4 # Copyright (C) 2006, 2007, 2008, 2009, 2010 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
24 This module implements the data structures which define the cluster
25 operations - the so-called opcodes.
27 Every operation which modifies the cluster state is expressed via
32 # this are practically structures, so disable the message about too
34 # pylint: disable-msg=R0903
37 class BaseOpCode(object):
38 """A simple serializable object.
40 This object serves as a parent class for OpCode without any custom
46 def __init__(self, **kwargs):
47 """Constructor for BaseOpCode.
49 The constructor takes only keyword arguments and will set
50 attributes on this object based on the passed arguments. As such,
51 it means that you should not pass arguments which are not in the
52 __slots__ attribute for this class.
55 slots = self._all_slots()
58 raise TypeError("Object %s doesn't support the parameter '%s'" %
59 (self.__class__.__name__, key))
60 setattr(self, key, kwargs[key])
62 def __getstate__(self):
63 """Generic serializer.
65 This method just returns the contents of the instance as a
69 @return: the instance attributes and their values
73 for name in self._all_slots():
74 if hasattr(self, name):
75 state[name] = getattr(self, name)
78 def __setstate__(self, state):
79 """Generic unserializer.
81 This method just restores from the serialized state the attributes
82 of the current instance.
84 @param state: the serialized opcode data
88 if not isinstance(state, dict):
89 raise ValueError("Invalid data to __setstate__: expected dict, got %s" %
92 for name in self._all_slots():
93 if name not in state and hasattr(self, name):
97 setattr(self, name, state[name])
101 """Compute the list of all declared slots for a class.
105 for parent in cls.__mro__:
106 slots.extend(getattr(parent, "__slots__", []))
110 class OpCode(BaseOpCode):
113 This is the root of the actual OpCode hierarchy. All clases derived
114 from this class should override OP_ID.
116 @cvar OP_ID: The ID of this opcode. This should be unique amongst all
117 children of this class.
118 @ivar dry_run: Whether the LU should be run in dry-run mode, i.e. just
122 OP_ID = "OP_ABSTRACT"
123 __slots__ = ["dry_run", "debug_level"]
125 def __getstate__(self):
126 """Specialized getstate for opcodes.
128 This method adds to the state dictionary the OP_ID of the class,
129 so that on unload we can identify the correct class for
130 instantiating the opcode.
133 @return: the state as a dictionary
136 data = BaseOpCode.__getstate__(self)
137 data["OP_ID"] = self.OP_ID
141 def LoadOpCode(cls, data):
142 """Generic load opcode method.
144 The method identifies the correct opcode class from the dict-form
145 by looking for a OP_ID key, if this is not found, or its value is
146 not available in this module as a child of this class, we fail.
149 @param data: the serialized opcode
152 if not isinstance(data, dict):
153 raise ValueError("Invalid data to LoadOpCode (%s)" % type(data))
154 if "OP_ID" not in data:
155 raise ValueError("Invalid data to LoadOpcode, missing OP_ID")
156 op_id = data["OP_ID"]
158 if op_id in OP_MAPPING:
159 op_class = OP_MAPPING[op_id]
161 raise ValueError("Invalid data to LoadOpCode: OP_ID %s unsupported" %
164 new_data = data.copy()
165 del new_data["OP_ID"]
166 op.__setstate__(new_data)
170 """Generates a summary description of this opcode.
173 # all OP_ID start with OP_, we remove that
175 field_name = getattr(self, "OP_DSC_FIELD", None)
177 field_value = getattr(self, field_name, None)
178 txt = "%s(%s)" % (txt, field_value)
184 class OpPostInitCluster(OpCode):
185 """Post cluster initialization.
187 This opcode does not touch the cluster at all. Its purpose is to run hooks
188 after the cluster has been initialized.
191 OP_ID = "OP_CLUSTER_POST_INIT"
195 class OpDestroyCluster(OpCode):
196 """Destroy the cluster.
198 This opcode has no other parameters. All the state is irreversibly
199 lost after the execution of this opcode.
202 OP_ID = "OP_CLUSTER_DESTROY"
206 class OpQueryClusterInfo(OpCode):
207 """Query cluster information."""
