4 # Copyright (C) 2006, 2007 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():
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"
302 "enabled_hypervisors",
306 "candidate_pool_size",
310 class OpRedistributeConfig(OpCode):
311 """Force a full push of the cluster configuration.
314 OP_ID = "OP_CLUSTER_REDIST_CONF"
319 class OpRemoveNode(OpCode):
322 @type node_name: C{str}
323 @ivar node_name: The name of the node to remove. If the node still has
324 instances on it, the operation will fail.
327 OP_ID = "OP_NODE_REMOVE"
328 OP_DSC_FIELD = "node_name"
329 __slots__ = ["node_name"]
332 class OpAddNode(OpCode):
333 """Add a node to the cluster.
335 @type node_name: C{str}
336 @ivar node_name: The name of the node to add. This can be a short name,
337 but it will be expanded to the FQDN.
338 @type primary_ip: IP address
339 @ivar primary_ip: The primary IP of the node. This will be ignored when the
340 opcode is submitted, but will be filled during the node
341 add (so it will be visible in the job query).
342 @type secondary_ip: IP address
343 @ivar secondary_ip: The secondary IP of the node. This needs to be passed
344 if the cluster has been initialized in 'dual-network'
345 mode, otherwise it must not be given.
347 @ivar readd: Whether to re-add an existing node to the cluster. If
348 this is not passed, then the operation will abort if the node
349 name is already in the cluster; use this parameter to 'repair'
350 a node that had its configuration broken, or was reinstalled
351 without removal from the cluster.
354 OP_ID = "OP_NODE_ADD"
355 OP_DSC_FIELD = "node_name"
356 __slots__ = ["node_name", "primary_ip", "secondary_ip", "readd"]
359 class OpQueryNodes(OpCode):
360 """Compute the list of nodes."""
361 OP_ID = "OP_NODE_QUERY"
362 __slots__ = ["output_fields", "names", "use_locking"]
365 class OpQueryNodeVolumes(OpCode):
366 """Get list of volumes on node."""
367 OP_ID = "OP_NODE_QUERYVOLS"
368 __slots__ = ["nodes", "output_fields"]
371 class OpQueryNodeStorage(OpCode):
372 """Get information on storage for node(s)."""
373 OP_ID = "OP_NODE_QUERY_STORAGE"
382 class OpModifyNodeStorage(OpCode):
383 """Modifies the properies of a storage unit"""
384 OP_ID = "OP_NODE_MODIFY_STORAGE"
393 class OpRepairNodeStorage(OpCode):
394 """Repairs the volume group on a node."""
395 OP_ID = "OP_REPAIR_NODE_STORAGE"
396 OP_DSC_FIELD = "node_name"
401 "ignore_consistency",
405 class OpSetNodeParams(OpCode):
406 """Change the parameters of a node."""
407 OP_ID = "OP_NODE_SET_PARAMS"
408 OP_DSC_FIELD = "node_name"
418 class OpPowercycleNode(OpCode):
419 """Tries to powercycle a node."""
420 OP_ID = "OP_NODE_POWERCYCLE"
421 OP_DSC_FIELD = "node_name"
428 class OpEvacuateNode(OpCode):
429 """Relocate secondary instances from a node."""
430 OP_ID = "OP_NODE_EVACUATE"
431 OP_DSC_FIELD = "node_name"
433 "node_name", "remote_node", "iallocator", "early_release",
437 class OpMigrateNode(OpCode):
438 """Migrate all instances from a node."""
439 OP_ID = "OP_NODE_MIGRATE"
440 OP_DSC_FIELD = "node_name"
447 class OpNodeEvacuationStrategy(OpCode):
448 """Compute the evacuation strategy for a list of nodes."""
449 OP_ID = "OP_NODE_EVAC_STRATEGY"
450 OP_DSC_FIELD = "nodes"
451 __slots__ = ["nodes", "iallocator", "remote_node"]
456 class OpCreateInstance(OpCode):
457 """Create an instance."""
