183 errors.ECODE_INVAL)
12185 def CheckArguments(self):
12186 if not (self.op.nics or self.op.disks or self.op.disk_template or
12187 self.op.hvparams or self.op.beparams or self.op.os_name or
12188 self.op.offline is not None or self.op.runtime_mem):
12189 raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
12191 if self.op.hvparams:
12192 _CheckGlobalHvParams(self.op.hvparams)
12195 self._UpgradeDiskNicMods("disk", self.op.disks,
12196 opcodes.OpInstanceSetParams.TestDiskModifications)
12198 self._UpgradeDiskNicMods("NIC", self.op.nics,
12199 opcodes.OpInstanceSetParams.TestNicModifications)
12201 # Check disk modifications
12202 self._CheckMods("disk", self.op.disks, constants.IDISK_PARAMS_TYPES,
12203 self._VerifyDiskModification)
12205 if self.op.disks and self.op.disk_template is not None:
12206 raise errors.OpPrereqError("Disk template conversion and other disk"
12207 " changes not supported at the same time",
12208 errors.ECODE_INVAL)
12210 if (self.op.disk_template and
12211 self.op.disk_template in constants.DTS_INT_MIRROR and
12212 self.op.remote_node is None):
12213 raise errors.OpPrereqError("Changing the disk template to a mirrored"
12214 " one requires specifying a secondary node",
12215 errors.ECODE_INVAL)
12217 # Check NIC modifications
12218 self._CheckMods("NIC", self.op.nics, constants.INIC_PARAMS_TYPES,
12219 self._VerifyNicModification)
12221 def ExpandNames(self):
12222 self._ExpandAndLockInstance()
12223 # Can't even acquire node locks in shared mode as upcoming changes in
12224 # Ganeti 2.6 will start to modify the node object on disk conversion
12225 self.needed_locks[locking.LEVEL_NODE] = []
12226 self.needed_locks[locking.LEVEL_NODE_RES] = []
12227 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
12229 def DeclareLocks(self, level):
12230 # TODO: Acquire group lock in shared mode (disk parameters)
12231 if level == locking.LEVEL_NODE:
12232 self._LockInstancesNodes()
12233 if self.op.disk_template and self.op.remote_node:
12234 self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
12235 self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
12236 elif level == locking.LEVEL_NODE_RES and self.op.disk_template:
12238 self.needed_locks[locking.LEVEL_NODE_RES] = \
12239 self.needed_locks[locking.LEVEL_NODE][:]
12241 def BuildHooksEnv(self):
12242 """Build hooks env.
12244 This runs on the master, primary and secondaries.
12248 if constants.BE_MINMEM in self.be_new:
12249 args["minmem"] = self.be_new[constants.BE_MINMEM]
12250 if constants.BE_MAXMEM in self.be_new:
12251 args["maxmem"] = self.be_new[constants.BE_MAXMEM]
12252 if constants.BE_VCPUS in self.be_new:
12253 args["vcpus"] = self.be_new[constants.BE_VCPUS]
12254 # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
12255 # information at all.
12257 if self._new_nics is not None:
12260 for nic in self._new_nics:
12261 nicparams = self.cluster.SimpleFillNIC(nic.nicparams)
12262 mode = nicparams[constants.NIC_MODE]
12263 link = nicparams[constants.NIC_LINK]
12264 nics.append((nic.ip, nic.mac, mode, link))
12266 args["nics"] = nics
12268 env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
12269 if self.op.disk_template:
12270 env["NEW_DISK_TEMPLATE"] = self.op.disk_template
12271 if self.op.runtime_mem:
12272 env["RUNTIME_MEMORY"] = self.op.runtime_mem
12276 def BuildHooksNodes(self):
12277 """Build hooks nodes.
12280 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
12283 def _PrepareNicModification(self, params, private, old_ip, old_params,
12285 update_params_dict = dict([(key, params[key])
12286 for key in constants.NICS_PARAMETERS
12289 if "bridge" in params:
12290 update_params_dict[constants.NIC_LINK] = params["bridge"]
12292 new_params = _GetUpdatedParams(old_params, update_params_dict)
12293 utils.ForceDictType(new_params, constants.NICS_PARAMETER_TYPES)
12295 new_filled_params = cluster.SimpleFillNIC(new_params)
12296 objects.NIC.CheckParameterSyntax(new_filled_params)
12298 new_mode = new_filled_params[constants.NIC_MODE]
12299 if new_mode == constants.NIC_MODE_BRIDGED:
12300 bridge = new_filled_params[constants.NIC_LINK]
12301 msg = self.rpc.call_bridges_exist(pnode, [bridge]).fail_msg
12303 msg = "Error checking bridges on node '%s': %s" % (pnode, msg)
12305 self.warn.append(msg)
12307 raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
12309 elif new_mode == constants.NIC_MODE_ROUTED:
12310 ip = params.get(constants.INIC_IP, old_ip)
12312 raise errors.OpPrereqError("Cannot set the NIC IP address to None"
12313 " on a routed NIC", errors.ECODE_INVAL)
12315 if constants.INIC_MAC in params:
12316 mac = params[constants.INIC_MAC]
12318 raise errors.OpPrereqError("Cannot unset the NIC MAC address",
12319 errors.ECODE_INVAL)
12320 elif mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
12321 # otherwise generate the MAC address
12322 params[constants.INIC_MAC] = \
12323 self.cfg.GenerateMAC(self.proc.GetECId())
12325 # or validate/reserve the current one
12327 self.cfg.ReserveMAC(mac, self.proc.GetECId())
12328 except errors.ReservationError:
12329 raise errors.OpPrereqError("MAC address '%s' already in use"
12330 " in cluster" % mac,
12331 errors.ECODE_NOTUNIQUE)
12333 private.params = new_params
12334 private.filled = new_filled_params
12336 return (None, None)
12338 def CheckPrereq(self):
12339 """Check prerequisites.
12341 This only checks the instance list against the existing names.
12344 # checking the new params on the primary/secondary nodes
12346 instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
12347 cluster = self.cluster = self.cfg.GetClusterInfo()
12348 assert self.instance is not None, \
12349 "Cannot retrieve locked instance %s" % self.op.instance_name
12350 pnode = instance.primary_node
12351 nodelist = list(instance.all_nodes)
12352 pnode_info = self.cfg.GetNodeInfo(pnode)
12353 self.diskparams = self.cfg.GetInstanceDiskParams(instance)
12355 # Prepare disk/NIC modifications
12356 self.diskmod = PrepareContainerMods(self.op.disks, None)
12357 self.nicmod = PrepareContainerMods(self.op.nics, _InstNicModPrivate)
12360 if self.op.os_name and not self.op.force:
12361 _CheckNodeHasOS(self, instance.primary_node, self.op.os_name,
12362 self.op.force_variant)
12363 instance_os = self.op.os_name
12365 instance_os = instance.os
12367 assert not (self.op.disk_template and self.op.disks), \
12368 "Can't modify disk template and apply disk changes at the same time"
12370 if self.op.disk_template:
12371 if instance.disk_template == self.op.disk_template:
12372 raise errors.OpPrereqError("Instance already has disk template %s" %
12373 instance.disk_template, errors.ECODE_INVAL)
12375 if (instance.disk_template,
12376 self.op.disk_template) not in self._DISK_CONVERSIONS:
12377 raise errors.OpPrereqError("Unsupported disk template conversion from"
12378 " %s to %s" % (instance.disk_template,
12379 self.op.disk_template),
12380 errors.ECODE_INVAL)
12381 _CheckInstanceState(self, instance, INSTANCE_DOWN,
12382 msg="cannot change disk template")
12383 if self.op.disk_template in constants.DTS_INT_MIRROR:
12384 if self.op.remote_node == pnode:
12385 raise errors.OpPrereqError("Given new secondary node %s is the same"
12386 " as the primary node of the instance" %
12387 self.op.remote_node, errors.ECODE_STATE)
12388 _CheckNodeOnline(self, self.op.remote_node)
12389 _CheckNodeNotDrained(self, self.op.remote_node)
12390 # FIXME: here we assume that the old instance type is DT_PLAIN
12391 assert instance.disk_template == constants.DT_PLAIN
12392 disks = [{constants.IDISK_SIZE: d.size,
12393 constants.IDISK_VG: d.logical_id[0]}
12394 for d in instance.disks]
12395 required = _ComputeDiskSizePerVG(self.op.disk_template, disks)
12396 _CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required)
12398 snode_info = self.cfg.GetNodeInfo(self.op.remote_node)
12399 snode_group = self.cfg.GetNodeGroup(snode_info.group)
12400 ipolicy = _CalculateGroupIPolicy(cluster, snode_group)
12401 _CheckTargetNodeIPolicy(self, ipolicy, instance, snode_info,
12402 ignore=self.op.ignore_ipolicy)
12403 if pnode_info.group != snode_info.group:
12404 self.LogWarning("The primary and secondary nodes are in two"
12405 " different node groups; the disk parameters"
12406 " from the first disk's node group will be"
12409 # hvparams processing
12410 if self.op.hvparams:
12411 hv_type = instance.hypervisor
12412 i_hvdict = _GetUpdatedParams(instance.hvparams, self.op.hvparams)
12413 utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
12414 hv_new = cluster.SimpleFillHV(hv_type, instance.os, i_hvdict)
12417 hypervisor.GetHypervisor(hv_type).CheckParameterSyntax(hv_new)
12418 _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
12419 self.hv_proposed = self.hv_new = hv_new # the new actual values
12420 self.hv_inst = i_hvdict # the new dict (without defaults)
12422 self.hv_proposed = cluster.SimpleFillHV(instance.hypervisor, instance.os,
12424 self.hv_new = self.hv_inst = {}
12426 # beparams processing
12427 if self.op.beparams:
12428 i_bedict = _GetUpdatedParams(instance.beparams, self.op.beparams,
12430 objects.UpgradeBeParams(i_bedict)
12431 utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
12432 be_new = cluster.SimpleFillBE(i_bedict)
12433 self.be_proposed = self.be_new = be_new # the new actual values
12434 self.be_inst = i_bedict # the new dict (without defaults)
12436 self.be_new = self.be_inst = {}
12437 self.be_proposed = cluster.SimpleFillBE(instance.beparams)
12438 be_old = cluster.FillBE(instance)
12440 # CPU param validation -- checking every time a parameter is
12441 # changed to cover all cases where either CPU mask or vcpus have
12443 if (constants.BE_VCPUS in self.be_proposed and
12444 constants.HV_CPU_MASK in self.hv_proposed):
12446 utils.ParseMultiCpuMask(self.hv_proposed[constants.HV_CPU_MASK])
12447 # Verify mask is consistent with number of vCPUs. Can skip this
12448 # test if only 1 entry in the CPU mask, which means same mask
12449 # is applied to all vCPUs.
