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"
449 class OpNodeEvacuationStrategy(OpCode):
450 """Compute the evacuation strategy for a list of nodes."""
451 OP_ID = "OP_NODE_EVAC_STRATEGY"
452 OP_DSC_FIELD = "nodes"
453 __slots__ = ["nodes", "iallocator", "remote_node"]
458 class OpCreateInstance(OpCode):
459 """Create an instance.
461 @ivar instance_name: Instance name
462 @ivar mode: Instance creation mode (one of L{constants.INSTANCE_CREATE_MODES})
463 @ivar source_handshake: Signed handshake from source (remote import only)
464 @ivar source_x509_ca: Source X509 CA in PEM format (remote import only)
465 @ivar source_instance_name: Previous name of instance (remote import only)
468 OP_ID = "OP_INSTANCE_CREATE"
469 OP_DSC_FIELD = "instance_name"
472 "os_type", "force_variant", "no_install",
473 "pnode", "disk_template", "snode", "mode",
475 "src_node", "src_path", "start", "identify_defaults",
476 "wait_for_sync", "ip_check", "name_check",
477 "file_storage_dir", "file_driver",
479 "hypervisor", "hvparams", "beparams", "osparams",
482 "source_instance_name",
487 class OpReinstallInstance(OpCode):
488 """Reinstall an instance's OS."""
489 OP_ID = "OP_INSTANCE_REINSTALL"
490 OP_DSC_FIELD = "instance_name"
491 __slots__ = ["instance_name", "os_type", "force_variant"]
494 class OpRemoveInstance(OpCode):
495 """Remove an instance."""
496 OP_ID = "OP_INSTANCE_REMOVE"
497 OP_DSC_FIELD = "instance_name"
505 class OpRenameInstance(OpCode):
506 """Rename an instance."""
507 OP_ID = "OP_INSTANCE_RENAME"
509 "instance_name", "ip_check", "new_name", "name_check",
513 class OpStartupInstance(OpCode):
514 """Startup an instance."""
515 OP_ID = "OP_INSTANCE_STARTUP"
516 OP_DSC_FIELD = "instance_name"
518 "instance_name", "force", "hvparams", "beparams",
522 class OpShutdownInstance(OpCode):
523 """Shutdown an instance."""
524 OP_ID = "OP_INSTANCE_SHUTDOWN"
525 OP_DSC_FIELD = "instance_name"
526 __slots__ = ["instance_name", "timeout"]
529 class OpRebootInstance(OpCode):
530 """Reboot an instance."""
531 OP_ID = "OP_INSTANCE_REBOOT"
532 OP_DSC_FIELD = "instance_name"
534 "instance_name", "reboot_type", "ignore_secondaries", "shutdown_timeout",
538 class OpReplaceDisks(OpCode):
539 """Replace the disks of an instance."""
540 OP_ID = "OP_INSTANCE_REPLACE_DISKS"
541 OP_DSC_FIELD = "instance_name"
543 "instance_name", "remote_node", "mode", "disks", "iallocator",
548 class OpFailoverInstance(OpCode):
549 """Failover an instance."""
550 OP_ID = "OP_INSTANCE_FAILOVER"
551 OP_DSC_FIELD = "instance_name"
553 "instance_name", "ignore_consistency", "shutdown_timeout",
557 class OpMigrateInstance(OpCode):
558 """Migrate an instance.
560 This migrates (without shutting down an instance) to its secondary
563 @ivar instance_name: the name of the instance
564 @ivar mode: the migration mode (live, non-live or None for auto)
567 OP_ID = "OP_INSTANCE_MIGRATE"
568 OP_DSC_FIELD = "instance_name"
569 __slots__ = ["instance_name", "mode", "cleanup", "live"]
572 class OpMoveInstance(OpCode):
575 This move (with shutting down an instance and data copying) to an
578 @ivar instance_name: the name of the instance
579 @ivar target_node: the destination node
582 OP_ID = "OP_INSTANCE_MOVE"
583 OP_DSC_FIELD = "instance_name"
585 "instance_name", "target_node", "shutdown_timeout",
589 class OpConnectConsole(OpCode):
590 """Connect to an instance's console."""
591 OP_ID = "OP_INSTANCE_CONSOLE"
592 OP_DSC_FIELD = "instance_name"
593 __slots__ = ["instance_name"]
596 class OpActivateInstanceDisks(OpCode):
597 """Activate an instance's disks."""
598 OP_ID = "OP_INSTANCE_ACTIVATE_DISKS"
599 OP_DSC_FIELD = "instance_name"
600 __slots__ = ["instance_name", "ignore_size"]
603 class OpDeactivateInstanceDisks(OpCode):
604 """Deactivate an instance's disks."""
605 OP_ID = "OP_INSTANCE_DEACTIVATE_DISKS"
606 OP_DSC_FIELD = "instance_name"
607 __slots__ = ["instance_name"]
610 class OpRecreateInstanceDisks(OpCode):
611 """Deactivate an instance's disks."""
612 OP_ID = "OP_INSTANCE_RECREATE_DISKS"
613 OP_DSC_FIELD = "instance_name"
614 __slots__ = ["instance_name", "disks"]
617 class OpQueryInstances(OpCode):
618 """Compute the list of instances."""
619 OP_ID = "OP_INSTANCE_QUERY"
620 __slots__ = ["output_fields", "names", "use_locking"]
623 class OpQueryInstanceData(OpCode):
624 """Compute the run-time status of instances."""
