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
43 class LockAcquireTimeout(Exception):
44 """Exception to report timeouts on acquiring locks.
49 def _CalculateLockAttemptTimeouts():
50 """Calculate timeouts for lock attempts.
55 # Wait for a total of at least 150s before doing a blocking acquire
56 while sum(result) < 150.0:
57 timeout = (result[-1] * 1.05) ** 1.25
59 # Cap timeout at 10 seconds. This gives other jobs a chance to run
60 # even if we're still trying to get our locks, before finally moving
61 # to a blocking acquire.
66 # Lower boundary for safety
69 result.append(timeout)
74 class _LockAttemptTimeoutStrategy(object):
75 """Class with lock acquire timeout strategy.
86 _TIMEOUT_PER_ATTEMPT = _CalculateLockAttemptTimeouts()
88 def __init__(self, attempt=0, _time_fn=time.time, _random_fn=random.random):
89 """Initializes this class.
92 @param attempt: Current attempt number
93 @param _time_fn: Time function for unittests
94 @param _random_fn: Random number generator for unittests
100 raise ValueError("Attempt must be zero or positive")
102 self._attempt = attempt
103 self._time_fn = _time_fn
104 self._random_fn = _random_fn
107 timeout = self._TIMEOUT_PER_ATTEMPT[attempt]
109 # No more timeouts, do blocking acquire
112 self._running_timeout = locking.RunningTimeout(timeout, False,
115 def NextAttempt(self):
116 """Returns the strategy for the next attempt.
119 return _LockAttemptTimeoutStrategy(attempt=self._attempt + 1,
120 _time_fn=self._time_fn,
121 _random_fn=self._random_fn)
123 def CalcRemainingTimeout(self):
124 """Returns the remaining timeout.
127 timeout = self._running_timeout.Remaining()
129 if timeout is not None:
130 # Add a small variation (-/+ 5%) to timeout. This helps in situations
131 # where two or more jobs are fighting for the same lock(s).
132 variation_range = timeout * 0.1
133 timeout += ((self._random_fn() * variation_range) -
134 (variation_range * 0.5))
139 class OpExecCbBase: # pylint: disable-msg=W0232
140 """Base class for OpCode execution callbacks.
143 def NotifyStart(self):
144 """Called when we are about to execute the LU.
146 This function is called when we're about to start the lu's Exec() method,
147 that is, after we have acquired all locks.
151 def Feedback(self, *args):
152 """Sends feedback from the LU code to the end-user.
156 def CheckCancel(self):
157 """Check whether job has been cancelled.
162 class Processor(object):
163 """Object which runs OpCodes"""
166 opcodes.OpPostInitCluster: cmdlib.LUPostInitCluster,
167 opcodes.OpDestroyCluster: cmdlib.LUDestroyCluster,
168 opcodes.OpQueryClusterInfo: cmdlib.LUQueryClusterInfo,
169 opcodes.OpVerifyCluster: cmdlib.LUVerifyCluster,
170 opcodes.OpQueryConfigValues: cmdlib.LUQueryConfigValues,
171 opcodes.OpRenameCluster: cmdlib.LURenameCluster,
172 opcodes.OpVerifyDisks: cmdlib.LUVerifyDisks,
173 opcodes.OpSetClusterParams: cmdlib.LUSetClusterParams,
174 opcodes.OpRedistributeConfig: cmdlib.LURedistributeConfig,
175 opcodes.OpRepairDiskSizes: cmdlib.LURepairDiskSizes,
177 opcodes.OpAddNode: cmdlib.LUAddNode,
178 opcodes.OpQueryNodes: cmdlib.LUQueryNodes,
179 opcodes.OpQueryNodeVolumes: cmdlib.LUQueryNodeVolumes,
180 opcodes.OpQueryNodeStorage: cmdlib.LUQueryNodeStorage,
181 opcodes.OpModifyNodeStorage: cmdlib.LUModifyNodeStorage,
182 opcodes.OpRepairNodeStorage: cmdlib.LURepairNodeStorage,
183 opcodes.OpRemoveNode: cmdlib.LURemoveNode,
184 opcodes.OpSetNodeParams: cmdlib.LUSetNodeParams,
185 opcodes.OpPowercycleNode: cmdlib.LUPowercycleNode,