12450 if (len(cpu_list) > 1 and
12451 len(cpu_list) != self.be_proposed[constants.BE_VCPUS]):
12452 raise errors.OpPrereqError("Number of vCPUs [%d] does not match the"
12454 (self.be_proposed[constants.BE_VCPUS],
12455 self.hv_proposed[constants.HV_CPU_MASK]),
12456 errors.ECODE_INVAL)
12458 # Only perform this test if a new CPU mask is given
12459 if constants.HV_CPU_MASK in self.hv_new:
12460 # Calculate the largest CPU number requested
12461 max_requested_cpu = max(map(max, cpu_list))
12462 # Check that all of the instance's nodes have enough physical CPUs to
12463 # satisfy the requested CPU mask
12464 _CheckNodesPhysicalCPUs(self, instance.all_nodes,
12465 max_requested_cpu + 1, instance.hypervisor)
12467 # osparams processing
12468 if self.op.osparams:
12469 i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
12470 _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
12471 self.os_inst = i_osdict # the new dict (without defaults)
12477 #TODO(dynmem): do the appropriate check involving MINMEM
12478 if (constants.BE_MAXMEM in self.op.beparams and not self.op.force and
12479 be_new[constants.BE_MAXMEM] > be_old[constants.BE_MAXMEM]):
12480 mem_check_list = [pnode]
12481 if be_new[constants.BE_AUTO_BALANCE]:
12482 # either we changed auto_balance to yes or it was from before
12483 mem_check_list.extend(instance.secondary_nodes)
12484 instance_info = self.rpc.call_instance_info(pnode, instance.name,
12485 instance.hypervisor)
12486 nodeinfo = self.rpc.call_node_info(mem_check_list, None,
12487 [instance.hypervisor])
12488 pninfo = nodeinfo[pnode]
12489 msg = pninfo.fail_msg
12491 # Assume the primary node is unreachable and go ahead
12492 self.warn.append("Can't get info from primary node %s: %s" %
12495 (_, _, (pnhvinfo, )) = pninfo.payload
12496 if not isinstance(pnhvinfo.get("memory_free", None), int):
12497 self.warn.append("Node data from primary node %s doesn't contain"
12498 " free memory information" % pnode)
12499 elif instance_info.fail_msg:
12500 self.warn.append("Can't get instance runtime information: %s" %
12501 instance_info.fail_msg)
12503 if instance_info.payload:
12504 current_mem = int(instance_info.payload["memory"])
12506 # Assume instance not running
12507 # (there is a slight race condition here, but it's not very
12508 # probable, and we have no other way to check)
12509 # TODO: Describe race condition
12511 #TODO(dynmem): do the appropriate check involving MINMEM
12512 miss_mem = (be_new[constants.BE_MAXMEM] - current_mem -
12513 pnhvinfo["memory_free"])
12515 raise errors.OpPrereqError("This change will prevent the instance"
12516 " from starting, due to %d MB of memory"
12517 " missing on its primary node" %
12519 errors.ECODE_NORES)
12521 if be_new[constants.BE_AUTO_BALANCE]:
12522 for node, nres in nodeinfo.items():
12523 if node not in instance.secondary_nodes:
12525 nres.Raise("Can't get info from secondary node %s" % node,
12526 prereq=True, ecode=errors.ECODE_STATE)
12527 (_, _, (nhvinfo, )) = nres.payload
12528 if not isinstance(nhvinfo.get("memory_free", None), int):
12529 raise errors.OpPrereqError("Secondary node %s didn't return free"
12530 " memory information" % node,
12531 errors.ECODE_STATE)
12532 #TODO(dynmem): do the appropriate check involving MINMEM
12533 elif be_new[constants.BE_MAXMEM] > nhvinfo["memory_free"]:
12534 raise errors.OpPrereqError("This change will prevent the instance"
12535 " from failover to its secondary node"
12536 " %s, due to not enough memory" % node,
12537 errors.ECODE_STATE)
12539 if self.op.runtime_mem:
12540 remote_info = self.rpc.call_instance_info(instance.primary_node,
12542 instance.hypervisor)
12543 remote_info.Raise("Error checking node %s" % instance.primary_node)
12544 if not remote_info.payload: # not running already
12545 raise errors.OpPrereqError("Instance %s is not running" % instance.name,
12546 errors.ECODE_STATE)
12548 current_memory = remote_info.payload["memory"]
12549 if (not self.op.force and
12550 (self.op.runtime_mem > self.be_proposed[constants.BE_MAXMEM] or
12551 self.op.runtime_mem < self.be_proposed[constants.BE_MINMEM])):
12552 raise errors.OpPrereqError("Instance %s must have memory between %d"
12553 " and %d MB of memory unless --force is"
12554 " given" % (instance.name,
12555 self.be_proposed[constants.BE_MINMEM],
12556 self.be_proposed[constants.BE_MAXMEM]),
12557 errors.ECODE_INVAL)
12559 if self.op.runtime_mem > current_memory:
12560 _CheckNodeFreeMemory(self, instance.primary_node,
12561 "ballooning memory for instance %s" %
12563 self.op.memory - current_memory,
12564 instance.hypervisor)
12566 if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
12567 raise errors.OpPrereqError("Disk operations not supported for"
12568 " diskless instances",
12569 errors.ECODE_INVAL)
12571 def _PrepareNicCreate(_, params, private):
12572 return self._PrepareNicModification(params, private, None, {},
12575 def _PrepareNicMod(_, nic, params, private):
12576 return self._PrepareNicModification(params, private, nic.ip,
12577 nic.nicparams, cluster, pnode)
12579 # Verify NIC changes (operating on copy)
12580 nics = instance.nics[:]
12581 ApplyContainerMods("NIC", nics, None, self.nicmod,
12582 _PrepareNicCreate, _PrepareNicMod, None)
12583 if len(nics) > constants.MAX_NICS:
12584 raise errors.OpPrereqError("Instance has too many network interfaces"
12585 " (%d), cannot add more" % constants.MAX_NICS,
12586 errors.ECODE_STATE)
12588 # Verify disk changes (operating on a copy)
12589 disks = instance.disks[:]
12590 ApplyContainerMods("disk", disks, None, self.diskmod, None, None, None)
12591 if len(disks) > constants.MAX_DISKS:
12592 raise errors.OpPrereqError("Instance has too many disks (%d), cannot add"
12593 " more" % constants.MAX_DISKS,
12594 errors.ECODE_STATE)
12596 if self.op.offline is not None:
12597 if self.op.offline:
12598 msg = "can't change to offline"
12600 msg = "can't change to online"
12601 _CheckInstanceState(self, instance, CAN_CHANGE_INSTANCE_OFFLINE, msg=msg)
12603 # Pre-compute NIC changes (necessary to use result in hooks)
12604 self._nic_chgdesc = []
12606 # Operate on copies as this is still in prereq
12607 nics = [nic.Copy() for nic in instance.nics]
12608 ApplyContainerMods("NIC", nics, self._nic_chgdesc, self.nicmod,
12609 self._CreateNewNic, self._ApplyNicMods, None)
12610 self._new_nics = nics
12612 self._new_nics = None
12614 def _ConvertPlainToDrbd(self, feedback_fn):
12615 """Converts an instance from plain to drbd.
12618 feedback_fn("Converting template to drbd")
12619 instance = self.instance
12620 pnode = instance.primary_node
12621 snode = self.op.remote_node
12623 assert instance.disk_template == constants.DT_PLAIN
12625 # create a fake disk info for _GenerateDiskTemplate
12626 disk_info = [{constants.IDISK_SIZE: d.size, constants.IDISK_MODE: d.mode,
12627 constants.IDISK_VG: d.logical_id[0]}
12628 for d in instance.disks]
12629 new_disks = _GenerateDiskTemplate(self, self.op.disk_template,
12630 instance.name, pnode, [snode],
12631 disk_info, None, None, 0, feedback_fn,
12633 anno_disks = rpc.AnnotateDiskParams(constants.DT_DRBD8, new_disks,
12635 info = _GetInstanceInfoText(instance)
12636 feedback_fn("Creating additional volumes...")
12637 # first, create the missing data and meta devices
12638 for disk in anno_disks:
12639 # unfortunately this is... not too nice
12640 _CreateSingleBlockDev(self, pnode, instance, disk.children[1],
12642 for child in disk.children:
12643 _CreateSingleBlockDev(self, snode, instance, child, info, True)
12644 # at this stage, all new LVs have been created, we can rename the
12646 feedback_fn("Renaming original volumes...")
12647 rename_list = [(o, n.children[0].logical_id)
12648 for (o, n) in zip(instance.disks, new_disks)]
12649 result = self.rpc.call_blockdev_rename(pnode, rename_list)
12650 result.Raise("Failed to rename original LVs")
12652 feedback_fn("Initializing DRBD devices...")
12653 # all child devices are in place, we can now create the DRBD devices
12654 for disk in anno_disks:
12655 for node in [pnode, snode]:
12656 f_create = node == pnode
12657 _CreateSingleBlockDev(self, node, instance, disk, info, f_create)
12659 # at this point, the instance has been modified
12660 instance.disk_template = constants.DT_DRBD8
12661 instance.disks = new_disks
12662 self.cfg.Update(instance, feedback_fn)
12664 # Release node locks while waiting for sync
12665 _ReleaseLocks(self, locking.LEVEL_NODE)
12667 # disks are created, waiting for sync
12668 disk_abort = not _WaitForSync(self, instance,
12669 oneshot=not self.op.wait_for_sync)
12671 raise errors.OpExecError("There are some degraded disks for"
12672 " this instance, please cleanup manually")
12674 # Node resource locks will be released by caller
12676 def _ConvertDrbdToPlain(self, feedback_fn):
12677 """Converts an instance from drbd to plain.
12680 instance = self.instance
12682 assert len(instance.secondary_nodes) == 1
12683 assert instance.disk_template == constants.DT_DRBD8
12685 pnode = instance.primary_node
12686 snode = instance.secondary_nodes[0]
12687 feedback_fn("Converting template to plain")
12689 old_disks = instance.disks
12690 new_disks = [d.children[0] for d in old_disks]
12692 # copy over size and mode
12693 for parent, child in zip(old_disks, new_disks):
12694 child.size = parent.size
12695 child.mode = parent.mode
12697 # this is a DRBD disk, return its port to the pool
12698 # NOTE: this must be done right before the call to cfg.Update!
12699 for disk in old_disks:
12700 tcp_port = disk.logical_id[2]
12701 self.cfg.AddTcpUdpPort(tcp_port)
12703 # update instance structure
12704 instance.disks = new_disks
12705 instance.disk_template = constants.DT_PLAIN
12706 self.cfg.Update(instance, feedback_fn)
12708 # Release locks in case removing disks takes a while
12709 _ReleaseLocks(self, locking.LEVEL_NODE)
12711 feedback_fn("Removing volumes on the secondary node...")
12712 for disk in old_disks:
12713 self.cfg.SetDiskID(disk, snode)
12714 msg = self.rpc.call_blockdev_remove(snode, disk).fail_msg
12716 self.LogWarning("Could not remove block device %s on node %s,"
12717 " continuing anyway: %s", disk.iv_name, snode, msg)
12719 feedback_fn("Removing unneeded volumes on the primary node...")
12720 for idx, disk in enumerate(old_disks):
12721 meta = disk.children[1]
12722 self.cfg.SetDiskID(meta, pnode)
12723 msg = self.rpc.call_blockdev_remove(pnode, meta).fail_msg
12725 self.LogWarning("Could not remove metadata for disk %d on node %s,"
12726 " continuing anyway: %s", idx, pnode, msg)
12728 def _CreateNewDisk(self, idx, params, _):
12729 """Creates a new disk.
12732 instance = self.instance
12735 if instance.disk_template in constants.DTS_FILEBASED:
12736 (file_driver, file_path) = instance.disks[0].logical_id
12737 file_path = os.path.dirname(file_path)
12739 file_driver = file_path = None
12742 _GenerateDiskTemplate(self, instance.disk_template, instance.name,
12743 instance.primary_node, instance.secondary_nodes,
12744 [params], file_path, file_driver, idx,
12745 self.Log, self.diskparams)[0]
12747 info = _GetInstanceInfoText(instance)
12749 logging.info("Creating volume %s for instance %s",
12750 disk.iv_name, instance.name)
12751 # Note: this needs to be kept in sync with _CreateDisks
12753 for node in instance.all_nodes:
12754 f_create = (node == instance.primary_node)
12756 _CreateBlockDev(self, node, instance, disk, f_create, info, f_create)
12757 except errors.OpExecError, err:
12758 self.LogWarning("Failed to create volume %s (%s) on node '%s': %s",
12759 disk.iv_name, disk, node, err)
12762 ("disk/%d" % idx, "add:size=%s,mode=%s" % (disk.size, disk.mode)),
12766 def _ModifyDisk(idx, disk, params, _):
12767 """Modifies a disk.
12770 disk.mode = params[constants.IDISK_MODE]
12773 ("disk.mode/%d" % idx, disk.mode),
12776 def _RemoveDisk(self, idx, root, _):
12780 for node, disk in root.ComputeNodeTree(self.instance.primary_node):
12781 self.cfg.SetDiskID(disk, node)
12782 msg = self.rpc.call_blockdev_remove(node, disk).fail_msg
12784 self.LogWarning("Could not remove disk/%d on node '%s': %s,"
12785 " continuing anyway", idx, node, msg)
12787 # if this is a DRBD disk, return its port to the pool
12788 if root.dev_type in constants.LDS_DRBD:
12789 self.cfg.AddTcpUdpPort(root.logical_id[2])
12792 def _CreateNewNic(idx, params, private):
12793 """Creates data structure for a new network interface.
12796 mac = params[constants.INIC_MAC]
12797 ip = params.get(constants.INIC_IP, None)
12798 nicparams = private.params
12800 return (objects.NIC(mac=mac, ip=ip, nicparams=nicparams), [
12802 "add:mac=%s,ip=%s,mode=%s,link=%s" %
12803 (mac, ip, private.filled[constants.NIC_MODE],
12804 private.filled[constants.NIC_LINK])),
12808 def _ApplyNicMods(idx, nic, params, private):
12809 """Modifies a network interface.
12814 for key in [constants.INIC_MAC, constants.INIC_IP]:
12816 changes.append(("nic.%s/%d" % (key, idx), params[key]))
12817 setattr(nic, key, params[key])
12820 nic.nicparams = private.params
12822 for (key, val) in params.items():
12823 changes.append(("nic.%s/%d" % (key, idx), val))
12827 def Exec(self, feedback_fn):
12828 """Modifies an instance.
12830 All parameters take effect only at the next restart of the instance.
12833 # Process here the warnings from CheckPrereq, as we don't have a
12834 # feedback_fn there.
