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
22 """Module implementing the logic behind the cluster operations
24 This module implements the logic for doing operations in the cluster. There
25 are two kinds of classes defined:
26 - logical units, which know how to deal with their specific opcode only
27 - the processor, which dispatches the opcodes to their logical units
35 from ganeti import opcodes
36 from ganeti import constants
37 from ganeti import errors
38 from ganeti import rpc
39 from ganeti import cmdlib
40 from ganeti import locking
41 from ganeti import utils
44 class LockAcquireTimeout(Exception):
45 """Exception to report timeouts on acquiring locks.
50 def _CalculateLockAttemptTimeouts():
51 """Calculate timeouts for lock attempts.
56 # Wait for a total of at least 150s before doing a blocking acquire
57 while sum(result) < 150.0:
58 timeout = (result[-1] * 1.05) ** 1.25
60 # Cap timeout at 10 seconds. This gives other jobs a chance to run
61 # even if we're still trying to get our locks, before finally moving
62 # to a blocking acquire.
67 # Lower boundary for safety
70 result.append(timeout)
75 class LockAttemptTimeoutStrategy(object):
76 """Class with lock acquire timeout strategy.
85 _TIMEOUT_PER_ATTEMPT = _CalculateLockAttemptTimeouts()
87 def __init__(self, _time_fn=time.time, _random_fn=random.random):
88 """Initializes this class.
90 @param _time_fn: Time function for unittests
91 @param _random_fn: Random number generator for unittests
96 self._timeouts = iter(self._TIMEOUT_PER_ATTEMPT)
97 self._time_fn = _time_fn
98 self._random_fn = _random_fn
100 def NextAttempt(self):
101 """Returns the timeout for the next attempt.
105 timeout = self._timeouts.next()
106 except StopIteration:
107 # No more timeouts, do blocking acquire
110 if timeout is not None:
111 # Add a small variation (-/+ 5%) to timeout. This helps in situations
112 # where two or more jobs are fighting for the same lock(s).
113 variation_range = timeout * 0.1
114 timeout += ((self._random_fn() * variation_range) -
115 (variation_range * 0.5))
120 class OpExecCbBase: # pylint: disable-msg=W0232
121 """Base class for OpCode execution callbacks.
124 def NotifyStart(self):
125 """Called when we are about to execute the LU.
127 This function is called when we're about to start the lu's Exec() method,
128 that is, after we have acquired all locks.
132 def Feedback(self, *args):
133 """Sends feedback from the LU code to the end-user.
137 def CheckCancel(self):
138 """Check whether job has been cancelled.
143 class Processor(object):
144 """Object which runs OpCodes"""
147 opcodes.OpPostInitCluster: cmdlib.LUPostInitCluster,
148 opcodes.OpDestroyCluster: cmdlib.LUDestroyCluster,
149 opcodes.OpQueryClusterInfo: cmdlib.LUQueryClusterInfo,
150 opcodes.OpVerifyCluster: cmdlib.LUVerifyCluster,
151 opcodes.OpQueryConfigValues: cmdlib.LUQueryConfigValues,
152 opcodes.OpRenameCluster: cmdlib.LURenameCluster,
153 opcodes.OpVerifyDisks: cmdlib.LUVerifyDisks,
154 opcodes.OpSetClusterParams: cmdlib.LUSetClusterParams,
155 opcodes.OpRedistributeConfig: cmdlib.LURedistributeConfig,
156 opcodes.OpRepairDiskSizes: cmdlib.LURepairDiskSizes,
157 opcodes.OpQuery: cmdlib.LUQuery,
158 opcodes.OpQueryFields: cmdlib.LUQueryFields,
160 opcodes.OpAddNode: cmdlib.LUAddNode,
161 opcodes.OpQueryNodes: cmdlib.LUQueryNodes,
162 opcodes.OpQueryNodeVolumes: cmdlib.LUQueryNodeVolumes,