186 opcodes.OpMigrateNode: cmdlib.LUMigrateNode,
187 opcodes.OpNodeEvacuationStrategy: cmdlib.LUNodeEvacuationStrategy,
189 opcodes.OpCreateInstance: cmdlib.LUCreateInstance,
190 opcodes.OpReinstallInstance: cmdlib.LUReinstallInstance,
191 opcodes.OpRemoveInstance: cmdlib.LURemoveInstance,
192 opcodes.OpRenameInstance: cmdlib.LURenameInstance,
193 opcodes.OpActivateInstanceDisks: cmdlib.LUActivateInstanceDisks,
194 opcodes.OpShutdownInstance: cmdlib.LUShutdownInstance,
195 opcodes.OpStartupInstance: cmdlib.LUStartupInstance,
196 opcodes.OpRebootInstance: cmdlib.LURebootInstance,
197 opcodes.OpDeactivateInstanceDisks: cmdlib.LUDeactivateInstanceDisks,
198 opcodes.OpReplaceDisks: cmdlib.LUReplaceDisks,
199 opcodes.OpRecreateInstanceDisks: cmdlib.LURecreateInstanceDisks,
200 opcodes.OpFailoverInstance: cmdlib.LUFailoverInstance,
201 opcodes.OpMigrateInstance: cmdlib.LUMigrateInstance,
202 opcodes.OpMoveInstance: cmdlib.LUMoveInstance,
203 opcodes.OpConnectConsole: cmdlib.LUConnectConsole,
204 opcodes.OpQueryInstances: cmdlib.LUQueryInstances,
205 opcodes.OpQueryInstanceData: cmdlib.LUQueryInstanceData,
206 opcodes.OpSetInstanceParams: cmdlib.LUSetInstanceParams,
207 opcodes.OpGrowDisk: cmdlib.LUGrowDisk,
209 opcodes.OpDiagnoseOS: cmdlib.LUDiagnoseOS,
211 opcodes.OpQueryExports: cmdlib.LUQueryExports,
212 opcodes.OpPrepareExport: cmdlib.LUPrepareExport,
213 opcodes.OpExportInstance: cmdlib.LUExportInstance,
214 opcodes.OpRemoveExport: cmdlib.LURemoveExport,
216 opcodes.OpGetTags: cmdlib.LUGetTags,
217 opcodes.OpSearchTags: cmdlib.LUSearchTags,
218 opcodes.OpAddTags: cmdlib.LUAddTags,
219 opcodes.OpDelTags: cmdlib.LUDelTags,
221 opcodes.OpTestDelay: cmdlib.LUTestDelay,
222 opcodes.OpTestAllocator: cmdlib.LUTestAllocator,
223 opcodes.OpTestJobqueue: cmdlib.LUTestJobqueue,
226 def __init__(self, context, ec_id):
227 """Constructor for Processor
229 @type context: GanetiContext
230 @param context: global Ganeti context
232 @param ec_id: execution context identifier
235 self.context = context
238 self.rpc = rpc.RpcRunner(context.cfg)
239 self.hmclass = HooksMaster
241 def _AcquireLocks(self, level, names, shared, timeout):
242 """Acquires locks via the Ganeti lock manager.
245 @param level: Lock level
246 @type names: list or string
247 @param names: Lock names
249 @param shared: Whether the locks should be acquired in shared mode
250 @type timeout: None or float
251 @param timeout: Timeout for acquiring the locks
255 self._cbs.CheckCancel()
257 acquired = self.context.glm.acquire(level, names, shared=shared,
262 def _ExecLU(self, lu):
263 """Logical Unit execution sequence.
266 write_count = self.context.cfg.write_count
268 hm = HooksMaster(self.rpc.call_hooks_runner, lu)
269 h_results = hm.RunPhase(constants.HOOKS_PHASE_PRE)
270 lu.HooksCallBack(constants.HOOKS_PHASE_PRE, h_results,
273 if getattr(lu.op, "dry_run", False):
274 # in this mode, no post-hooks are run, and the config is not
275 # written (as it might have been modified by another LU, and we
276 # shouldn't do writeout on behalf of other threads
277 self.LogInfo("dry-run mode requested, not actually executing"
279 return lu.dry_run_result
282 result = lu.Exec(self.Log)
283 h_results = hm.RunPhase(constants.HOOKS_PHASE_POST)
284 result = lu.HooksCallBack(constants.HOOKS_PHASE_POST, h_results,
287 # FIXME: This needs locks if not lu_class.REQ_BGL
288 if write_count != self.context.cfg.write_count:
293 def _LockAndExecLU(self, lu, level, calc_timeout):
294 """Execute a Logical Unit, with the needed locks.