458 OP_ID = "OP_INSTANCE_CREATE"
459 OP_DSC_FIELD = "instance_name"
461 "instance_name", "os_type", "force_variant",
462 "pnode", "disk_template", "snode", "mode",
464 "src_node", "src_path", "start",
465 "wait_for_sync", "ip_check", "name_check",
466 "file_storage_dir", "file_driver",
468 "hypervisor", "hvparams", "beparams",
473 class OpReinstallInstance(OpCode):
474 """Reinstall an instance's OS."""
475 OP_ID = "OP_INSTANCE_REINSTALL"
476 OP_DSC_FIELD = "instance_name"
477 __slots__ = ["instance_name", "os_type", "force_variant"]
480 class OpRemoveInstance(OpCode):
481 """Remove an instance."""
482 OP_ID = "OP_INSTANCE_REMOVE"
483 OP_DSC_FIELD = "instance_name"
491 class OpRenameInstance(OpCode):
492 """Rename an instance."""
493 OP_ID = "OP_INSTANCE_RENAME"
495 "instance_name", "ignore_ip", "new_name",
499 class OpStartupInstance(OpCode):
500 """Startup an instance."""
501 OP_ID = "OP_INSTANCE_STARTUP"
502 OP_DSC_FIELD = "instance_name"
504 "instance_name", "force", "hvparams", "beparams",
508 class OpShutdownInstance(OpCode):
509 """Shutdown an instance."""
510 OP_ID = "OP_INSTANCE_SHUTDOWN"
511 OP_DSC_FIELD = "instance_name"
512 __slots__ = ["instance_name", "timeout"]
515 class OpRebootInstance(OpCode):
516 """Reboot an instance."""
517 OP_ID = "OP_INSTANCE_REBOOT"
518 OP_DSC_FIELD = "instance_name"
520 "instance_name", "reboot_type", "ignore_secondaries", "shutdown_timeout",
524 class OpReplaceDisks(OpCode):
525 """Replace the disks of an instance."""
526 OP_ID = "OP_INSTANCE_REPLACE_DISKS"
527 OP_DSC_FIELD = "instance_name"
529 "instance_name", "remote_node", "mode", "disks", "iallocator",
534 class OpFailoverInstance(OpCode):
535 """Failover an instance."""
536 OP_ID = "OP_INSTANCE_FAILOVER"
537 OP_DSC_FIELD = "instance_name"
539 "instance_name", "ignore_consistency", "shutdown_timeout",
543 class OpMigrateInstance(OpCode):
544 """Migrate an instance.
546 This migrates (without shutting down an instance) to its secondary
549 @ivar instance_name: the name of the instance
552 OP_ID = "OP_INSTANCE_MIGRATE"
553 OP_DSC_FIELD = "instance_name"
554 __slots__ = ["instance_name", "live", "cleanup"]
557 class OpMoveInstance(OpCode):
560 This move (with shutting down an instance and data copying) to an
563 @ivar instance_name: the name of the instance
564 @ivar target_node: the destination node
567 OP_ID = "OP_INSTANCE_MOVE"
568 OP_DSC_FIELD = "instance_name"
570 "instance_name", "target_node", "shutdown_timeout",
574 class OpConnectConsole(OpCode):
575 """Connect to an instance's console."""
576 OP_ID = "OP_INSTANCE_CONSOLE"
577 OP_DSC_FIELD = "instance_name"
578 __slots__ = ["instance_name"]
581 class OpActivateInstanceDisks(OpCode):
582 """Activate an instance's disks."""
583 OP_ID = "OP_INSTANCE_ACTIVATE_DISKS"
584 OP_DSC_FIELD = "instance_name"
585 __slots__ = ["instance_name", "ignore_size"]
588 class OpDeactivateInstanceDisks(OpCode):
589 """Deactivate an instance's disks."""
590 OP_ID = "OP_INSTANCE_DEACTIVATE_DISKS"
591 OP_DSC_FIELD = "instance_name"
592 __slots__ = ["instance_name"]
595 class OpRecreateInstanceDisks(OpCode):
596 """Deactivate an instance's disks."""
597 OP_ID = "OP_INSTANCE_RECREATE_DISKS"
598 OP_DSC_FIELD = "instance_name"
599 __slots__ = ["instance_name", "disks"]
602 class OpQueryInstances(OpCode):
603 """Compute the list of instances."""