163 opcodes.OpQueryNodeStorage: cmdlib.LUQueryNodeStorage,
164 opcodes.OpModifyNodeStorage: cmdlib.LUModifyNodeStorage,
165 opcodes.OpRepairNodeStorage: cmdlib.LURepairNodeStorage,
166 opcodes.OpRemoveNode: cmdlib.LURemoveNode,
167 opcodes.OpSetNodeParams: cmdlib.LUSetNodeParams,
168 opcodes.OpPowercycleNode: cmdlib.LUPowercycleNode,
169 opcodes.OpMigrateNode: cmdlib.LUMigrateNode,
170 opcodes.OpNodeEvacuationStrategy: cmdlib.LUNodeEvacuationStrategy,
172 opcodes.OpCreateInstance: cmdlib.LUCreateInstance,
173 opcodes.OpReinstallInstance: cmdlib.LUReinstallInstance,
174 opcodes.OpRemoveInstance: cmdlib.LURemoveInstance,
175 opcodes.OpRenameInstance: cmdlib.LURenameInstance,
176 opcodes.OpActivateInstanceDisks: cmdlib.LUActivateInstanceDisks,
177 opcodes.OpShutdownInstance: cmdlib.LUShutdownInstance,
178 opcodes.OpStartupInstance: cmdlib.LUStartupInstance,
179 opcodes.OpRebootInstance: cmdlib.LURebootInstance,
180 opcodes.OpDeactivateInstanceDisks: cmdlib.LUDeactivateInstanceDisks,
181 opcodes.OpReplaceDisks: cmdlib.LUReplaceDisks,
182 opcodes.OpRecreateInstanceDisks: cmdlib.LURecreateInstanceDisks,
183 opcodes.OpFailoverInstance: cmdlib.LUFailoverInstance,
184 opcodes.OpMigrateInstance: cmdlib.LUMigrateInstance,
185 opcodes.OpMoveInstance: cmdlib.LUMoveInstance,
186 opcodes.OpConnectConsole: cmdlib.LUConnectConsole,
187 opcodes.OpQueryInstances: cmdlib.LUQueryInstances,
188 opcodes.OpQueryInstanceData: cmdlib.LUQueryInstanceData,
189 opcodes.OpSetInstanceParams: cmdlib.LUSetInstanceParams,
190 opcodes.OpGrowDisk: cmdlib.LUGrowDisk,
192 opcodes.OpAddGroup: cmdlib.LUAddGroup,
193 opcodes.OpQueryGroups: cmdlib.LUQueryGroups,
194 opcodes.OpSetGroupParams: cmdlib.LUSetGroupParams,
195 opcodes.OpRemoveGroup: cmdlib.LURemoveGroup,
196 opcodes.OpRenameGroup: cmdlib.LURenameGroup,
198 opcodes.OpDiagnoseOS: cmdlib.LUDiagnoseOS,
200 opcodes.OpQueryExports: cmdlib.LUQueryExports,
201 opcodes.OpPrepareExport: cmdlib.LUPrepareExport,
202 opcodes.OpExportInstance: cmdlib.LUExportInstance,
203 opcodes.OpRemoveExport: cmdlib.LURemoveExport,
205 opcodes.OpGetTags: cmdlib.LUGetTags,
206 opcodes.OpSearchTags: cmdlib.LUSearchTags,
207 opcodes.OpAddTags: cmdlib.LUAddTags,
208 opcodes.OpDelTags: cmdlib.LUDelTags,
210 opcodes.OpTestDelay: cmdlib.LUTestDelay,
211 opcodes.OpTestAllocator: cmdlib.LUTestAllocator,
212 opcodes.OpTestJobqueue: cmdlib.LUTestJobqueue,
214 opcodes.OpOobCommand: cmdlib.LUOobCommand,
217 def __init__(self, context, ec_id):
218 """Constructor for Processor
220 @type context: GanetiContext
221 @param context: global Ganeti context
223 @param ec_id: execution context identifier
226 self.context = context
229 self.rpc = rpc.RpcRunner(context.cfg)
230 self.hmclass = HooksMaster
232 def _AcquireLocks(self, level, names, shared, timeout, priority):
233 """Acquires locks via the Ganeti lock manager.
236 @param level: Lock level
237 @type names: list or string
238 @param names: Lock names
240 @param shared: Whether the locks should be acquired in shared mode
241 @type timeout: None or float
242 @param timeout: Timeout for acquiring the locks
243 @raise LockAcquireTimeout: In case locks couldn't be acquired in specified
248 self._cbs.CheckCancel()
250 acquired = self.context.glm.acquire(level, names, shared=shared,
251 timeout=timeout, priority=priority)
254 raise LockAcquireTimeout()
258 def _ExecLU(self, lu):
259 """Logical Unit execution sequence.