208 OP_ID = "OP_CLUSTER_QUERY"
212 class OpVerifyCluster(OpCode):
213 """Verify the cluster state.
215 @type skip_checks: C{list}
216 @ivar skip_checks: steps to be skipped from the verify process; this
217 needs to be a subset of
218 L{constants.VERIFY_OPTIONAL_CHECKS}; currently
219 only L{constants.VERIFY_NPLUSONE_MEM} can be passed
222 OP_ID = "OP_CLUSTER_VERIFY"
223 __slots__ = ["skip_checks", "verbose", "error_codes",
224 "debug_simulate_errors"]
227 class OpVerifyDisks(OpCode):
228 """Verify the cluster disks.
232 Result: a tuple of four elements:
233 - list of node names with bad data returned (unreachable, etc.)
234 - dict of node names with broken volume groups (values: error msg)
235 - list of instances with degraded disks (that should be activated)
236 - dict of instances with missing logical volumes (values: (node, vol)
237 pairs with details about the missing volumes)
239 In normal operation, all lists should be empty. A non-empty instance
240 list (3rd element of the result) is still ok (errors were fixed) but
241 non-empty node list means some node is down, and probably there are
242 unfixable drbd errors.
244 Note that only instances that are drbd-based are taken into
245 consideration. This might need to be revisited in the future.
248 OP_ID = "OP_CLUSTER_VERIFY_DISKS"
252 class OpRepairDiskSizes(OpCode):
253 """Verify the disk sizes of the instances and fixes configuration
256 Parameters: optional instances list, in case we want to restrict the
257 checks to only a subset of the instances.
259 Result: a list of tuples, (instance, disk, new-size) for changed
262 In normal operation, the list should be empty.
264 @type instances: list
265 @ivar instances: the list of instances to check, or empty for all instances
268 OP_ID = "OP_CLUSTER_REPAIR_DISK_SIZES"
269 __slots__ = ["instances"]
272 class OpQueryConfigValues(OpCode):
273 """Query cluster configuration values."""
274 OP_ID = "OP_CLUSTER_CONFIG_QUERY"
275 __slots__ = ["output_fields"]
278 class OpRenameCluster(OpCode):
279 """Rename the cluster.
282 @ivar name: The new name of the cluster. The name and/or the master IP
283 address will be changed to match the new name and its IP
287 OP_ID = "OP_CLUSTER_RENAME"
288 OP_DSC_FIELD = "name"
292 class OpSetClusterParams(OpCode):
293 """Change the parameters of the cluster.
295 @type vg_name: C{str} or C{None}
296 @ivar vg_name: The new volume group name or None to disable LVM usage.
299 OP_ID = "OP_CLUSTER_SET_PARAMS"
303 "enabled_hypervisors",
309 "candidate_pool_size",
310 "maintain_node_health",
314 "default_iallocator",
319 class OpRedistributeConfig(OpCode):
320 """Force a full push of the cluster configuration.
323 OP_ID = "OP_CLUSTER_REDIST_CONF"
328 class OpRemoveNode(OpCode):
331 @type node_name: C{str}
332 @ivar node_name: The name of the node to remove. If the node still has
333 instances on it, the operation will fail.
336 OP_ID = "OP_NODE_REMOVE"
337 OP_DSC_FIELD = "node_name"
338 __slots__ = ["node_name"]
341 class OpAddNode(OpCode):
342 """Add a node to the cluster.
344 @type node_name: C{str}
345 @ivar node_name: The name of the node to add. This can be a short name,
346 but it will be expanded to the FQDN.
347 @type primary_ip: IP address
348 @ivar primary_ip: The primary IP of the node. This will be ignored when the
349 opcode is submitted, but will be filled during the node
350 add (so it will be visible in the job query).
351 @type secondary_ip: IP address
352 @ivar secondary_ip: The secondary IP of the node. This needs to be passed
353 if the cluster has been initialized in 'dual-network'
354 mode, otherwise it must not be given.
356 @ivar readd: Whether to re-add an existing node to the cluster. If
357 this is not passed, then the operation will abort if the node
358 name is already in the cluster; use this parameter to 'repair'
359 a node that had its configuration broken, or was reinstalled
360 without removal from the cluster.