12835 # TODO: Replace with self.LogWarning
12836 for warn in self.warn:
12837 feedback_fn("WARNING: %s" % warn)
12839 assert ((self.op.disk_template is None) ^
12840 bool(self.owned_locks(locking.LEVEL_NODE_RES))), \
12841 "Not owning any node resource locks"
12844 instance = self.instance
12847 if self.op.runtime_mem:
12848 rpcres = self.rpc.call_instance_balloon_memory(instance.primary_node,
12850 self.op.runtime_mem)
12851 rpcres.Raise("Cannot modify instance runtime memory")
12852 result.append(("runtime_memory", self.op.runtime_mem))
12854 # Apply disk changes
12855 ApplyContainerMods("disk", instance.disks, result, self.diskmod,
12856 self._CreateNewDisk, self._ModifyDisk, self._RemoveDisk)
12857 _UpdateIvNames(0, instance.disks)
12859 if self.op.disk_template:
12861 check_nodes = set(instance.all_nodes)
12862 if self.op.remote_node:
12863 check_nodes.add(self.op.remote_node)
12864 for level in [locking.LEVEL_NODE, locking.LEVEL_NODE_RES]:
12865 owned = self.owned_locks(level)
12866 assert not (check_nodes - owned), \
12867 ("Not owning the correct locks, owning %r, expected at least %r" %
12868 (owned, check_nodes))
12870 r_shut = _ShutdownInstanceDisks(self, instance)
12872 raise errors.OpExecError("Cannot shutdown instance disks, unable to"
12873 " proceed with disk template conversion")
12874 mode = (instance.disk_template, self.op.disk_template)
12876 self._DISK_CONVERSIONS[mode](self, feedback_fn)
12878 self.cfg.ReleaseDRBDMinors(instance.name)
12880 result.append(("disk_template", self.op.disk_template))
12882 assert instance.disk_template == self.op.disk_template, \
12883 ("Expected disk template '%s', found '%s'" %
12884 (self.op.disk_template, instance.disk_template))
12886 # Release node and resource locks if there are any (they might already have
12887 # been released during disk conversion)
12888 _ReleaseLocks(self, locking.LEVEL_NODE)
12889 _ReleaseLocks(self, locking.LEVEL_NODE_RES)
12891 # Apply NIC changes
12892 if self._new_nics is not None:
12893 instance.nics = self._new_nics
12894 result.extend(self._nic_chgdesc)
12897 if self.op.hvparams:
12898 instance.hvparams = self.hv_inst
12899 for key, val in self.op.hvparams.iteritems():
12900 result.append(("hv/%s" % key, val))
12903 if self.op.beparams:
12904 instance.beparams = self.be_inst
12905 for key, val in self.op.beparams.iteritems():
12906 result.append(("be/%s" % key, val))
12909 if self.op.os_name:
12910 instance.os = self.op.os_name
12913 if self.op.osparams:
12914 instance.osparams = self.os_inst
12915 for key, val in self.op.osparams.iteritems():
12916 result.append(("os/%s" % key, val))
12918 if self.op.offline is None:
12921 elif self.op.offline:
12922 # Mark instance as offline
12923 self.cfg.MarkInstanceOffline(instance.name)
12924 result.append(("admin_state", constants.ADMINST_OFFLINE))
12926 # Mark instance as online, but stopped
12927 self.cfg.MarkInstanceDown(instance.name)
12928 result.append(("admin_state", constants.ADMINST_DOWN))
12930 self.cfg.Update(instance, feedback_fn)
12932 assert not (self.owned_locks(locking.LEVEL_NODE_RES) or
12933 self.owned_locks(locking.LEVEL_NODE)), \
12934 "All node locks should have been released by now"
12938 _DISK_CONVERSIONS = {
12939 (constants.DT_PLAIN, constants.DT_DRBD8): _ConvertPlainToDrbd,
12940 (constants.DT_DRBD8, constants.DT_PLAIN): _ConvertDrbdToPlain,
12944 class LUInstanceChangeGroup(LogicalUnit):
12945 HPATH = "instance-change-group"
12946 HTYPE = constants.HTYPE_INSTANCE
12949 def ExpandNames(self):
12950 self.share_locks = _ShareAll()
12951 self.needed_locks = {
12952 locking.LEVEL_NODEGROUP: [],
12953 locking.LEVEL_NODE: [],
12956 self._ExpandAndLockInstance()
12958 if self.op.target_groups:
12959 self.req_target_uuids = map(self.cfg.LookupNodeGroup,
12960 self.op.target_groups)
12962 self.req_target_uuids = None
12964 self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
12966 def DeclareLocks(self, level):
12967 if level == locking.LEVEL_NODEGROUP:
12968 assert not self.needed_locks[locking.LEVEL_NODEGROUP]
12970 if self.req_target_uuids:
12971 lock_groups = set(self.req_target_uuids)
12973 # Lock all groups used by instance optimistically; this requires going
12974 # via the node before it's locked, requiring verification later on
12975 instance_groups = self.cfg.GetInstanceNodeGroups(self.op.instance_name)
12976 lock_groups.update(instance_groups)
12978 # No target groups, need to lock all of them
12979 lock_groups = locking.ALL_SET
12981 self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
12983 elif level == locking.LEVEL_NODE:
12984 if self.req_target_uuids:
12985 # Lock all nodes used by instances
12986 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
12987 self._LockInstancesNodes()
12989 # Lock all nodes in all potential target groups
12990 lock_groups = (frozenset(self.owned_locks(locking.LEVEL_NODEGROUP)) -
12991 self.cfg.GetInstanceNodeGroups(self.op.instance_name))
12992 member_nodes = [node_name
12993 for group in lock_groups
12994 for node_name in self.cfg.GetNodeGroup(group).members]
12995 self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
12997 # Lock all nodes as all groups are potential targets
12998 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
13000 def CheckPrereq(self):
13001 owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
13002 owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
13003 owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
13005 assert (self.req_target_uuids is None or
13006 owned_groups.issuperset(self.req_target_uuids))
13007 assert owned_instances == set([self.op.instance_name])
13009 # Get instance information
13010 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
13012 # Check if node groups for locked instance are still correct
13013 assert owned_nodes.issuperset(self.instance.all_nodes), \
13014 ("Instance %s's nodes changed while we kept the lock" %
13015 self.op.instance_name)
13017 inst_groups = _CheckInstanceNodeGroups(self.cfg, self.op.instance_name,
13020 if self.req_target_uuids:
13021 # User requested specific target groups
13022 self.target_uuids = frozenset(self.req_target_uuids)
13024 # All groups except those used by the instance are potential targets
13025 self.target_uuids = owned_groups - inst_groups
13027 conflicting_groups = self.target_uuids & inst_groups
13028 if conflicting_groups:
13029 raise errors.OpPrereqError("Can't use group(s) '%s' as targets, they are"
13030 " used by the instance '%s'" %
13031 (utils.CommaJoin(conflicting_groups),
13032 self.op.instance_name),
13033 errors.ECODE_INVAL)
13035 if not self.target_uuids:
13036 raise errors.OpPrereqError("There are no possible target groups",
13037 errors.ECODE_INVAL)
13039 def BuildHooksEnv(self):
13040 """Build hooks env.
13043 assert self.target_uuids
13046 "TARGET_GROUPS": " ".join(self.target_uuids),
13049 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
13053 def BuildHooksNodes(self):
13054 """Build hooks nodes.
13057 mn = self.cfg.GetMasterNode()
13058 return ([mn], [mn])
13060 def Exec(self, feedback_fn):
13061 instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
13063 assert instances == [self.op.instance_name], "Instance not locked"
13065 ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_CHG_GROUP,
13066 instances=instances, target_groups=list(self.target_uuids))
13068 ial.Run(self.op.iallocator)
13070 if not ial.success:
13071 raise errors.OpPrereqError("Can't compute solution for changing group of"
13072 " instance '%s' using iallocator '%s': %s" %
13073 (self.op.instance_name, self.op.iallocator,
13075 errors.ECODE_NORES)
13077 jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, False)
13079 self.LogInfo("Iallocator returned %s job(s) for changing group of"
13080 " instance '%s'", len(jobs), self.op.instance_name)
13082 return ResultWithJobs(jobs)
13085 class LUBackupQuery(NoHooksLU):
13086 """Query the exports list
13091 def CheckArguments(self):
13092 self.expq = _ExportQuery(qlang.MakeSimpleFilter("node", self.op.nodes),
13093 ["node", "export"], self.op.use_locking)
13095 def ExpandNames(self):
13096 self.expq.ExpandNames(self)
13098 def DeclareLocks(self, level):
13099 self.expq.DeclareLocks(self, level)
13101 def Exec(self, feedback_fn):
13104 for (node, expname) in self.expq.OldStyleQuery(self):
13105 if expname is None:
13106 result[node] = False
13108 result.setdefault(node, []).append(expname)
13113 class _ExportQuery(_QueryBase):
13114 FIELDS = query.EXPORT_FIELDS
13116 #: The node name is not a unique key for this query
13117 SORT_FIELD = "node"
13119 def ExpandNames(self, lu):
13120 lu.needed_locks = {}
13122 # The following variables interact with _QueryBase._GetNames
13124 self.wanted = _GetWantedNodes(lu, self.names)
13126 self.wanted = locking.ALL_SET
13128 self.do_locking = self.use_locking
13130 if self.do_locking:
13131 lu.share_locks = _ShareAll()
13132 lu.needed_locks = {
13133 locking.LEVEL_NODE: self.wanted,
13136 def DeclareLocks(self, lu, level):
13139 def _GetQueryData(self, lu):
13140 """Computes the list of nodes and their attributes.
13143 # Locking is not used
13145 assert not (compat.any(lu.glm.is_owned(level)
13146 for level in locking.LEVELS
13147 if level != locking.LEVEL_CLUSTER) or
13148 self.do_locking or self.use_locking)
13150 nodes = self._GetNames(lu, lu.cfg.GetNodeList(), locking.LEVEL_NODE)
13154 for (node, nres) in lu.rpc.call_export_list(nodes).items():
13156 result.append((node, None))
13158 result.extend((node, expname) for expname in nres.payload)
13163 class LUBackupPrepare(NoHooksLU):
13164 """Prepares an instance for an export and returns useful information.
13169 def ExpandNames(self):
13170 self._ExpandAndLockInstance()
13172 def CheckPrereq(self):
13173 """Check prerequisites.
13176 instance_name = self.op.instance_name
13178 self.instance = self.cfg.GetInstanceInfo(instance_name)
13179 assert self.instance is not None, \
13180 "Cannot retrieve locked instance %s" % self.op.instance_name
13181 _CheckNodeOnline(self, self.instance.primary_node)
13183 self._cds = _GetClusterDomainSecret()
13185 def Exec(self, feedback_fn):
13186 """Prepares an instance for an export.
13189 instance = self.instance
13191 if self.op.mode == constants.EXPORT_MODE_REMOTE:
13192 salt = utils.GenerateSecret(8)
13194 feedback_fn("Generating X509 certificate on %s" % instance.primary_node)
13195 result = self.rpc.call_x509_cert_create(instance.primary_node,
13196 constants.RIE_CERT_VALIDITY)
13197 result.Raise("Can't create X509 key and certificate on %s" % result.node)
13199 (name, cert_pem) = result.payload
13201 cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
13205 "handshake": masterd.instance.ComputeRemoteExportHandshake(self._cds),
13206 "x509_key_name": (name, utils.Sha1Hmac(self._cds, name, salt=salt),
13208 "x509_ca": utils.SignX509Certificate(cert, self._cds, salt),
13214 class LUBackupExport(LogicalUnit):
13215 """Export an instance to an image in the cluster.
13218 HPATH = "instance-export"
13219 HTYPE = constants.HTYPE_INSTANCE
13222 def CheckArguments(self):
13223 """Check the arguments.
13226 self.x509_key_name = self.op.x509_key_name
13227 self.dest_x509_ca_pem = self.op.destination_x509_ca
13229 if self.op.mode == constants.EXPORT_MODE_REMOTE:
13230 if not self.x509_key_name:
13231 raise errors.OpPrereqError("Missing X509 key name for encryption",
13232 errors.ECODE_INVAL)
13234 if not self.dest_x509_ca_pem:
13235 raise errors.OpPrereqError("Missing destination X509 CA",
13236 errors.ECODE_INVAL)
13238 def ExpandNames(self):
13239 self._ExpandAndLockInstance()
13241 # Lock all nodes for local exports
13242 if self.op.mode == constants.EXPORT_MODE_LOCAL:
13243 # FIXME: lock only instance primary and destination node
13245 # Sad but true, for now we have do lock all nodes, as we don't know where
13246 # the previous export might be, and in this LU we search for it and
13247 # remove it from its current node. In the future we could fix this by:
13248 # - making a tasklet to search (share-lock all), then create the
13249 # new one, then one to remove, after
13250 # - removing the removal operation altogether
13251 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
13253 def DeclareLocks(self, level):
13254 """Last minute lock declaration."""
13255 # All nodes are locked anyway, so nothing to do here.
13257 def BuildHooksEnv(self):
13258 """Build hooks env.