262 write_count = self.context.cfg.write_count
264 hm = HooksMaster(self.rpc.call_hooks_runner, lu)
265 h_results = hm.RunPhase(constants.HOOKS_PHASE_PRE)
266 lu.HooksCallBack(constants.HOOKS_PHASE_PRE, h_results,
269 if getattr(lu.op, "dry_run", False):
270 # in this mode, no post-hooks are run, and the config is not
271 # written (as it might have been modified by another LU, and we
272 # shouldn't do writeout on behalf of other threads
273 self.LogInfo("dry-run mode requested, not actually executing"
275 return lu.dry_run_result
278 result = lu.Exec(self.Log)
279 h_results = hm.RunPhase(constants.HOOKS_PHASE_POST)
280 result = lu.HooksCallBack(constants.HOOKS_PHASE_POST, h_results,
283 # FIXME: This needs locks if not lu_class.REQ_BGL
284 if write_count != self.context.cfg.write_count:
289 def _LockAndExecLU(self, lu, level, calc_timeout, priority):
290 """Execute a Logical Unit, with the needed locks.
292 This is a recursive function that starts locking the given level, and
293 proceeds up, till there are no more locks to acquire. Then it executes the
294 given LU and its opcodes.
297 adding_locks = level in lu.add_locks
298 acquiring_locks = level in lu.needed_locks
299 if level not in locking.LEVELS:
301 self._cbs.NotifyStart()
303 result = self._ExecLU(lu)
305 elif adding_locks and acquiring_locks:
306 # We could both acquire and add locks at the same level, but for now we
307 # don't need this, so we'll avoid the complicated code needed.
308 raise NotImplementedError("Can't declare locks to acquire when adding"
311 elif adding_locks or acquiring_locks:
312 lu.DeclareLocks(level)
313 share = lu.share_locks[level]
316 assert adding_locks ^ acquiring_locks, \
317 "Locks must be either added or acquired"
321 needed_locks = lu.needed_locks[level]
323 acquired = self._AcquireLocks(level, needed_locks, share,
324 calc_timeout(), priority)
327 add_locks = lu.add_locks[level]
328 lu.remove_locks[level] = add_locks
331 self.context.glm.add(level, add_locks, acquired=1, shared=share)
332 except errors.LockError:
333 raise errors.OpPrereqError(
334 "Couldn't add locks (%s), probably because of a race condition"
335 " with another job, who added them first" % add_locks,
341 lu.acquired_locks[level] = acquired
343 result = self._LockAndExecLU(lu, level + 1, calc_timeout, priority)
345 if level in lu.remove_locks:
346 self.context.glm.remove(level, lu.remove_locks[level])
348 if self.context.glm.is_owned(level):
349 self.context.glm.release(level)
352 result = self._LockAndExecLU(lu, level + 1, calc_timeout, priority)
356 def ExecOpCode(self, op, cbs, timeout=None, priority=None):
357 """Execute an opcode.
359 @type op: an OpCode instance
360 @param op: the opcode to be executed
361 @type cbs: L{OpExecCbBase}
362 @param cbs: Runtime callbacks
363 @type timeout: float or None
364 @param timeout: Maximum time to acquire all locks, None for no timeout
365 @type priority: number or None
366 @param priority: Priority for acquiring lock(s)
367 @raise LockAcquireTimeout: In case locks couldn't be acquired in specified
371 if not isinstance(op, opcodes.OpCode):
372 raise errors.ProgrammerError("Non-opcode instance passed"
375 lu_class = self.DISPATCH_TABLE.get(op.__class__, None)
377 raise errors.OpCodeUnknown("Unknown opcode")
380 calc_timeout = lambda: None
382 calc_timeout = utils.RunningTimeout(timeout, False).Remaining
386 # Acquire the Big Ganeti Lock exclusively if this LU requires it,
387 # and in a shared fashion otherwise (to prevent concurrent run with
389 self._AcquireLocks(locking.LEVEL_CLUSTER, locking.BGL,
390 not lu_class.REQ_BGL, calc_timeout(),
393 lu = lu_class(self, op, self.context, self.rpc)
395 assert lu.needed_locks is not None, "needed_locks not set by LU"
398 return self._LockAndExecLU(lu, locking.LEVEL_INSTANCE, calc_timeout,
402 self.context.cfg.DropECReservations(self._ec_id)
404 self.context.glm.release(locking.LEVEL_CLUSTER)
408 def Log(self, *args):
409 """Forward call to feedback callback function.
413 self._cbs.Feedback(*args)
415 def LogStep(self, current, total, message):
416 """Log a change in LU execution progress.
419 logging.debug("Step %d/%d %s", current, total, message)
420 self.Log("STEP %d/%d %s" % (current, total, message))
422 def LogWarning(self, message, *args, **kwargs):
423 """Log a warning to the logs and the user.
425 The optional keyword argument is 'hint' and can be used to show a
426 hint to the user (presumably related to the warning). If the
427 message is empty, it will not be printed at all, allowing one to
431 assert not kwargs or (len(kwargs) == 1 and "hint" in kwargs), \
432 "Invalid keyword arguments for LogWarning (%s)" % str(kwargs)
434 message = message % tuple(args)
436 logging.warning(message)
437 self.Log(" - WARNING: %s" % message)
439 self.Log(" Hint: %s" % kwargs["hint"])
441 def LogInfo(self, message, *args):
442 """Log an informational message to the logs and the user.