363 OP_ID = "OP_NODE_ADD"
364 OP_DSC_FIELD = "node_name"
365 __slots__ = ["node_name", "primary_ip", "secondary_ip", "readd"]
368 class OpQueryNodes(OpCode):
369 """Compute the list of nodes."""
370 OP_ID = "OP_NODE_QUERY"
371 __slots__ = ["output_fields", "names", "use_locking"]
374 class OpQueryNodeVolumes(OpCode):
375 """Get list of volumes on node."""
376 OP_ID = "OP_NODE_QUERYVOLS"
377 __slots__ = ["nodes", "output_fields"]
380 class OpQueryNodeStorage(OpCode):
381 """Get information on storage for node(s)."""
382 OP_ID = "OP_NODE_QUERY_STORAGE"
391 class OpModifyNodeStorage(OpCode):
392 """Modifies the properies of a storage unit"""
393 OP_ID = "OP_NODE_MODIFY_STORAGE"
402 class OpRepairNodeStorage(OpCode):
403 """Repairs the volume group on a node."""
404 OP_ID = "OP_REPAIR_NODE_STORAGE"
405 OP_DSC_FIELD = "node_name"
410 "ignore_consistency",
414 class OpSetNodeParams(OpCode):
415 """Change the parameters of a node."""
416 OP_ID = "OP_NODE_SET_PARAMS"
417 OP_DSC_FIELD = "node_name"
428 class OpPowercycleNode(OpCode):
429 """Tries to powercycle a node."""
430 OP_ID = "OP_NODE_POWERCYCLE"
431 OP_DSC_FIELD = "node_name"
438 class OpMigrateNode(OpCode):
439 """Migrate all instances from a node."""
440 OP_ID = "OP_NODE_MIGRATE"
441 OP_DSC_FIELD = "node_name"
448 class OpNodeEvacuationStrategy(OpCode):
449 """Compute the evacuation strategy for a list of nodes."""
450 OP_ID = "OP_NODE_EVAC_STRATEGY"
451 OP_DSC_FIELD = "nodes"
452 __slots__ = ["nodes", "iallocator", "remote_node"]
457 class OpCreateInstance(OpCode):
458 """Create an instance.
460 @ivar instance_name: Instance name
461 @ivar mode: Instance creation mode (one of L{constants.INSTANCE_CREATE_MODES})
462 @ivar source_handshake: Signed handshake from source (remote import only)
463 @ivar source_x509_ca: Source X509 CA in PEM format (remote import only)
464 @ivar source_instance_name: Previous name of instance (remote import only)
467 OP_ID = "OP_INSTANCE_CREATE"
468 OP_DSC_FIELD = "instance_name"
471 "os_type", "force_variant", "no_install",
472 "pnode", "disk_template", "snode", "mode",
474 "src_node", "src_path", "start", "identify_defaults",
475 "wait_for_sync", "ip_check", "name_check",
476 "file_storage_dir", "file_driver",
478 "hypervisor", "hvparams", "beparams", "osparams",
481 "source_instance_name",
486 class OpReinstallInstance(OpCode):
487 """Reinstall an instance's OS."""
488 OP_ID = "OP_INSTANCE_REINSTALL"
489 OP_DSC_FIELD = "instance_name"
490 __slots__ = ["instance_name", "os_type", "force_variant"]
493 class OpRemoveInstance(OpCode):
494 """Remove an instance."""
495 OP_ID = "OP_INSTANCE_REMOVE"
496 OP_DSC_FIELD = "instance_name"
504 class OpRenameInstance(OpCode):
505 """Rename an instance."""
506 OP_ID = "OP_INSTANCE_RENAME"
508 "instance_name", "ip_check", "new_name", "name_check",
512 class OpStartupInstance(OpCode):
513 """Startup an instance."""
514 OP_ID = "OP_INSTANCE_STARTUP"
515 OP_DSC_FIELD = "instance_name"
517 "instance_name", "force", "hvparams", "beparams",
521 class OpShutdownInstance(OpCode):
522 """Shutdown an instance."""
523 OP_ID = "OP_INSTANCE_SHUTDOWN"
524 OP_DSC_FIELD = "instance_name"
525 __slots__ = ["instance_name", "timeout"]
528 class OpRebootInstance(OpCode):
529 """Reboot an instance."""