296 This is a recursive function that starts locking the given level, and
297 proceeds up, till there are no more locks to acquire. Then it executes the
298 given LU and its opcodes.
301 adding_locks = level in lu.add_locks
302 acquiring_locks = level in lu.needed_locks
303 if level not in locking.LEVELS:
305 self._cbs.NotifyStart()
307 result = self._ExecLU(lu)
309 elif adding_locks and acquiring_locks:
310 # We could both acquire and add locks at the same level, but for now we
311 # don't need this, so we'll avoid the complicated code needed.
312 raise NotImplementedError("Can't declare locks to acquire when adding"
315 elif adding_locks or acquiring_locks:
316 lu.DeclareLocks(level)
317 share = lu.share_locks[level]
320 assert adding_locks ^ acquiring_locks, \
321 "Locks must be either added or acquired"
325 needed_locks = lu.needed_locks[level]
327 acquired = self._AcquireLocks(level, needed_locks, share,
331 raise LockAcquireTimeout()
335 add_locks = lu.add_locks[level]
336 lu.remove_locks[level] = add_locks
339 self.context.glm.add(level, add_locks, acquired=1, shared=share)
340 except errors.LockError:
341 raise errors.OpPrereqError(
342 "Couldn't add locks (%s), probably because of a race condition"
343 " with another job, who added them first" % add_locks,
349 lu.acquired_locks[level] = acquired
351 result = self._LockAndExecLU(lu, level + 1, calc_timeout)
353 if level in lu.remove_locks:
354 self.context.glm.remove(level, lu.remove_locks[level])
356 if self.context.glm.is_owned(level):
357 self.context.glm.release(level)
360 result = self._LockAndExecLU(lu, level + 1, calc_timeout)
364 def ExecOpCode(self, op, cbs, timeout=None):
365 """Execute an opcode.
367 @type op: an OpCode instance
368 @param op: the opcode to be executed
369 @type cbs: L{OpExecCbBase}
370 @param cbs: Runtime callbacks
371 @type timeout: float or None
372 @param timeout: Maximum time to acquire all locks, None for no timeout
373 @raise LockAcquireTimeout: In case locks couldn't be acquired in specified
377 if not isinstance(op, opcodes.OpCode):
378 raise errors.ProgrammerError("Non-opcode instance passed"
381 lu_class = self.DISPATCH_TABLE.get(op.__class__, None)
383 raise errors.OpCodeUnknown("Unknown opcode")
386 calc_timeout = lambda: None
388 calc_timeout = locking.RunningTimeout(timeout, False).Remaining
392 # Acquire the Big Ganeti Lock exclusively if this LU requires it,
393 # and in a shared fashion otherwise (to prevent concurrent run with
395 if self._AcquireLocks(locking.LEVEL_CLUSTER, locking.BGL,
396 not lu_class.REQ_BGL, calc_timeout()) is None:
397 raise LockAcquireTimeout()
400 lu = lu_class(self, op, self.context, self.rpc)
402 assert lu.needed_locks is not None, "needed_locks not set by LU"
405 return self._LockAndExecLU(lu, locking.LEVEL_INSTANCE, calc_timeout)
408 self.context.cfg.DropECReservations(self._ec_id)
410 self.context.glm.release(locking.LEVEL_CLUSTER)
414 def Log(self, *args):
415 """Forward call to feedback callback function.
419 self._cbs.Feedback(*args)
421 def LogStep(self, current, total, message):
422 """Log a change in LU execution progress.
425 logging.debug("Step %d/%d %s", current, total, message)
426 self.Log("STEP %d/%d %s" % (current, total, message))
428 def LogWarning(self, message, *args, **kwargs):
429 """Log a warning to the logs and the user.
431 The optional keyword argument is 'hint' and can be used to show a
432 hint to the user (presumably related to the warning). If the
433 message is empty, it will not be printed at all, allowing one to
437 assert not kwargs or (len(kwargs) == 1 and "hint" in kwargs), \
438 "Invalid keyword arguments for LogWarning (%s)" % str(kwargs)
440 message = message % tuple(args)
442 logging.warning(message)
443 self.Log(" - WARNING: %s" % message)
445 self.Log(" Hint: %s" % kwargs["hint"])
447 def LogInfo(self, message, *args):
448 """Log an informational message to the logs and the user.