604 OP_ID = "OP_INSTANCE_QUERY"
605 __slots__ = ["output_fields", "names", "use_locking"]
608 class OpQueryInstanceData(OpCode):
609 """Compute the run-time status of instances."""
610 OP_ID = "OP_INSTANCE_QUERY_DATA"
611 __slots__ = ["instances", "static"]
614 class OpSetInstanceParams(OpCode):
615 """Change the parameters of an instance."""
616 OP_ID = "OP_INSTANCE_SET_PARAMS"
617 OP_DSC_FIELD = "instance_name"
620 "hvparams", "beparams", "force",
625 class OpGrowDisk(OpCode):
626 """Grow a disk of an instance."""
627 OP_ID = "OP_INSTANCE_GROW_DISK"
628 OP_DSC_FIELD = "instance_name"
630 "instance_name", "disk", "amount", "wait_for_sync",
635 class OpDiagnoseOS(OpCode):
636 """Compute the list of guest operating systems."""
637 OP_ID = "OP_OS_DIAGNOSE"
638 __slots__ = ["output_fields", "names"]
642 class OpQueryExports(OpCode):
643 """Compute the list of exported images."""
644 OP_ID = "OP_BACKUP_QUERY"
645 __slots__ = ["nodes", "use_locking"]
648 class OpExportInstance(OpCode):
649 """Export an instance."""
650 OP_ID = "OP_BACKUP_EXPORT"
651 OP_DSC_FIELD = "instance_name"
653 "instance_name", "target_node", "shutdown", "shutdown_timeout",
657 class OpRemoveExport(OpCode):
658 """Remove an instance's export."""
659 OP_ID = "OP_BACKUP_REMOVE"
660 OP_DSC_FIELD = "instance_name"
661 __slots__ = ["instance_name"]
665 class OpGetTags(OpCode):
666 """Returns the tags of the given object."""
667 OP_ID = "OP_TAGS_GET"
668 OP_DSC_FIELD = "name"
669 __slots__ = ["kind", "name"]
672 class OpSearchTags(OpCode):
673 """Searches the tags in the cluster for a given pattern."""
674 OP_ID = "OP_TAGS_SEARCH"
675 OP_DSC_FIELD = "pattern"
676 __slots__ = ["pattern"]
679 class OpAddTags(OpCode):
680 """Add a list of tags on a given object."""
681 OP_ID = "OP_TAGS_SET"
682 __slots__ = ["kind", "name", "tags"]
685 class OpDelTags(OpCode):
686 """Remove a list of tags from a given object."""
687 OP_ID = "OP_TAGS_DEL"
688 __slots__ = ["kind", "name", "tags"]
692 class OpTestDelay(OpCode):
693 """Sleeps for a configured amount of time.
695 This is used just for debugging and testing.
698 - duration: the time to sleep
699 - on_master: if true, sleep on the master
700 - on_nodes: list of nodes in which to sleep
702 If the on_master parameter is true, it will execute a sleep on the
703 master (before any node sleep).
705 If the on_nodes list is not empty, it will sleep on those nodes
706 (after the sleep on the master, if that is enabled).
708 As an additional feature, the case of duration < 0 will be reported
709 as an execution error, so this opcode can be used as a failure
710 generator. The case of duration == 0 will not be treated specially.
713 OP_ID = "OP_TEST_DELAY"
714 OP_DSC_FIELD = "duration"
715 __slots__ = ["duration", "on_master", "on_nodes"]
718 class OpTestAllocator(OpCode):
719 """Allocator framework testing.
721 This opcode has two modes:
722 - gather and return allocator input for a given mode (allocate new
723 or replace secondary) and a given instance definition (direction
725 - run a selected allocator for a given operation (as above) and
726 return the allocator output (direction 'out')
729 OP_ID = "OP_TEST_ALLOCATOR"
730 OP_DSC_FIELD = "allocator"
732 "direction", "mode", "allocator", "name",
733 "mem_size", "disks", "disk_template",
734 "os", "tags", "nics", "vcpus", "hypervisor",
739 OP_MAPPING = dict([(v.OP_ID, v) for v in globals().values()
740 if (isinstance(v, type) and issubclass(v, OpCode) and
741 hasattr(v, "OP_ID"))])