13260 This will run on the master, primary node and target node.
13264 "EXPORT_MODE": self.op.mode,
13265 "EXPORT_NODE": self.op.target_node,
13266 "EXPORT_DO_SHUTDOWN": self.op.shutdown,
13267 "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
13268 # TODO: Generic function for boolean env variables
13269 "REMOVE_INSTANCE": str(bool(self.op.remove_instance)),
13272 env.update(_BuildInstanceHookEnvByObject(self, self.instance))
13276 def BuildHooksNodes(self):
13277 """Build hooks nodes.
13280 nl = [self.cfg.GetMasterNode(), self.instance.primary_node]
13282 if self.op.mode == constants.EXPORT_MODE_LOCAL:
13283 nl.append(self.op.target_node)
13287 def CheckPrereq(self):
13288 """Check prerequisites.
13290 This checks that the instance and node names are valid.
13293 instance_name = self.op.instance_name
13295 self.instance = self.cfg.GetInstanceInfo(instance_name)
13296 assert self.instance is not None, \
13297 "Cannot retrieve locked instance %s" % self.op.instance_name
13298 _CheckNodeOnline(self, self.instance.primary_node)
13300 if (self.op.remove_instance and
13301 self.instance.admin_state == constants.ADMINST_UP and
13302 not self.op.shutdown):
13303 raise errors.OpPrereqError("Can not remove instance without shutting it"
13306 if self.op.mode == constants.EXPORT_MODE_LOCAL:
13307 self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
13308 self.dst_node = self.cfg.GetNodeInfo(self.op.target_node)
13309 assert self.dst_node is not None
13311 _CheckNodeOnline(self, self.dst_node.name)
13312 _CheckNodeNotDrained(self, self.dst_node.name)
13315 self.dest_disk_info = None
13316 self.dest_x509_ca = None
13318 elif self.op.mode == constants.EXPORT_MODE_REMOTE:
13319 self.dst_node = None
13321 if len(self.op.target_node) != len(self.instance.disks):
13322 raise errors.OpPrereqError(("Received destination information for %s"
13323 " disks, but instance %s has %s disks") %
13324 (len(self.op.target_node), instance_name,
13325 len(self.instance.disks)),
13326 errors.ECODE_INVAL)
13328 cds = _GetClusterDomainSecret()
13330 # Check X509 key name
13332 (key_name, hmac_digest, hmac_salt) = self.x509_key_name
13333 except (TypeError, ValueError), err:
13334 raise errors.OpPrereqError("Invalid data for X509 key name: %s" % err)
13336 if not utils.VerifySha1Hmac(cds, key_name, hmac_digest, salt=hmac_salt):
13337 raise errors.OpPrereqError("HMAC for X509 key name is wrong",
13338 errors.ECODE_INVAL)
13340 # Load and verify CA
13342 (cert, _) = utils.LoadSignedX509Certificate(self.dest_x509_ca_pem, cds)
13343 except OpenSSL.crypto.Error, err:
13344 raise errors.OpPrereqError("Unable to load destination X509 CA (%s)" %
13345 (err, ), errors.ECODE_INVAL)
13347 (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
13348 if errcode is not None:
13349 raise errors.OpPrereqError("Invalid destination X509 CA (%s)" %
13350 (msg, ), errors.ECODE_INVAL)
13352 self.dest_x509_ca = cert
13354 # Verify target information
13356 for idx, disk_data in enumerate(self.op.target_node):
13358 (host, port, magic) = \
13359 masterd.instance.CheckRemoteExportDiskInfo(cds, idx, disk_data)
13360 except errors.GenericError, err:
13361 raise errors.OpPrereqError("Target info for disk %s: %s" %
13362 (idx, err), errors.ECODE_INVAL)
13364 disk_info.append((host, port, magic))
13366 assert len(disk_info) == len(self.op.target_node)
13367 self.dest_disk_info = disk_info
13370 raise errors.ProgrammerError("Unhandled export mode %r" %
13373 # instance disk type verification
13374 # TODO: Implement export support for file-based disks
13375 for disk in self.instance.disks:
13376 if disk.dev_type == constants.LD_FILE:
13377 raise errors.OpPrereqError("Export not supported for instances with"
13378 " file-based disks", errors.ECODE_INVAL)
13380 def _CleanupExports(self, feedback_fn):
13381 """Removes exports of current instance from all other nodes.
13383 If an instance in a cluster with nodes A..D was exported to node C, its
13384 exports will be removed from the nodes A, B and D.
13387 assert self.op.mode != constants.EXPORT_MODE_REMOTE
13389 nodelist = self.cfg.GetNodeList()
13390 nodelist.remove(self.dst_node.name)
13392 # on one-node clusters nodelist will be empty after the removal
13393 # if we proceed the backup would be removed because OpBackupQuery
13394 # substitutes an empty list with the full cluster node list.
13395 iname = self.instance.name
13397 feedback_fn("Removing old exports for instance %s" % iname)
13398 exportlist = self.rpc.call_export_list(nodelist)
13399 for node in exportlist:
13400 if exportlist[node].fail_msg:
13402 if iname in exportlist[node].payload:
13403 msg = self.rpc.call_export_remove(node, iname).fail_msg
13405 self.LogWarning("Could not remove older export for instance %s"
13406 " on node %s: %s", iname, node, msg)
13408 def Exec(self, feedback_fn):
13409 """Export an instance to an image in the cluster.
13412 assert self.op.mode in constants.EXPORT_MODES
13414 instance = self.instance
13415 src_node = instance.primary_node
13417 if self.op.shutdown:
13418 # shutdown the instance, but not the disks
13419 feedback_fn("Shutting down instance %s" % instance.name)
13420 result = self.rpc.call_instance_shutdown(src_node, instance,
13421 self.op.shutdown_timeout)
13422 # TODO: Maybe ignore failures if ignore_remove_failures is set
13423 result.Raise("Could not shutdown instance %s on"
13424 " node %s" % (instance.name, src_node))
13426 # set the disks ID correctly since call_instance_start needs the
13427 # correct drbd minor to create the symlinks
13428 for disk in instance.disks:
13429 self.cfg.SetDiskID(disk, src_node)
13431 activate_disks = (instance.admin_state != constants.ADMINST_UP)
13434 # Activate the instance disks if we'exporting a stopped instance
13435 feedback_fn("Activating disks for %s" % instance.name)
13436 _StartInstanceDisks(self, instance, None)
13439 helper = masterd.instance.ExportInstanceHelper(self, feedback_fn,
13442 helper.CreateSnapshots()
13444 if (self.op.shutdown and
13445 instance.admin_state == constants.ADMINST_UP and
13446 not self.op.remove_instance):
13447 assert not activate_disks
13448 feedback_fn("Starting instance %s" % instance.name)
13449 result = self.rpc.call_instance_start(src_node,
13450 (instance, None, None), False)
13451 msg = result.fail_msg
13453 feedback_fn("Failed to start instance: %s" % msg)
13454 _ShutdownInstanceDisks(self, instance)
13455 raise errors.OpExecError("Could not start instance: %s" % msg)
13457 if self.op.mode == constants.EXPORT_MODE_LOCAL:
13458 (fin_resu, dresults) = helper.LocalExport(self.dst_node)
13459 elif self.op.mode == constants.EXPORT_MODE_REMOTE:
13460 connect_timeout = constants.RIE_CONNECT_TIMEOUT
13461 timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
13463 (key_name, _, _) = self.x509_key_name
13466 OpenSSL.crypto.dump_certificate(OpenSSL.crypto.FILETYPE_PEM,
13469 (fin_resu, dresults) = helper.RemoteExport(self.dest_disk_info,
13470 key_name, dest_ca_pem,
13475 # Check for backwards compatibility
13476 assert len(dresults) == len(instance.disks)
13477 assert compat.all(isinstance(i, bool) for i in dresults), \
13478 "Not all results are boolean: %r" % dresults
13482 feedback_fn("Deactivating disks for %s" % instance.name)
13483 _ShutdownInstanceDisks(self, instance)
13485 if not (compat.all(dresults) and fin_resu):
13488 failures.append("export finalization")
13489 if not compat.all(dresults):
13490 fdsk = utils.CommaJoin(idx for (idx, dsk) in enumerate(dresults)
13492 failures.append("disk export: disk(s) %s" % fdsk)
13494 raise errors.OpExecError("Export failed, errors in %s" %
13495 utils.CommaJoin(failures))
13497 # At this point, the export was successful, we can cleanup/finish
13499 # Remove instance if requested
13500 if self.op.remove_instance:
13501 feedback_fn("Removing instance %s" % instance.name)
13502 _RemoveInstance(self, feedback_fn, instance,
13503 self.op.ignore_remove_failures)
13505 if self.op.mode == constants.EXPORT_MODE_LOCAL:
13506 self._CleanupExports(feedback_fn)
13508 return fin_resu, dresults
13511 class LUBackupRemove(NoHooksLU):
13512 """Remove exports related to the named instance.
13517 def ExpandNames(self):
13518 self.needed_locks = {}
13519 # We need all nodes to be locked in order for RemoveExport to work, but we
13520 # don't need to lock the instance itself, as nothing will happen to it (and
13521 # we can remove exports also for a removed instance)
13522 self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
13524 def Exec(self, feedback_fn):
13525 """Remove any export.
13528 instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
13529 # If the instance was not found we'll try with the name that was passed in.
13530 # This will only work if it was an FQDN, though.
13532 if not instance_name:
13534 instance_name = self.op.instance_name
13536 locked_nodes = self.owned_locks(locking.LEVEL_NODE)
13537 exportlist = self.rpc.call_export_list(locked_nodes)
13539 for node in exportlist:
13540 msg = exportlist[node].fail_msg
13542 self.LogWarning("Failed to query node %s (continuing): %s", node, msg)
13544 if instance_name in exportlist[node].payload:
13546 result = self.rpc.call_export_remove(node, instance_name)
13547 msg = result.fail_msg
13549 logging.error("Could not remove export for instance %s"
13550 " on node %s: %s", instance_name, node, msg)
13552 if fqdn_warn and not found:
13553 feedback_fn("Export not found. If trying to remove an export belonging"
13554 " to a deleted instance please use its Fully Qualified"
13558 class LUGroupAdd(LogicalUnit):
13559 """Logical unit for creating node groups.
13562 HPATH = "group-add"
13563 HTYPE = constants.HTYPE_GROUP
13566 def ExpandNames(self):
13567 # We need the new group's UUID here so that we can create and acquire the
13568 # corresponding lock. Later, in Exec(), we'll indicate to cfg.AddNodeGroup
13569 # that it should not check whether the UUID exists in the configuration.
13570 self.group_uuid = self.cfg.GenerateUniqueID(self.proc.GetECId())
13571 self.needed_locks = {}
13572 self.add_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
13574 def CheckPrereq(self):
13575 """Check prerequisites.
13577 This checks that the given group name is not an existing node group
13582 existing_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
13583 except errors.OpPrereqError:
13586 raise errors.OpPrereqError("Desired group name '%s' already exists as a"
13587 " node group (UUID: %s)" %
13588 (self.op.group_name, existing_uuid),
13589 errors.ECODE_EXISTS)
13591 if self.op.ndparams:
13592 utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
13594 if self.op.hv_state:
13595 self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state, None)
13597 self.new_hv_state = None
13599 if self.op.disk_state:
13600 self.new_disk_state = _MergeAndVerifyDiskState(self.op.disk_state, None)
13602 self.new_disk_state = None
13604 if self.op.diskparams:
13605 for templ in constants.DISK_TEMPLATES:
13606 if templ in self.op.diskparams:
13607 utils.ForceDictType(self.op.diskparams[templ],
13608 constants.DISK_DT_TYPES)
13609 self.new_diskparams = self.op.diskparams
13611 self.new_diskparams = {}
13613 if self.op.ipolicy:
13614 cluster = self.cfg.GetClusterInfo()
13615 full_ipolicy = cluster.SimpleFillIPolicy(self.op.ipolicy)
13617 objects.InstancePolicy.CheckParameterSyntax(full_ipolicy)
13618 except errors.ConfigurationError, err:
13619 raise errors.OpPrereqError("Invalid instance policy: %s" % err,
13620 errors.ECODE_INVAL)
13622 def BuildHooksEnv(self):
13623 """Build hooks env.
13627 "GROUP_NAME": self.op.group_name,
13630 def BuildHooksNodes(self):
13631 """Build hooks nodes.
13634 mn = self.cfg.GetMasterNode()
13635 return ([mn], [mn])
13637 def Exec(self, feedback_fn):
13638 """Add the node group to the cluster.
13641 group_obj = objects.NodeGroup(name=self.op.group_name, members=[],
13642 uuid=self.group_uuid,
13643 alloc_policy=self.op.alloc_policy,
13644 ndparams=self.op.ndparams,
13645 diskparams=self.new_diskparams,
13646 ipolicy=self.op.ipolicy,
13647 hv_state_static=self.new_hv_state,
13648 disk_state_static=self.new_disk_state)
13650 self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False)
13651 del self.remove_locks[locking.LEVEL_NODEGROUP]
13654 class LUGroupAssignNodes(NoHooksLU):
13655 """Logical unit for assigning nodes to groups.