446 message = message % tuple(args)
447 logging.info(message)
448 self.Log(" - INFO: %s" % message)
452 errors.ProgrammerError("Tried to use execution context id when not set")
456 class HooksMaster(object):
459 This class distributes the run commands to the nodes based on the
462 In order to remove the direct dependency on the rpc module, the
463 constructor needs a function which actually does the remote
464 call. This will usually be rpc.call_hooks_runner, but any function
465 which behaves the same works.
468 def __init__(self, callfn, lu):
472 self.env, node_list_pre, node_list_post = self._BuildEnv()
473 self.node_list = {constants.HOOKS_PHASE_PRE: node_list_pre,
474 constants.HOOKS_PHASE_POST: node_list_post}
477 """Compute the environment and the target nodes.
479 Based on the opcode and the current node list, this builds the
480 environment for the hooks and the target node list for the run.
484 "PATH": "/sbin:/bin:/usr/sbin:/usr/bin",
485 "GANETI_HOOKS_VERSION": constants.HOOKS_VERSION,
486 "GANETI_OP_CODE": self.op.OP_ID,
487 "GANETI_OBJECT_TYPE": self.lu.HTYPE,
488 "GANETI_DATA_DIR": constants.DATA_DIR,
491 if self.lu.HPATH is not None:
492 lu_env, lu_nodes_pre, lu_nodes_post = self.lu.BuildHooksEnv()
495 env["GANETI_" + key] = lu_env[key]
497 lu_nodes_pre = lu_nodes_post = []
499 return env, frozenset(lu_nodes_pre), frozenset(lu_nodes_post)
501 def _RunWrapper(self, node_list, hpath, phase):
502 """Simple wrapper over self.callfn.
504 This method fixes the environment before doing the rpc call.
507 env = self.env.copy()
508 env["GANETI_HOOKS_PHASE"] = phase
509 env["GANETI_HOOKS_PATH"] = hpath
510 if self.lu.cfg is not None:
511 env["GANETI_CLUSTER"] = self.lu.cfg.GetClusterName()
512 env["GANETI_MASTER"] = self.lu.cfg.GetMasterNode()
514 env = dict([(str(key), str(val)) for key, val in env.iteritems()])
516 return self.callfn(node_list, hpath, phase, env)
518 def RunPhase(self, phase, nodes=None):
519 """Run all the scripts for a phase.
521 This is the main function of the HookMaster.
523 @param phase: one of L{constants.HOOKS_PHASE_POST} or
524 L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
525 @param nodes: overrides the predefined list of nodes for the given phase
526 @return: the processed results of the hooks multi-node rpc call
527 @raise errors.HooksFailure: on communication failure to the nodes
528 @raise errors.HooksAbort: on failure of one of the hooks
531 if not self.node_list[phase] and not nodes:
532 # empty node list, we should not attempt to run this as either
533 # we're in the cluster init phase and the rpc client part can't
534 # even attempt to run, or this LU doesn't do hooks at all
536 hpath = self.lu.HPATH
537 if nodes is not None:
538 results = self._RunWrapper(nodes, hpath, phase)
540 results = self._RunWrapper(self.node_list[phase], hpath, phase)
543 msg = "Communication Failure"
544 if phase == constants.HOOKS_PHASE_PRE:
545 raise errors.HooksFailure(msg)
547 self.lu.LogWarning(msg)
549 for node_name in results:
550 res = results[node_name]
555 self.lu.LogWarning("Communication failure to node %s: %s",
558 for script, hkr, output in res.payload:
559 if hkr == constants.HKR_FAIL:
560 if phase == constants.HOOKS_PHASE_PRE:
561 errs.append((node_name, script, output))
564 output = "(no output)"
565 self.lu.LogWarning("On %s script %s failed, output: %s" %
566 (node_name, script, output))
567 if errs and phase == constants.HOOKS_PHASE_PRE:
568 raise errors.HooksAbort(errs)
571 def RunConfigUpdate(self):
572 """Run the special configuration update hook
574 This is a special hook that runs only on the master after each
575 top-level LI if the configuration has been updated.
578 phase = constants.HOOKS_PHASE_POST
579 hpath = constants.HOOKS_NAME_CFGUPDATE
580 nodes = [self.lu.cfg.GetMasterNode()]
581 self._RunWrapper(nodes, hpath, phase)