625 OP_ID = "OP_INSTANCE_QUERY_DATA"
626 __slots__ = ["instances", "static"]
629 class OpSetInstanceParams(OpCode):
630 """Change the parameters of an instance."""
631 OP_ID = "OP_INSTANCE_SET_PARAMS"
632 OP_DSC_FIELD = "instance_name"
635 "hvparams", "beparams", "osparams", "force",
636 "nics", "disks", "disk_template",
637 "remote_node", "os_name", "force_variant",
641 class OpGrowDisk(OpCode):
642 """Grow a disk of an instance."""
643 OP_ID = "OP_INSTANCE_GROW_DISK"
644 OP_DSC_FIELD = "instance_name"
646 "instance_name", "disk", "amount", "wait_for_sync",
651 class OpDiagnoseOS(OpCode):
652 """Compute the list of guest operating systems."""
653 OP_ID = "OP_OS_DIAGNOSE"
654 __slots__ = ["output_fields", "names"]
658 class OpQueryExports(OpCode):
659 """Compute the list of exported images."""
660 OP_ID = "OP_BACKUP_QUERY"
661 __slots__ = ["nodes", "use_locking"]
664 class OpPrepareExport(OpCode):
665 """Prepares an instance export.
667 @ivar instance_name: Instance name
668 @ivar mode: Export mode (one of L{constants.EXPORT_MODES})
671 OP_ID = "OP_BACKUP_PREPARE"
672 OP_DSC_FIELD = "instance_name"
674 "instance_name", "mode",
678 class OpExportInstance(OpCode):
679 """Export an instance.
681 For local exports, the export destination is the node name. For remote
682 exports, the export destination is a list of tuples, each consisting of
683 hostname/IP address, port, HMAC and HMAC salt. The HMAC is calculated using
684 the cluster domain secret over the value "${index}:${hostname}:${port}". The
685 destination X509 CA must be a signed certificate.
687 @ivar mode: Export mode (one of L{constants.EXPORT_MODES})
688 @ivar target_node: Export destination
689 @ivar x509_key_name: X509 key to use (remote export only)
690 @ivar destination_x509_ca: Destination X509 CA in PEM format (remote export
694 OP_ID = "OP_BACKUP_EXPORT"
695 OP_DSC_FIELD = "instance_name"
697 # TODO: Rename target_node as it changes meaning for different export modes
698 # (e.g. "destination")
699 "instance_name", "target_node", "shutdown", "shutdown_timeout",
701 "ignore_remove_failures",
704 "destination_x509_ca",
708 class OpRemoveExport(OpCode):
709 """Remove an instance's export."""
710 OP_ID = "OP_BACKUP_REMOVE"
711 OP_DSC_FIELD = "instance_name"
712 __slots__ = ["instance_name"]
716 class OpGetTags(OpCode):
717 """Returns the tags of the given object."""
718 OP_ID = "OP_TAGS_GET"
719 OP_DSC_FIELD = "name"
720 __slots__ = ["kind", "name"]
723 class OpSearchTags(OpCode):
724 """Searches the tags in the cluster for a given pattern."""
725 OP_ID = "OP_TAGS_SEARCH"
726 OP_DSC_FIELD = "pattern"
727 __slots__ = ["pattern"]
730 class OpAddTags(OpCode):
731 """Add a list of tags on a given object."""
732 OP_ID = "OP_TAGS_SET"
733 __slots__ = ["kind", "name", "tags"]
736 class OpDelTags(OpCode):
737 """Remove a list of tags from a given object."""
738 OP_ID = "OP_TAGS_DEL"
739 __slots__ = ["kind", "name", "tags"]
743 class OpTestDelay(OpCode):
744 """Sleeps for a configured amount of time.
746 This is used just for debugging and testing.
749 - duration: the time to sleep
750 - on_master: if true, sleep on the master
751 - on_nodes: list of nodes in which to sleep
753 If the on_master parameter is true, it will execute a sleep on the
754 master (before any node sleep).
756 If the on_nodes list is not empty, it will sleep on those nodes
757 (after the sleep on the master, if that is enabled).
759 As an additional feature, the case of duration < 0 will be reported
760 as an execution error, so this opcode can be used as a failure
761 generator. The case of duration == 0 will not be treated specially.
764 OP_ID = "OP_TEST_DELAY"
765 OP_DSC_FIELD = "duration"
766 __slots__ = ["duration", "on_master", "on_nodes", "repeat"]
769 class OpTestAllocator(OpCode):
770 """Allocator framework testing.
772 This opcode has two modes:
773 - gather and return allocator input for a given mode (allocate new
774 or replace secondary) and a given instance definition (direction
776 - run a selected allocator for a given operation (as above) and
777 return the allocator output (direction 'out')
780 OP_ID = "OP_TEST_ALLOCATOR"
781 OP_DSC_FIELD = "allocator"
783 "direction", "mode", "allocator", "name",
784 "mem_size", "disks", "disk_template",
785 "os", "tags", "nics", "vcpus", "hypervisor",
790 class OpTestJobqueue(OpCode):
791 """Utility opcode to test some aspects of the job queue.
794 OP_ID = "OP_TEST_JQUEUE"
803 OP_MAPPING = dict([(v.OP_ID, v) for v in globals().values()
804 if (isinstance(v, type) and issubclass(v, OpCode) and
805 hasattr(v, "OP_ID"))])