13660 def ExpandNames(self):
13661 # These raise errors.OpPrereqError on their own:
13662 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
13663 self.op.nodes = _GetWantedNodes(self, self.op.nodes)
13665 # We want to lock all the affected nodes and groups. We have readily
13666 # available the list of nodes, and the *destination* group. To gather the
13667 # list of "source" groups, we need to fetch node information later on.
13668 self.needed_locks = {
13669 locking.LEVEL_NODEGROUP: set([self.group_uuid]),
13670 locking.LEVEL_NODE: self.op.nodes,
13673 def DeclareLocks(self, level):
13674 if level == locking.LEVEL_NODEGROUP:
13675 assert len(self.needed_locks[locking.LEVEL_NODEGROUP]) == 1
13677 # Try to get all affected nodes' groups without having the group or node
13678 # lock yet. Needs verification later in the code flow.
13679 groups = self.cfg.GetNodeGroupsFromNodes(self.op.nodes)
13681 self.needed_locks[locking.LEVEL_NODEGROUP].update(groups)
13683 def CheckPrereq(self):
13684 """Check prerequisites.
13687 assert self.needed_locks[locking.LEVEL_NODEGROUP]
13688 assert (frozenset(self.owned_locks(locking.LEVEL_NODE)) ==
13689 frozenset(self.op.nodes))
13691 expected_locks = (set([self.group_uuid]) |
13692 self.cfg.GetNodeGroupsFromNodes(self.op.nodes))
13693 actual_locks = self.owned_locks(locking.LEVEL_NODEGROUP)
13694 if actual_locks != expected_locks:
13695 raise errors.OpExecError("Nodes changed groups since locks were acquired,"
13696 " current groups are '%s', used to be '%s'" %
13697 (utils.CommaJoin(expected_locks),
13698 utils.CommaJoin(actual_locks)))
13700 self.node_data = self.cfg.GetAllNodesInfo()
13701 self.group = self.cfg.GetNodeGroup(self.group_uuid)
13702 instance_data = self.cfg.GetAllInstancesInfo()
13704 if self.group is None:
13705 raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
13706 (self.op.group_name, self.group_uuid))
13708 (new_splits, previous_splits) = \
13709 self.CheckAssignmentForSplitInstances([(node, self.group_uuid)
13710 for node in self.op.nodes],
13711 self.node_data, instance_data)
13714 fmt_new_splits = utils.CommaJoin(utils.NiceSort(new_splits))
13716 if not self.op.force:
13717 raise errors.OpExecError("The following instances get split by this"
13718 " change and --force was not given: %s" %
13721 self.LogWarning("This operation will split the following instances: %s",
13724 if previous_splits:
13725 self.LogWarning("In addition, these already-split instances continue"
13726 " to be split across groups: %s",
13727 utils.CommaJoin(utils.NiceSort(previous_splits)))
13729 def Exec(self, feedback_fn):
13730 """Assign nodes to a new group.
13733 mods = [(node_name, self.group_uuid) for node_name in self.op.nodes]
13735 self.cfg.AssignGroupNodes(mods)
13738 def CheckAssignmentForSplitInstances(changes, node_data, instance_data):
13739 """Check for split instances after a node assignment.
13741 This method considers a series of node assignments as an atomic operation,
13742 and returns information about split instances after applying the set of
13745 In particular, it returns information about newly split instances, and
13746 instances that were already split, and remain so after the change.
13748 Only instances whose disk template is listed in constants.DTS_INT_MIRROR are
13751 @type changes: list of (node_name, new_group_uuid) pairs.
13752 @param changes: list of node assignments to consider.
13753 @param node_data: a dict with data for all nodes
13754 @param instance_data: a dict with all instances to consider
13755 @rtype: a two-tuple
13756 @return: a list of instances that were previously okay and result split as a
13757 consequence of this change, and a list of instances that were previously
13758 split and this change does not fix.
13761 changed_nodes = dict((node, group) for node, group in changes
13762 if node_data[node].group != group)
13764 all_split_instances = set()
13765 previously_split_instances = set()
13767 def InstanceNodes(instance):
13768 return [instance.primary_node] + list(instance.secondary_nodes)
13770 for inst in instance_data.values():
13771 if inst.disk_template not in constants.DTS_INT_MIRROR:
13774 instance_nodes = InstanceNodes(inst)
13776 if len(set(node_data[node].group for node in instance_nodes)) > 1:
13777 previously_split_instances.add(inst.name)
13779 if len(set(changed_nodes.get(node, node_data[node].group)
13780 for node in instance_nodes)) > 1:
13781 all_split_instances.add(inst.name)
13783 return (list(all_split_instances - previously_split_instances),
13784 list(previously_split_instances & all_split_instances))
13787 class _GroupQuery(_QueryBase):
13788 FIELDS = query.GROUP_FIELDS
13790 def ExpandNames(self, lu):
13791 lu.needed_locks = {}
13793 self._all_groups = lu.cfg.GetAllNodeGroupsInfo()
13794 self._cluster = lu.cfg.GetClusterInfo()
13795 name_to_uuid = dict((g.name, g.uuid) for g in self._all_groups.values())
13798 self.wanted = [name_to_uuid[name]
13799 for name in utils.NiceSort(name_to_uuid.keys())]
13801 # Accept names to be either names or UUIDs.
13804 all_uuid = frozenset(self._all_groups.keys())
13806 for name in self.names:
13807 if name in all_uuid:
13808 self.wanted.append(name)
13809 elif name in name_to_uuid:
13810 self.wanted.append(name_to_uuid[name])
13812 missing.append(name)
13815 raise errors.OpPrereqError("Some groups do not exist: %s" %
13816 utils.CommaJoin(missing),
13817 errors.ECODE_NOENT)
13819 def DeclareLocks(self, lu, level):
13822 def _GetQueryData(self, lu):
13823 """Computes the list of node groups and their attributes.
13826 do_nodes = query.GQ_NODE in self.requested_data
13827 do_instances = query.GQ_INST in self.requested_data
13829 group_to_nodes = None
13830 group_to_instances = None
13832 # For GQ_NODE, we need to map group->[nodes], and group->[instances] for
13833 # GQ_INST. The former is attainable with just GetAllNodesInfo(), but for the
13834 # latter GetAllInstancesInfo() is not enough, for we have to go through
13835 # instance->node. Hence, we will need to process nodes even if we only need
13836 # instance information.
13837 if do_nodes or do_instances:
13838 all_nodes = lu.cfg.GetAllNodesInfo()
13839 group_to_nodes = dict((uuid, []) for uuid in self.wanted)
13842 for node in all_nodes.values():
13843 if node.group in group_to_nodes:
13844 group_to_nodes[node.group].append(node.name)
13845 node_to_group[node.name] = node.group
13848 all_instances = lu.cfg.GetAllInstancesInfo()
13849 group_to_instances = dict((uuid, []) for uuid in self.wanted)
13851 for instance in all_instances.values():
13852 node = instance.primary_node
13853 if node in node_to_group:
13854 group_to_instances[node_to_group[node]].append(instance.name)
13857 # Do not pass on node information if it was not requested.
13858 group_to_nodes = None
13860 return query.GroupQueryData(self._cluster,
13861 [self._all_groups[uuid]
13862 for uuid in self.wanted],
13863 group_to_nodes, group_to_instances,
13864 query.GQ_DISKPARAMS in self.requested_data)
13867 class LUGroupQuery(NoHooksLU):
13868 """Logical unit for querying node groups.
13873 def CheckArguments(self):
13874 self.gq = _GroupQuery(qlang.MakeSimpleFilter("name", self.op.names),
13875 self.op.output_fields, False)
13877 def ExpandNames(self):
13878 self.gq.ExpandNames(self)
13880 def DeclareLocks(self, level):
13881 self.gq.DeclareLocks(self, level)
13883 def Exec(self, feedback_fn):
13884 return self.gq.OldStyleQuery(self)
13887 class LUGroupSetParams(LogicalUnit):
13888 """Modifies the parameters of a node group.
13891 HPATH = "group-modify"
13892 HTYPE = constants.HTYPE_GROUP
13895 def CheckArguments(self):
13898 self.op.diskparams,
13899 self.op.alloc_policy,
13901 self.op.disk_state,
13905 if all_changes.count(None) == len(all_changes):
13906 raise errors.OpPrereqError("Please pass at least one modification",
13907 errors.ECODE_INVAL)
13909 def ExpandNames(self):
13910 # This raises errors.OpPrereqError on its own:
13911 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
13913 self.needed_locks = {
13914 locking.LEVEL_INSTANCE: [],
13915 locking.LEVEL_NODEGROUP: [self.group_uuid],
13918 self.share_locks[locking.LEVEL_INSTANCE] = 1
13920 def DeclareLocks(self, level):
13921 if level == locking.LEVEL_INSTANCE:
13922 assert not self.needed_locks[locking.LEVEL_INSTANCE]
13924 # Lock instances optimistically, needs verification once group lock has
13926 self.needed_locks[locking.LEVEL_INSTANCE] = \
13927 self.cfg.GetNodeGroupInstances(self.group_uuid)
13930 def _UpdateAndVerifyDiskParams(old, new):
13931 """Updates and verifies disk parameters.
13934 new_params = _GetUpdatedParams(old, new)
13935 utils.ForceDictType(new_params, constants.DISK_DT_TYPES)
13938 def CheckPrereq(self):
13939 """Check prerequisites.
13942 owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
13944 # Check if locked instances are still correct
13945 _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
13947 self.group = self.cfg.GetNodeGroup(self.group_uuid)
13948 cluster = self.cfg.GetClusterInfo()
13950 if self.group is None:
13951 raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
13952 (self.op.group_name, self.group_uuid))
13954 if self.op.ndparams:
13955 new_ndparams = _GetUpdatedParams(self.group.ndparams, self.op.ndparams)
13956 utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
13957 self.new_ndparams = new_ndparams
13959 if self.op.diskparams:
13960 diskparams = self.group.diskparams
13961 uavdp = self._UpdateAndVerifyDiskParams
13962 # For each disktemplate subdict update and verify the values
13963 new_diskparams = dict((dt,
13964 uavdp(diskparams.get(dt, {}),
13965 self.op.diskparams[dt]))
13966 for dt in constants.DISK_TEMPLATES
13967 if dt in self.op.diskparams)
13968 # As we've all subdicts of diskparams ready, lets merge the actual
13969 # dict with all updated subdicts
13970 self.new_diskparams = objects.FillDict(diskparams, new_diskparams)
13972 if self.op.hv_state:
13973 self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state,
13974 self.group.hv_state_static)
13976 if self.op.disk_state:
13977 self.new_disk_state = \
13978 _MergeAndVerifyDiskState(self.op.disk_state,
13979 self.group.disk_state_static)
13981 if self.op.ipolicy:
13982 self.new_ipolicy = _GetUpdatedIPolicy(self.group.ipolicy,
13986 new_ipolicy = cluster.SimpleFillIPolicy(self.new_ipolicy)
13987 inst_filter = lambda inst: inst.name in owned_instances
13988 instances = self.cfg.GetInstancesInfoByFilter(inst_filter).values()
13990 _ComputeNewInstanceViolations(_CalculateGroupIPolicy(cluster,
13992 new_ipolicy, instances)
13995 self.LogWarning("After the ipolicy change the following instances"
13996 " violate them: %s",
13997 utils.CommaJoin(violations))
13999 def BuildHooksEnv(self):
14000 """Build hooks env.
14004 "GROUP_NAME": self.op.group_name,
14005 "NEW_ALLOC_POLICY": self.op.alloc_policy,
14008 def BuildHooksNodes(self):
14009 """Build hooks nodes.
14012 mn = self.cfg.GetMasterNode()
14013 return ([mn], [mn])
14015 def Exec(self, feedback_fn):
14016 """Modifies the node group.
14021 if self.op.ndparams:
14022 self.group.ndparams = self.new_ndparams
14023 result.append(("ndparams", str(self.group.ndparams)))
14025 if self.op.diskparams:
14026 self.group.diskparams = self.new_diskparams
14027 result.append(("diskparams", str(self.group.diskparams)))
14029 if self.op.alloc_policy:
14030 self.group.alloc_policy = self.op.alloc_policy
14032 if self.op.hv_state:
14033 self.group.hv_state_static = self.new_hv_state
14035 if self.op.disk_state:
14036 self.group.disk_state_static = self.new_disk_state
14038 if self.op.ipolicy:
14039 self.group.ipolicy = self.new_ipolicy
14041 self.cfg.Update(self.group, feedback_fn)
14045 class LUGroupRemove(LogicalUnit):
14046 HPATH = "group-remove"
14047 HTYPE = constants.HTYPE_GROUP
14050 def ExpandNames(self):
14051 # This will raises errors.OpPrereqError on its own:
14052 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
14053 self.needed_locks = {
14054 locking.LEVEL_NODEGROUP: [self.group_uuid],
14057 def CheckPrereq(self):
14058 """Check prerequisites.