530 OP_ID = "OP_INSTANCE_REBOOT"
531 OP_DSC_FIELD = "instance_name"
533 "instance_name", "reboot_type", "ignore_secondaries", "shutdown_timeout",
537 class OpReplaceDisks(OpCode):
538 """Replace the disks of an instance."""
539 OP_ID = "OP_INSTANCE_REPLACE_DISKS"
540 OP_DSC_FIELD = "instance_name"
542 "instance_name", "remote_node", "mode", "disks", "iallocator",
547 class OpFailoverInstance(OpCode):
548 """Failover an instance."""
549 OP_ID = "OP_INSTANCE_FAILOVER"
550 OP_DSC_FIELD = "instance_name"
552 "instance_name", "ignore_consistency", "shutdown_timeout",
556 class OpMigrateInstance(OpCode):
557 """Migrate an instance.
559 This migrates (without shutting down an instance) to its secondary
562 @ivar instance_name: the name of the instance
563 @ivar mode: the migration mode (live, non-live or None for auto)
566 OP_ID = "OP_INSTANCE_MIGRATE"
567 OP_DSC_FIELD = "instance_name"
568 __slots__ = ["instance_name", "mode", "cleanup"]
571 class OpMoveInstance(OpCode):
574 This move (with shutting down an instance and data copying) to an
577 @ivar instance_name: the name of the instance
578 @ivar target_node: the destination node
581 OP_ID = "OP_INSTANCE_MOVE"
582 OP_DSC_FIELD = "instance_name"
584 "instance_name", "target_node", "shutdown_timeout",
588 class OpConnectConsole(OpCode):
589 """Connect to an instance's console."""
590 OP_ID = "OP_INSTANCE_CONSOLE"
591 OP_DSC_FIELD = "instance_name"
592 __slots__ = ["instance_name"]
595 class OpActivateInstanceDisks(OpCode):
596 """Activate an instance's disks."""
597 OP_ID = "OP_INSTANCE_ACTIVATE_DISKS"
598 OP_DSC_FIELD = "instance_name"
599 __slots__ = ["instance_name", "ignore_size"]
602 class OpDeactivateInstanceDisks(OpCode):
603 """Deactivate an instance's disks."""
604 OP_ID = "OP_INSTANCE_DEACTIVATE_DISKS"
605 OP_DSC_FIELD = "instance_name"
606 __slots__ = ["instance_name"]
609 class OpRecreateInstanceDisks(OpCode):
610 """Deactivate an instance's disks."""
611 OP_ID = "OP_INSTANCE_RECREATE_DISKS"
612 OP_DSC_FIELD = "instance_name"
613 __slots__ = ["instance_name", "disks"]
616 class OpQueryInstances(OpCode):
617 """Compute the list of instances."""
618 OP_ID = "OP_INSTANCE_QUERY"
619 __slots__ = ["output_fields", "names", "use_locking"]
622 class OpQueryInstanceData(OpCode):
623 """Compute the run-time status of instances."""
624 OP_ID = "OP_INSTANCE_QUERY_DATA"
625 __slots__ = ["instances", "static"]
628 class OpSetInstanceParams(OpCode):
629 """Change the parameters of an instance."""
630 OP_ID = "OP_INSTANCE_SET_PARAMS"
631 OP_DSC_FIELD = "instance_name"
634 "hvparams", "beparams", "osparams", "force",
635 "nics", "disks", "disk_template",
636 "remote_node", "os_name", "force_variant",
640 class OpGrowDisk(OpCode):
641 """Grow a disk of an instance."""
642 OP_ID = "OP_INSTANCE_GROW_DISK"
643 OP_DSC_FIELD = "instance_name"
645 "instance_name", "disk", "amount", "wait_for_sync",
650 class OpDiagnoseOS(OpCode):
651 """Compute the list of guest operating systems."""
652 OP_ID = "OP_OS_DIAGNOSE"
653 __slots__ = ["output_fields", "names"]
657 class OpQueryExports(OpCode):
658 """Compute the list of exported images."""
659 OP_ID = "OP_BACKUP_QUERY"
660 __slots__ = ["nodes", "use_locking"]
663 class OpPrepareExport(OpCode):
664 """Prepares an instance export.