452 message = message % tuple(args)
453 logging.info(message)
454 self.Log(" - INFO: %s" % message)
458 errors.ProgrammerError("Tried to use execution context id when not set")
462 class HooksMaster(object):
465 This class distributes the run commands to the nodes based on the
468 In order to remove the direct dependency on the rpc module, the
469 constructor needs a function which actually does the remote
470 call. This will usually be rpc.call_hooks_runner, but any function
471 which behaves the same works.
474 def __init__(self, callfn, lu):
478 self.env, node_list_pre, node_list_post = self._BuildEnv()
479 self.node_list = {constants.HOOKS_PHASE_PRE: node_list_pre,
480 constants.HOOKS_PHASE_POST: node_list_post}
483 """Compute the environment and the target nodes.
485 Based on the opcode and the current node list, this builds the
486 environment for the hooks and the target node list for the run.
490 "PATH": "/sbin:/bin:/usr/sbin:/usr/bin",
491 "GANETI_HOOKS_VERSION": constants.HOOKS_VERSION,
492 "GANETI_OP_CODE": self.op.OP_ID,
493 "GANETI_OBJECT_TYPE": self.lu.HTYPE,
494 "GANETI_DATA_DIR": constants.DATA_DIR,
497 if self.lu.HPATH is not None:
498 lu_env, lu_nodes_pre, lu_nodes_post = self.lu.BuildHooksEnv()
501 env["GANETI_" + key] = lu_env[key]
503 lu_nodes_pre = lu_nodes_post = []
505 return env, frozenset(lu_nodes_pre), frozenset(lu_nodes_post)
507 def _RunWrapper(self, node_list, hpath, phase):
508 """Simple wrapper over self.callfn.
510 This method fixes the environment before doing the rpc call.
513 env = self.env.copy()
514 env["GANETI_HOOKS_PHASE"] = phase
515 env["GANETI_HOOKS_PATH"] = hpath
516 if self.lu.cfg is not None:
517 env["GANETI_CLUSTER"] = self.lu.cfg.GetClusterName()
518 env["GANETI_MASTER"] = self.lu.cfg.GetMasterNode()
520 env = dict([(str(key), str(val)) for key, val in env.iteritems()])
522 return self.callfn(node_list, hpath, phase, env)
524 def RunPhase(self, phase, nodes=None):
525 """Run all the scripts for a phase.
527 This is the main function of the HookMaster.
529 @param phase: one of L{constants.HOOKS_PHASE_POST} or
530 L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
531 @param nodes: overrides the predefined list of nodes for the given phase
532 @return: the processed results of the hooks multi-node rpc call
533 @raise errors.HooksFailure: on communication failure to the nodes
534 @raise errors.HooksAbort: on failure of one of the hooks
537 if not self.node_list[phase] and not nodes:
538 # empty node list, we should not attempt to run this as either
539 # we're in the cluster init phase and the rpc client part can't
540 # even attempt to run, or this LU doesn't do hooks at all
542 hpath = self.lu.HPATH
543 if nodes is not None:
544 results = self._RunWrapper(nodes, hpath, phase)
546 results = self._RunWrapper(self.node_list[phase], hpath, phase)
549 msg = "Communication Failure"
550 if phase == constants.HOOKS_PHASE_PRE:
551 raise errors.HooksFailure(msg)
553 self.lu.LogWarning(msg)
555 for node_name in results:
556 res = results[node_name]
561 self.lu.LogWarning("Communication failure to node %s: %s",
564 for script, hkr, output in res.payload:
565 if hkr == constants.HKR_FAIL:
566 if phase == constants.HOOKS_PHASE_PRE:
567 errs.append((node_name, script, output))
570 output = "(no output)"
571 self.lu.LogWarning("On %s script %s failed, output: %s" %
572 (node_name, script, output))
573 if errs and phase == constants.HOOKS_PHASE_PRE:
574 raise errors.HooksAbort(errs)
577 def RunConfigUpdate(self):
578 """Run the special configuration update hook
580 This is a special hook that runs only on the master after each
581 top-level LI if the configuration has been updated.
584 phase = constants.HOOKS_PHASE_POST
585 hpath = constants.HOOKS_NAME_CFGUPDATE
586 nodes = [self.lu.cfg.GetMasterNode()]
587 self._RunWrapper(nodes, hpath, phase)