14060 This checks that the given group name exists as a node group, that is
14061 empty (i.e., contains no nodes), and that is not the last group of the
14065 # Verify that the group is empty.
14066 group_nodes = [node.name
14067 for node in self.cfg.GetAllNodesInfo().values()
14068 if node.group == self.group_uuid]
14071 raise errors.OpPrereqError("Group '%s' not empty, has the following"
14073 (self.op.group_name,
14074 utils.CommaJoin(utils.NiceSort(group_nodes))),
14075 errors.ECODE_STATE)
14077 # Verify the cluster would not be left group-less.
14078 if len(self.cfg.GetNodeGroupList()) == 1:
14079 raise errors.OpPrereqError("Group '%s' is the only group,"
14080 " cannot be removed" %
14081 self.op.group_name,
14082 errors.ECODE_STATE)
14084 def BuildHooksEnv(self):
14085 """Build hooks env.
14089 "GROUP_NAME": self.op.group_name,
14092 def BuildHooksNodes(self):
14093 """Build hooks nodes.
14096 mn = self.cfg.GetMasterNode()
14097 return ([mn], [mn])
14099 def Exec(self, feedback_fn):
14100 """Remove the node group.
14104 self.cfg.RemoveNodeGroup(self.group_uuid)
14105 except errors.ConfigurationError:
14106 raise errors.OpExecError("Group '%s' with UUID %s disappeared" %
14107 (self.op.group_name, self.group_uuid))
14109 self.remove_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
14112 class LUGroupRename(LogicalUnit):
14113 HPATH = "group-rename"
14114 HTYPE = constants.HTYPE_GROUP
14117 def ExpandNames(self):
14118 # This raises errors.OpPrereqError on its own:
14119 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
14121 self.needed_locks = {
14122 locking.LEVEL_NODEGROUP: [self.group_uuid],
14125 def CheckPrereq(self):
14126 """Check prerequisites.
14128 Ensures requested new name is not yet used.
14132 new_name_uuid = self.cfg.LookupNodeGroup(self.op.new_name)
14133 except errors.OpPrereqError:
14136 raise errors.OpPrereqError("Desired new name '%s' clashes with existing"
14137 " node group (UUID: %s)" %
14138 (self.op.new_name, new_name_uuid),
14139 errors.ECODE_EXISTS)
14141 def BuildHooksEnv(self):
14142 """Build hooks env.
14146 "OLD_NAME": self.op.group_name,
14147 "NEW_NAME": self.op.new_name,
14150 def BuildHooksNodes(self):
14151 """Build hooks nodes.
14154 mn = self.cfg.GetMasterNode()
14156 all_nodes = self.cfg.GetAllNodesInfo()
14157 all_nodes.pop(mn, None)
14160 run_nodes.extend(node.name for node in all_nodes.values()
14161 if node.group == self.group_uuid)
14163 return (run_nodes, run_nodes)
14165 def Exec(self, feedback_fn):
14166 """Rename the node group.
14169 group = self.cfg.GetNodeGroup(self.group_uuid)
14172 raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
14173 (self.op.group_name, self.group_uuid))
14175 group.name = self.op.new_name
14176 self.cfg.Update(group, feedback_fn)
14178 return self.op.new_name
14181 class LUGroupEvacuate(LogicalUnit):
14182 HPATH = "group-evacuate"
14183 HTYPE = constants.HTYPE_GROUP
14186 def ExpandNames(self):
14187 # This raises errors.OpPrereqError on its own:
14188 self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
14190 if self.op.target_groups:
14191 self.req_target_uuids = map(self.cfg.LookupNodeGroup,
14192 self.op.target_groups)
14194 self.req_target_uuids = []
14196 if self.group_uuid in self.req_target_uuids:
14197 raise errors.OpPrereqError("Group to be evacuated (%s) can not be used"
14198 " as a target group (targets are %s)" %
14200 utils.CommaJoin(self.req_target_uuids)),
14201 errors.ECODE_INVAL)
14203 self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
14205 self.share_locks = _ShareAll()
14206 self.needed_locks = {
14207 locking.LEVEL_INSTANCE: [],
14208 locking.LEVEL_NODEGROUP: [],
14209 locking.LEVEL_NODE: [],
14212 def DeclareLocks(self, level):
14213 if level == locking.LEVEL_INSTANCE:
14214 assert not self.needed_locks[locking.LEVEL_INSTANCE]
14216 # Lock instances optimistically, needs verification once node and group
14217 # locks have been acquired
14218 self.needed_locks[locking.LEVEL_INSTANCE] = \
14219 self.cfg.GetNodeGroupInstances(self.group_uuid)
14221 elif level == locking.LEVEL_NODEGROUP:
14222 assert not self.needed_locks[locking.LEVEL_NODEGROUP]
14224 if self.req_target_uuids:
14225 lock_groups = set([self.group_uuid] + self.req_target_uuids)
14227 # Lock all groups used by instances optimistically; this requires going
14228 # via the node before it's locked, requiring verification later on
14229 lock_groups.update(group_uuid
14230 for instance_name in
14231 self.owned_locks(locking.LEVEL_INSTANCE)
14233 self.cfg.GetInstanceNodeGroups(instance_name))
14235 # No target groups, need to lock all of them
14236 lock_groups = locking.ALL_SET
14238 self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
14240 elif level == locking.LEVEL_NODE:
14241 # This will only lock the nodes in the group to be evacuated which
14242 # contain actual instances
14243 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
14244 self._LockInstancesNodes()
14246 # Lock all nodes in group to be evacuated and target groups
14247 owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
14248 assert self.group_uuid in owned_groups
14249 member_nodes = [node_name
14250 for group in owned_groups
14251 for node_name in self.cfg.GetNodeGroup(group).members]
14252 self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
14254 def CheckPrereq(self):
14255 owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
14256 owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
14257 owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
14259 assert owned_groups.issuperset(self.req_target_uuids)
14260 assert self.group_uuid in owned_groups
14262 # Check if locked instances are still correct
14263 _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
14265 # Get instance information
14266 self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
14268 # Check if node groups for locked instances are still correct
14269 _CheckInstancesNodeGroups(self.cfg, self.instances,
14270 owned_groups, owned_nodes, self.group_uuid)
14272 if self.req_target_uuids:
14273 # User requested specific target groups
14274 self.target_uuids = self.req_target_uuids
14276 # All groups except the one to be evacuated are potential targets
14277 self.target_uuids = [group_uuid for group_uuid in owned_groups
14278 if group_uuid != self.group_uuid]
14280 if not self.target_uuids:
14281 raise errors.OpPrereqError("There are no possible target groups",
14282 errors.ECODE_INVAL)
14284 def BuildHooksEnv(self):
14285 """Build hooks env.
14289 "GROUP_NAME": self.op.group_name,
14290 "TARGET_GROUPS": " ".join(self.target_uuids),
14293 def BuildHooksNodes(self):
14294 """Build hooks nodes.
14297 mn = self.cfg.GetMasterNode()
14299 assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
14301 run_nodes = [mn] + self.cfg.GetNodeGroup(self.group_uuid).members
14303 return (run_nodes, run_nodes)
14305 def Exec(self, feedback_fn):
14306 instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
14308 assert self.group_uuid not in self.target_uuids
14310 ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_CHG_GROUP,
14311 instances=instances, target_groups=self.target_uuids)
14313 ial.Run(self.op.iallocator)
14315 if not ial.success:
14316 raise errors.OpPrereqError("Can't compute group evacuation using"
14317 " iallocator '%s': %s" %
14318 (self.op.iallocator, ial.info),
14319 errors.ECODE_NORES)
14321 jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, False)
14323 self.LogInfo("Iallocator returned %s job(s) for evacuating node group %s",
14324 len(jobs), self.op.group_name)
14326 return ResultWithJobs(jobs)
14329 class TagsLU(NoHooksLU): # pylint: disable=W0223
14330 """Generic tags LU.
14332 This is an abstract class which is the parent of all the other tags LUs.
14335 def ExpandNames(self):
14336 self.group_uuid = None
14337 self.needed_locks = {}
14339 if self.op.kind == constants.TAG_NODE:
14340 self.op.name = _ExpandNodeName(self.cfg, self.op.name)
14341 lock_level = locking.LEVEL_NODE
14342 lock_name = self.op.name
14343 elif self.op.kind == constants.TAG_INSTANCE:
14344 self.op.name = _ExpandInstanceName(self.cfg, self.op.name)
14345 lock_level = locking.LEVEL_INSTANCE
14346 lock_name = self.op.name
14347 elif self.op.kind == constants.TAG_NODEGROUP:
14348 self.group_uuid = self.cfg.LookupNodeGroup(self.op.name)
14349 lock_level = locking.LEVEL_NODEGROUP
14350 lock_name = self.group_uuid
14355 if lock_level and getattr(self.op, "use_locking", True):
14356 self.needed_locks[lock_level] = lock_name
14358 # FIXME: Acquire BGL for cluster tag operations (as of this writing it's
14359 # not possible to acquire the BGL based on opcode parameters)
14361 def CheckPrereq(self):
14362 """Check prerequisites.
14365 if self.op.kind == constants.TAG_CLUSTER:
14366 self.target = self.cfg.GetClusterInfo()
14367 elif self.op.kind == constants.TAG_NODE:
14368 self.target = self.cfg.GetNodeInfo(self.op.name)
14369 elif self.op.kind == constants.TAG_INSTANCE:
14370 self.target = self.cfg.GetInstanceInfo(self.op.name)
14371 elif self.op.kind == constants.TAG_NODEGROUP:
14372 self.target = self.cfg.GetNodeGroup(self.group_uuid)
14374 raise errors.OpPrereqError("Wrong tag type requested (%s)" %
14375 str(self.op.kind), errors.ECODE_INVAL)
14378 class LUTagsGet(TagsLU):
14379 """Returns the tags of a given object.
14384 def ExpandNames(self):
14385 TagsLU.ExpandNames(self)
14387 # Share locks as this is only a read operation
14388 self.share_locks = _ShareAll()
14390 def Exec(self, feedback_fn):
14391 """Returns the tag list.
14394 return list(self.target.GetTags())
14397 class LUTagsSearch(NoHooksLU):
14398 """Searches the tags for a given pattern.
14403 def ExpandNames(self):
14404 self.needed_locks = {}
14406 def CheckPrereq(self):
14407 """Check prerequisites.
14409 This checks the pattern passed for validity by compiling it.
14413 self.re = re.compile(self.op.pattern)
14414 except re.error, err:
14415 raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
14416 (self.op.pattern, err), errors.ECODE_INVAL)
14418 def Exec(self, feedback_fn):
14419 """Returns the tag list.
14423 tgts = [("/cluster", cfg.GetClusterInfo())]
14424 ilist = cfg.GetAllInstancesInfo().values()
14425 tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
14426 nlist = cfg.GetAllNodesInfo().values()
14427 tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
14428 tgts.extend(("/nodegroup/%s" % n.name, n)
14429 for n in cfg.GetAllNodeGroupsInfo().values())
14431 for path, target in tgts:
14432 for tag in target.GetTags():
14433 if self.re.search(tag):
14434 results.append((path, tag))
14438 class LUTagsSet(TagsLU):
14439 """Sets a tag on a given object.
14444 def CheckPrereq(self):
14445 """Check prerequisites.
14447 This checks the type and length of the tag name and value.
14450 TagsLU.CheckPrereq(self)
14451 for tag in self.op.tags:
14452 objects.TaggableObject.ValidateTag(tag)
14454 def Exec(self, feedback_fn):
14459 for tag in self.op.tags:
14460 self.target.AddTag(tag)
14461 except errors.TagError, err:
14462 raise errors.OpExecError("Error while setting tag: %s" % str(err))
14463 self.cfg.Update(self.target, feedback_fn)
14466 class LUTagsDel(TagsLU):
14467 """Delete a list of tags from a given object.
14472 def CheckPrereq(self):
14473 """Check prerequisites.
14475 This checks that we have the given tag.
14478 TagsLU.CheckPrereq(self)
14479 for tag in self.op.tags:
14480 objects.TaggableObject.ValidateTag(tag)
14481 del_tags = frozenset(self.op.tags)
14482 cur_tags = self.target.GetTags()
14484 diff_tags = del_tags - cur_tags
14486 diff_names = ("'%s'" % i for i in sorted(diff_tags))
14487 raise errors.OpPrereqError("Tag(s) %s not found" %
14488 (utils.CommaJoin(diff_names), ),
14489 errors.ECODE_NOENT)
14491 def Exec(self, feedback_fn):
14492 """Remove the tag from the object.
14495 for tag in self.op.tags:
14496 self.target.RemoveTag(tag)
14497 self.cfg.Update(self.target, feedback_fn)
14500 class LUTestDelay(NoHooksLU):
14501 """Sleep for a specified amount of time.
14503 This LU sleeps on the master and/or nodes for a specified amount of
14509 def ExpandNames(self):
14510 """Expand names and set required locks.