666 @ivar instance_name: Instance name
667 @ivar mode: Export mode (one of L{constants.EXPORT_MODES})
670 OP_ID = "OP_BACKUP_PREPARE"
671 OP_DSC_FIELD = "instance_name"
673 "instance_name", "mode",
677 class OpExportInstance(OpCode):
678 """Export an instance.
680 For local exports, the export destination is the node name. For remote
681 exports, the export destination is a list of tuples, each consisting of
682 hostname/IP address, port, HMAC and HMAC salt. The HMAC is calculated using
683 the cluster domain secret over the value "${index}:${hostname}:${port}". The
684 destination X509 CA must be a signed certificate.
686 @ivar mode: Export mode (one of L{constants.EXPORT_MODES})
687 @ivar target_node: Export destination
688 @ivar x509_key_name: X509 key to use (remote export only)
689 @ivar destination_x509_ca: Destination X509 CA in PEM format (remote export
693 OP_ID = "OP_BACKUP_EXPORT"
694 OP_DSC_FIELD = "instance_name"
696 # TODO: Rename target_node as it changes meaning for different export modes
697 # (e.g. "destination")
698 "instance_name", "target_node", "shutdown", "shutdown_timeout",
700 "ignore_remove_failures",
703 "destination_x509_ca",
707 class OpRemoveExport(OpCode):
708 """Remove an instance's export."""
709 OP_ID = "OP_BACKUP_REMOVE"
710 OP_DSC_FIELD = "instance_name"
711 __slots__ = ["instance_name"]
715 class OpGetTags(OpCode):
716 """Returns the tags of the given object."""
717 OP_ID = "OP_TAGS_GET"
718 OP_DSC_FIELD = "name"
719 __slots__ = ["kind", "name"]
722 class OpSearchTags(OpCode):
723 """Searches the tags in the cluster for a given pattern."""
724 OP_ID = "OP_TAGS_SEARCH"
725 OP_DSC_FIELD = "pattern"
726 __slots__ = ["pattern"]
729 class OpAddTags(OpCode):
730 """Add a list of tags on a given object."""
731 OP_ID = "OP_TAGS_SET"
732 __slots__ = ["kind", "name", "tags"]
735 class OpDelTags(OpCode):
736 """Remove a list of tags from a given object."""
737 OP_ID = "OP_TAGS_DEL"
738 __slots__ = ["kind", "name", "tags"]
742 class OpTestDelay(OpCode):
743 """Sleeps for a configured amount of time.
745 This is used just for debugging and testing.
748 - duration: the time to sleep
749 - on_master: if true, sleep on the master
750 - on_nodes: list of nodes in which to sleep
752 If the on_master parameter is true, it will execute a sleep on the
753 master (before any node sleep).
755 If the on_nodes list is not empty, it will sleep on those nodes
756 (after the sleep on the master, if that is enabled).
758 As an additional feature, the case of duration < 0 will be reported
759 as an execution error, so this opcode can be used as a failure
760 generator. The case of duration == 0 will not be treated specially.
763 OP_ID = "OP_TEST_DELAY"
764 OP_DSC_FIELD = "duration"
765 __slots__ = ["duration", "on_master", "on_nodes", "repeat"]
768 class OpTestAllocator(OpCode):
769 """Allocator framework testing.
771 This opcode has two modes:
772 - gather and return allocator input for a given mode (allocate new
773 or replace secondary) and a given instance definition (direction
775 - run a selected allocator for a given operation (as above) and
776 return the allocator output (direction 'out')
779 OP_ID = "OP_TEST_ALLOCATOR"
780 OP_DSC_FIELD = "allocator"
782 "direction", "mode", "allocator", "name",
783 "mem_size", "disks", "disk_template",
784 "os", "tags", "nics", "vcpus", "hypervisor",
789 class OpTestJobqueue(OpCode):
790 """Utility opcode to test some aspects of the job queue.
793 OP_ID = "OP_TEST_JQUEUE"
802 OP_MAPPING = dict([(v.OP_ID, v) for v in globals().values()
803 if (isinstance(v, type) and issubclass(v, OpCode) and
804 hasattr(v, "OP_ID"))])