14512 This expands the node list, if any.
14515 self.needed_locks = {}
14516 if self.op.on_nodes:
14517 # _GetWantedNodes can be used here, but is not always appropriate to use
14518 # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
14519 # more information.
14520 self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
14521 self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
14523 def _TestDelay(self):
14524 """Do the actual sleep.
14527 if self.op.on_master:
14528 if not utils.TestDelay(self.op.duration):
14529 raise errors.OpExecError("Error during master delay test")
14530 if self.op.on_nodes:
14531 result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
14532 for node, node_result in result.items():
14533 node_result.Raise("Failure during rpc call to node %s" % node)
14535 def Exec(self, feedback_fn):
14536 """Execute the test delay opcode, with the wanted repetitions.
14539 if self.op.repeat == 0:
14542 top_value = self.op.repeat - 1
14543 for i in range(self.op.repeat):
14544 self.LogInfo("Test delay iteration %d/%d" % (i, top_value))
14548 class LUTestJqueue(NoHooksLU):
14549 """Utility LU to test some aspects of the job queue.
14554 # Must be lower than default timeout for WaitForJobChange to see whether it
14555 # notices changed jobs
14556 _CLIENT_CONNECT_TIMEOUT = 20.0
14557 _CLIENT_CONFIRM_TIMEOUT = 60.0
14560 def _NotifyUsingSocket(cls, cb, errcls):
14561 """Opens a Unix socket and waits for another program to connect.
14564 @param cb: Callback to send socket name to client
14565 @type errcls: class
14566 @param errcls: Exception class to use for errors
14569 # Using a temporary directory as there's no easy way to create temporary
14570 # sockets without writing a custom loop around tempfile.mktemp and
14572 tmpdir = tempfile.mkdtemp()
14574 tmpsock = utils.PathJoin(tmpdir, "sock")
14576 logging.debug("Creating temporary socket at %s", tmpsock)
14577 sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
14582 # Send details to client
14585 # Wait for client to connect before continuing
14586 sock.settimeout(cls._CLIENT_CONNECT_TIMEOUT)
14588 (conn, _) = sock.accept()
14589 except socket.error, err:
14590 raise errcls("Client didn't connect in time (%s)" % err)
14594 # Remove as soon as client is connected
14595 shutil.rmtree(tmpdir)
14597 # Wait for client to close
14600 # pylint: disable=E1101
14601 # Instance of '_socketobject' has no ... member
14602 conn.settimeout(cls._CLIENT_CONFIRM_TIMEOUT)
14604 except socket.error, err:
14605 raise errcls("Client failed to confirm notification (%s)" % err)
14609 def _SendNotification(self, test, arg, sockname):
14610 """Sends a notification to the client.
14613 @param test: Test name
14614 @param arg: Test argument (depends on test)
14615 @type sockname: string
14616 @param sockname: Socket path
14619 self.Log(constants.ELOG_JQUEUE_TEST, (sockname, test, arg))
14621 def _Notify(self, prereq, test, arg):
14622 """Notifies the client of a test.
14625 @param prereq: Whether this is a prereq-phase test
14627 @param test: Test name
14628 @param arg: Test argument (depends on test)
14632 errcls = errors.OpPrereqError
14634 errcls = errors.OpExecError
14636 return self._NotifyUsingSocket(compat.partial(self._SendNotification,
14640 def CheckArguments(self):
14641 self.checkargs_calls = getattr(self, "checkargs_calls", 0) + 1
14642 self.expandnames_calls = 0
14644 def ExpandNames(self):
14645 checkargs_calls = getattr(self, "checkargs_calls", 0)
14646 if checkargs_calls < 1:
14647 raise errors.ProgrammerError("CheckArguments was not called")
14649 self.expandnames_calls += 1
14651 if self.op.notify_waitlock:
14652 self._Notify(True, constants.JQT_EXPANDNAMES, None)
14654 self.LogInfo("Expanding names")
14656 # Get lock on master node (just to get a lock, not for a particular reason)
14657 self.needed_locks = {
14658 locking.LEVEL_NODE: self.cfg.GetMasterNode(),
14661 def Exec(self, feedback_fn):
14662 if self.expandnames_calls < 1:
14663 raise errors.ProgrammerError("ExpandNames was not called")
14665 if self.op.notify_exec:
14666 self._Notify(False, constants.JQT_EXEC, None)
14668 self.LogInfo("Executing")
14670 if self.op.log_messages:
14671 self._Notify(False, constants.JQT_STARTMSG, len(self.op.log_messages))
14672 for idx, msg in enumerate(self.op.log_messages):
14673 self.LogInfo("Sending log message %s", idx + 1)
14674 feedback_fn(constants.JQT_MSGPREFIX + msg)
14675 # Report how many test messages have been sent
14676 self._Notify(False, constants.JQT_LOGMSG, idx + 1)
14679 raise errors.OpExecError("Opcode failure was requested")
14684 class IAllocator(object):
14685 """IAllocator framework.
14687 An IAllocator instance has three sets of attributes:
14688 - cfg that is needed to query the cluster
14689 - input data (all members of the _KEYS class attribute are required)
14690 - four buffer attributes (in|out_data|text), that represent the
14691 input (to the external script) in text and data structure format,
14692 and the output from it, again in two formats
14693 - the result variables from the script (success, info, nodes) for
14697 # pylint: disable=R0902
14698 # lots of instance attributes
14700 def __init__(self, cfg, rpc_runner, mode, **kwargs):
14702 self.rpc = rpc_runner
14703 # init buffer variables
14704 self.in_text = self.out_text = self.in_data = self.out_data = None
14705 # init all input fields so that pylint is happy
14707 self.memory = self.disks = self.disk_template = self.spindle_use = None
14708 self.os = self.tags = self.nics = self.vcpus = None
14709 self.hypervisor = None
14710 self.relocate_from = None
14712 self.instances = None
14713 self.evac_mode = None
14714 self.target_groups = []
14716 self.required_nodes = None
14717 # init result fields
14718 self.success = self.info = self.result = None
14721 (fn, keydata, self._result_check) = self._MODE_DATA[self.mode]
14723 raise errors.ProgrammerError("Unknown mode '%s' passed to the"
14724 " IAllocator" % self.mode)
14726 keyset = [n for (n, _) in keydata]
14729 if key not in keyset:
14730 raise errors.ProgrammerError("Invalid input parameter '%s' to"
14731 " IAllocator" % key)
14732 setattr(self, key, kwargs[key])
14735 if key not in kwargs:
14736 raise errors.ProgrammerError("Missing input parameter '%s' to"
14737 " IAllocator" % key)
14738 self._BuildInputData(compat.partial(fn, self), keydata)
14740 def _ComputeClusterData(self):
14741 """Compute the generic allocator input data.
14743 This is the data that is independent of the actual operation.
14747 cluster_info = cfg.GetClusterInfo()
14750 "version": constants.IALLOCATOR_VERSION,
14751 "cluster_name": cfg.GetClusterName(),
14752 "cluster_tags": list(cluster_info.GetTags()),
14753 "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
14754 "ipolicy": cluster_info.ipolicy,
14756 ninfo = cfg.GetAllNodesInfo()
14757 iinfo = cfg.GetAllInstancesInfo().values()
14758 i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
14761 node_list = [n.name for n in ninfo.values() if n.vm_capable]
14763 if self.mode == constants.IALLOCATOR_MODE_ALLOC:
14764 hypervisor_name = self.hypervisor
14765 elif self.mode == constants.IALLOCATOR_MODE_RELOC:
14766 hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
14768 hypervisor_name = cluster_info.primary_hypervisor
14770 node_data = self.rpc.call_node_info(node_list, [cfg.GetVGName()],
14773 self.rpc.call_all_instances_info(node_list,
14774 cluster_info.enabled_hypervisors)
14776 data["nodegroups"] = self._ComputeNodeGroupData(cfg)
14778 config_ndata = self._ComputeBasicNodeData(cfg, ninfo)
14779 data["nodes"] = self._ComputeDynamicNodeData(ninfo, node_data, node_iinfo,
14780 i_list, config_ndata)
14781 assert len(data["nodes"]) == len(ninfo), \
14782 "Incomplete node data computed"
14784 data["instances"] = self._ComputeInstanceData(cluster_info, i_list)
14786 self.in_data = data
14789 def _ComputeNodeGroupData(cfg):
14790 """Compute node groups data.
14793 cluster = cfg.GetClusterInfo()
14794 ng = dict((guuid, {
14795 "name": gdata.name,
14796 "alloc_policy": gdata.alloc_policy,
14797 "ipolicy": _CalculateGroupIPolicy(cluster, gdata),
14799 for guuid, gdata in cfg.GetAllNodeGroupsInfo().items())
14804 def _ComputeBasicNodeData(cfg, node_cfg):
14805 """Compute global node data.
14808 @returns: a dict of name: (node dict, node config)
14811 # fill in static (config-based) values
14812 node_results = dict((ninfo.name, {
14813 "tags": list(ninfo.GetTags()),
14814 "primary_ip": ninfo.primary_ip,
14815 "secondary_ip": ninfo.secondary_ip,
14816 "offline": ninfo.offline,
14817 "drained": ninfo.drained,
14818 "master_candidate": ninfo.master_candidate,
14819 "group": ninfo.group,
14820 "master_capable": ninfo.master_capable,
14821 "vm_capable": ninfo.vm_capable,
14822 "ndparams": cfg.GetNdParams(ninfo),
14824 for ninfo in node_cfg.values())
14826 return node_results
14829 def _ComputeDynamicNodeData(node_cfg, node_data, node_iinfo, i_list,
14831 """Compute global node data.
14833 @param node_results: the basic node structures as filled from the config
14836 #TODO(dynmem): compute the right data on MAX and MIN memory
14837 # make a copy of the current dict
14838 node_results = dict(node_results)
14839 for nname, nresult in node_data.items():
14840 assert nname in node_results, "Missing basic data for node %s" % nname
14841 ninfo = node_cfg[nname]
14843 if not (ninfo.offline or ninfo.drained):
14844 nresult.Raise("Can't get data for node %s" % nname)
14845 node_iinfo[nname].Raise("Can't get node instance info from node %s" %
14847 remote_info = _MakeLegacyNodeInfo(nresult.payload)
14849 for attr in ["memory_total", "memory_free", "memory_dom0",
14850 "vg_size", "vg_free", "cpu_total"]:
14851 if attr not in remote_info:
14852 raise errors.OpExecError("Node '%s' didn't return attribute"
14853 " '%s'" % (nname, attr))
14854 if not isinstance(remote_info[attr], int):
14855 raise errors.OpExecError("Node '%s' returned invalid value"
14857 (nname, attr, remote_info[attr]))
14858 # compute memory used by primary instances
14859 i_p_mem = i_p_up_mem = 0
14860 for iinfo, beinfo in i_list:
14861 if iinfo.primary_node == nname:
14862 i_p_mem += beinfo[constants.BE_MAXMEM]
14863 if iinfo.name not in node_iinfo[nname].payload:
14866 i_used_mem = int(node_iinfo[nname].payload[iinfo.name]["memory"])
14867 i_mem_diff = beinfo[constants.BE_MAXMEM] - i_used_mem
14868 remote_info["memory_free"] -= max(0, i_mem_diff)
14870 if iinfo.admin_state == constants.ADMINST_UP:
14871 i_p_up_mem += beinfo[constants.BE_MAXMEM]
14873 # compute memory used by instances
14875 "total_memory": remote_info["memory_total"],
14876 "reserved_memory": remote_info["memory_dom0"],
14877 "free_memory": remote_info["memory_free"],
14878 "total_disk": remote_info["vg_size"],
14879 "free_disk": remote_info["vg_free"],
14880 "total_cpus": remote_info["cpu_total"],
14881 "i_pri_memory": i_p_mem,
14882 "i_pri_up_memory": i_p_up_mem,
14884 pnr_dyn.update(node_results[nname])
14885 node_results[nname] = pnr_dyn
14887 return node_results
14890 def _ComputeInstanceData(cluster_info, i_list):
14891 """Compute global instance data.
14895 for iinfo, beinfo in i_list:
14897 for nic in iinfo.nics:
14898 filled_params = cluster_info.SimpleFillNIC(nic.nicparams)
14902 "mode": filled_params[constants.NIC_MODE],
14903 "link": filled_params[constants.NIC_LINK],
14905 if filled_params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
14906 nic_dict["bridge"] = filled_params[constants.NIC_LINK]
14907 nic_data.append(nic_dict)
14909 "tags": list(iinfo.GetTags()),
14910 "admin_state": iinfo.admin_state,
14911 "vcpus": beinfo[constants.BE_VCPUS],
14912 "memory": beinfo[constants.BE_MAXMEM],
14913 "spindle_use": beinfo[constants.BE_SPINDLE_USE],
14915 "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
14917 "disks": [{constants.IDISK_SIZE: dsk.size,
14918 constants.IDISK_MODE: dsk.mode}
14919 for dsk in iinfo.disks],
14920 "disk_template": iinfo.disk_template,
14921 "hypervisor": iinfo.hypervisor,
14923 pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
14925 instance_data[iinfo.name] = pir
14927 return instance_data
14929 def _AddNewInstance(self):
14930 """Add new instance data to allocator structure.
14932 This in combination with _AllocatorGetClusterData will create the
14933 correct structure needed as input for the allocator.
14935 The checks for the completeness of the opcode must have already been
14939 disk_space = _ComputeDiskSize(self.disk_template, self.disks)
14941 if self.disk_template in constants.DTS_INT_MIRROR:
14942 self.required_nodes = 2
14944 self.required_nodes = 1
14948 "disk_template": self.disk_template,
14951 "vcpus": self.vcpus,
14952 "memory": self.memory,
14953 "spindle_use": self.spindle_use,
14954 "disks": self.disks,
14955 "disk_space_total": disk_space,
14957 "required_nodes": self.required_nodes,
14958 "hypervisor": self.hypervisor,
14963 def _AddRelocateInstance(self):
14964 """Add relocate instance data to allocator structure.
14966 This in combination with _IAllocatorGetClusterData will create the
14967 correct structure needed as input for the allocator.
14969 The checks for the completeness of the opcode must have already been
14973 instance = self.cfg.GetInstanceInfo(self.name)
14974 if instance is None:
14975 raise errors.ProgrammerError("Unknown instance '%s' passed to"
14976 " IAllocator" % self.name)
14978 if instance.disk_template not in constants.DTS_MIRRORED:
14979 raise errors.OpPrereqError("Can't relocate non-mirrored instances",
14980 errors.ECODE_INVAL)
14982 if instance.disk_template in constants.DTS_INT_MIRROR and \
14983 len(instance.secondary_nodes) != 1:
14984 raise errors.OpPrereqError("Instance has not exactly one secondary node",
14985 errors.ECODE_STATE)
14987 self.required_nodes = 1
14988 disk_sizes = [{constants.IDISK_SIZE: disk.size} for disk in instance.disks]
14989 disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
14993 "disk_space_total": disk_space,
14994 "required_nodes": self.required_nodes,
14995 "relocate_from": self.relocate_from,
14999 def _AddNodeEvacuate(self):
15000 """Get data for node-evacuate requests.
15004 "instances": self.instances,
15005 "evac_mode": self.evac_mode,
15008 def _AddChangeGroup(self):
15009 """Get data for node-evacuate requests.
15013 "instances": self.instances,
15014 "target_groups": self.target_groups,
15017 def _BuildInputData(self, fn, keydata):
15018 """Build input data structures.
15021 self._ComputeClusterData()
15024 request["type"] = self.mode
15025 for keyname, keytype in keydata:
15026 if keyname not in request:
15027 raise errors.ProgrammerError("Request parameter %s is missing" %
15029 val = request[keyname]
15030 if not keytype(val):
15031 raise errors.ProgrammerError("Request parameter %s doesn't pass"
15032 " validation, value %s, expected"
15033 " type %s" % (keyname, val, keytype))
15034 self.in_data["request"] = request
15036 self.in_text = serializer.Dump(self.in_data)
15038 _STRING_LIST = ht.TListOf(ht.TString)
15039 _JOB_LIST = ht.TListOf(ht.TListOf(ht.TStrictDict(True, False, {
15040 # pylint: disable=E1101
15041 # Class '...' has no 'OP_ID' member
15042 "OP_ID": ht.TElemOf([opcodes.OpInstanceFailover.OP_ID,
15043 opcodes.OpInstanceMigrate.OP_ID,
15044 opcodes.OpInstanceReplaceDisks.OP_ID])
15048 ht.TListOf(ht.TAnd(ht.TIsLength(3),
15049 ht.TItems([ht.TNonEmptyString,
15050 ht.TNonEmptyString,
15051 ht.TListOf(ht.TNonEmptyString),
15054 ht.TListOf(ht.TAnd(ht.TIsLength(2),
15055 ht.TItems([ht.TNonEmptyString,
15058 _NEVAC_RESULT = ht.TAnd(ht.TIsLength(3),
15059 ht.TItems([_NEVAC_MOVED, _NEVAC_FAILED, _JOB_LIST]))
15062 constants.IALLOCATOR_MODE_ALLOC:
15065 ("name", ht.TString),
15066 ("memory", ht.TInt),
15067 ("spindle_use", ht.TInt),
15068 ("disks", ht.TListOf(ht.TDict)),
15069 ("disk_template", ht.TString),
15070 ("os", ht.TString),
15071 ("tags", _STRING_LIST),
15072 ("nics", ht.TListOf(ht.TDict)),
15073 ("vcpus", ht.TInt),
15074 ("hypervisor", ht.TString),
15076 constants.IALLOCATOR_MODE_RELOC:
15077 (_AddRelocateInstance,
15078 [("name", ht.TString), ("relocate_from", _STRING_LIST)],
15080 constants.IALLOCATOR_MODE_NODE_EVAC:
15081 (_AddNodeEvacuate, [
15082 ("instances", _STRING_LIST),
15083 ("evac_mode", ht.TElemOf(constants.IALLOCATOR_NEVAC_MODES)),
15085 constants.IALLOCATOR_MODE_CHG_GROUP:
15086 (_AddChangeGroup, [
15087 ("instances", _STRING_LIST),
15088 ("target_groups", _STRING_LIST),
15092 def Run(self, name, validate=True, call_fn=None):
15093 """Run an instance allocator and return the results.
15096 if call_fn is None:
15097 call_fn = self.rpc.call_iallocator_runner
15099 result = call_fn(self.cfg.GetMasterNode(), name, self.in_text)
15100 result.Raise("Failure while running the iallocator script")
15102 self.out_text = result.payload
15104 self._ValidateResult()
15106 def _ValidateResult(self):
15107 """Process the allocator results.
15109 This will process and if successful save the result in
15110 self.out_data and the other parameters.
15114 rdict = serializer.Load(self.out_text)
15115 except Exception, err:
15116 raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
15118 if not isinstance(rdict, dict):
15119 raise errors.OpExecError("Can't parse iallocator results: not a dict")
15121 # TODO: remove backwards compatiblity in later versions
15122 if "nodes" in rdict and "result" not in rdict:
15123 rdict["result"] = rdict["nodes"]
15126 for key in "success", "info", "result":
15127 if key not in rdict:
15128 raise errors.OpExecError("Can't parse iallocator results:"
15129 " missing key '%s'" % key)
15130 setattr(self, key, rdict[key])
15132 if not self._result_check(self.result):
15133 raise errors.OpExecError("Iallocator returned invalid result,"
15134 " expected %s, got %s" %
15135 (self._result_check, self.result),
15136 errors.ECODE_INVAL)
15138 if self.mode == constants.IALLOCATOR_MODE_RELOC:
15139 assert self.relocate_from is not None
15140 assert self.required_nodes == 1
15142 node2group = dict((name, ndata["group"])
15143 for (name, ndata) in self.in_data["nodes"].items())
15145 fn = compat.partial(self._NodesToGroups, node2group,
15146 self.in_data["nodegroups"])
15148 instance = self.cfg.GetInstanceInfo(self.name)
15149 request_groups = fn(self.relocate_from + [instance.primary_node])
15150 result_groups = fn(rdict["result"] + [instance.primary_node])
15152 if self.success and not set(result_groups).issubset(request_groups):
15153 raise errors.OpExecError("Groups of nodes returned by iallocator (%s)"
15154 " differ from original groups (%s)" %
15155 (utils.CommaJoin(result_groups),
15156 utils.CommaJoin(request_groups)))
15158 elif self.mode == constants.IALLOCATOR_MODE_NODE_EVAC:
15159 assert self.evac_mode in constants.IALLOCATOR_NEVAC_MODES
15161 self.out_data = rdict
15164 def _NodesToGroups(node2group, groups, nodes):
15165 """Returns a list of unique group names for a list of nodes.
15167 @type node2group: dict
15168 @param node2group: Map from node name to group UUID
15170 @param groups: Group information
15172 @param nodes: Node names
15179 group_uuid = node2group[node]
15181 # Ignore unknown node
15185 group = groups[group_uuid]
15187 # Can't find group, let's use UUID
15188 group_name = group_uuid
15190 group_name = group["name"]
15192 result.add(group_name)
15194 return sorted(result)
15197 class LUTestAllocator(NoHooksLU):
15198 """Run allocator tests.
15200 This LU runs the allocator tests
15203 def CheckPrereq(self):
15204 """Check prerequisites.
15206 This checks the opcode parameters depending on the director and mode test.
15209 if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
15210 for attr in ["memory", "disks", "disk_template",
15211 "os", "tags", "nics", "vcpus"]:
15212 if not hasattr(self.op, attr):
15213 raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
15214 attr, errors.ECODE_INVAL)
15215 iname = self.cfg.ExpandInstanceName(self.op.name)
15216 if iname is not None:
15217 raise errors.OpPrereqError("Instance '%s' already in the cluster" %
15218 iname, errors.ECODE_EXISTS)
15219 if not isinstance(self.op.nics, list):
15220 raise errors.OpPrereqError("Invalid parameter 'nics'",
15221 errors.ECODE_INVAL)
15222 if not isinstance(self.op.disks, list):
15223 raise errors.OpPrereqError("Invalid parameter 'disks'",
15224 errors.ECODE_INVAL)
15225 for row in self.op.disks:
15226 if (not isinstance(row, dict) or
15227 constants.IDISK_SIZE not in row or
15228 not isinstance(row[constants.IDISK_SIZE], int) or
15229 constants.IDISK_MODE not in row or
15230 row[constants.IDISK_MODE] not in constants.DISK_ACCESS_SET):
15231 raise errors.OpPrereqError("Invalid contents of the 'disks'"
15232 " parameter", errors.ECODE_INVAL)
15233 if self.op.hypervisor is None:
15234 self.op.hypervisor = self.cfg.GetHypervisorType()
15235 elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
15236 fname = _ExpandInstanceName(self.cfg, self.op.name)
15237 self.op.name = fname
15238 self.relocate_from = \
15239 list(self.cfg.GetInstanceInfo(fname).secondary_nodes)
15240 elif self.op.mode in (constants.IALLOCATOR_MODE_CHG_GROUP,
15241 constants.IALLOCATOR_MODE_NODE_EVAC):
15242 if not self.op.instances:
15243 raise errors.OpPrereqError("Missing instances", errors.ECODE_INVAL)
15244 self.op.instances = _GetWantedInstances(self, self.op.instances)
15246 raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
15247 self.op.mode, errors.ECODE_INVAL)
15249 if self.op.direction == constants.IALLOCATOR_DIR_OUT:
15250 if self.op.allocator is None:
15251 raise errors.OpPrereqError("Missing allocator name",
15252 errors.ECODE_INVAL)
15253 elif self.op.direction != constants.IALLOCATOR_DIR_IN:
15254 raise errors.OpPrereqError("Wrong allocator test '%s'" %
15255 self.op.direction, errors.ECODE_INVAL)
15257 def Exec(self, feedback_fn):
15258 """Run the allocator test.
15261 if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
15262 ial = IAllocator(self.cfg, self.rpc,
15265 memory=self.op.memory,
15266 disks=self.op.disks,
15267 disk_template=self.op.disk_template,
15271 vcpus=self.op.vcpus,
15272 hypervisor=self.op.hypervisor,
15274 elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
15275 ial = IAllocator(self.cfg, self.rpc,
15278 relocate_from=list(self.relocate_from),
15280 elif self.op.mode == constants.IALLOCATOR_MODE_CHG_GROUP:
15281 ial = IAllocator(self.cfg, self.rpc,
15283 instances=self.op.instances,
15284 target_groups=self.op.target_groups)
15285 elif self.op.mode == constants.IALLOCATOR_MODE_NODE_EVAC:
15286 ial = IAllocator(self.cfg, self.rpc,
15288 instances=self.op.instances,
15289 evac_mode=self.op.evac_mode)
15291 raise errors.ProgrammerError("Uncatched mode %s in"
15292 " LUTestAllocator.Exec", self.op.mode)
15294 if self.op.direction == constants.IALLOCATOR_DIR_IN:
15295 result = ial.in_text
15297 ial.Run(self.op.allocator, validate=False)
15298 result = ial.out_text
15302 #: Query type implementations
15304 constants.QR_CLUSTER: _ClusterQuery,
15305 constants.QR_INSTANCE: _InstanceQuery,
15306 constants.QR_NODE: _NodeQuery,
15307 constants.QR_GROUP: _GroupQuery,
15308 constants.QR_OS: _OsQuery,
15309 constants.QR_EXPORT: _ExportQuery,
15312 assert set(_QUERY_IMPL.keys()) == constants.QR_VIA_OP
15315 def _GetQueryImplementation(name):
15316 """Returns the implemtnation for a query type.
15318 @param name: Query type, must be one of L{constants.QR_VIA_OP}
15322 return _QUERY_IMPL[name]
15324 raise errors.OpPrereqError("Unknown query resource '%s'" % name,
15325 errors.ECODE_INVAL)