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 """Internal 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 ReportLocks(self, msg):
157 """Report lock operations.
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.OpEvacuateNode: cmdlib.LUEvacuateNode,
187 opcodes.OpMigrateNode: cmdlib.LUMigrateNode,
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.OpExportInstance: cmdlib.LUExportInstance,
213 opcodes.OpRemoveExport: cmdlib.LURemoveExport,
215 opcodes.OpGetTags: cmdlib.LUGetTags,
216 opcodes.OpSearchTags: cmdlib.LUSearchTags,
217 opcodes.OpAddTags: cmdlib.LUAddTags,
218 opcodes.OpDelTags: cmdlib.LUDelTags,
220 opcodes.OpTestDelay: cmdlib.LUTestDelay,
221 opcodes.OpTestAllocator: cmdlib.LUTestAllocator,
224 def __init__(self, context, ec_id):
225 """Constructor for Processor
227 @type context: GanetiContext
228 @param context: global Ganeti context
230 @param ec_id: execution context identifier
233 self.context = context
236 self.rpc = rpc.RpcRunner(context.cfg)
237 self.hmclass = HooksMaster
239 def _ReportLocks(self, level, names, shared, timeout, acquired, result):
240 """Reports lock operations.
243 @param level: Lock level
244 @type names: list or string
245 @param names: Lock names
247 @param shared: Whether the locks should be acquired in shared mode
248 @type timeout: None or float
249 @param timeout: Timeout for acquiring the locks
251 @param acquired: Whether the locks have already been acquired
252 @type result: None or set
253 @param result: Result from L{locking.GanetiLockManager.acquire}
261 parts.append("timeout")
263 parts.append("acquired")
265 parts.append("waiting")
267 parts.append("blocking")
269 parts.append("timeout=%0.6fs" % timeout)
271 parts.append(locking.LEVEL_NAMES[level])
273 if names == locking.ALL_SET:
275 elif isinstance(names, basestring):
278 parts.append(",".join(sorted(names)))
281 parts.append("shared")
283 parts.append("exclusive")
285 msg = "/".join(parts)
287 logging.debug("LU locks %s", msg)
290 self._cbs.ReportLocks(msg)
292 def _AcquireLocks(self, level, names, shared, timeout):
293 """Acquires locks via the Ganeti lock manager.
296 @param level: Lock level
297 @type names: list or string
298 @param names: Lock names
300 @param shared: Whether the locks should be acquired in shared mode
301 @type timeout: None or float
302 @param timeout: Timeout for acquiring the locks
305 self._ReportLocks(level, names, shared, timeout, False, None)
307 acquired = self.context.glm.acquire(level, names, shared=shared,
310 self._ReportLocks(level, names, shared, timeout, True, acquired)
314 def _ExecLU(self, lu):
315 """Logical Unit execution sequence.
318 write_count = self.context.cfg.write_count
320 hm = HooksMaster(self.rpc.call_hooks_runner, lu)
321 h_results = hm.RunPhase(constants.HOOKS_PHASE_PRE)
322 lu.HooksCallBack(constants.HOOKS_PHASE_PRE, h_results,
323 self._Feedback, None)
325 if getattr(lu.op, "dry_run", False):
326 # in this mode, no post-hooks are run, and the config is not
327 # written (as it might have been modified by another LU, and we
328 # shouldn't do writeout on behalf of other threads
329 self.LogInfo("dry-run mode requested, not actually executing"
331 return lu.dry_run_result
334 result = lu.Exec(self._Feedback)
335 h_results = hm.RunPhase(constants.HOOKS_PHASE_POST)
336 result = lu.HooksCallBack(constants.HOOKS_PHASE_POST, h_results,
337 self._Feedback, result)
339 # FIXME: This needs locks if not lu_class.REQ_BGL
340 if write_count != self.context.cfg.write_count:
345 def _LockAndExecLU(self, lu, level, calc_timeout):
346 """Execute a Logical Unit, with the needed locks.
348 This is a recursive function that starts locking the given level, and
349 proceeds up, till there are no more locks to acquire. Then it executes the
350 given LU and its opcodes.
353 adding_locks = level in lu.add_locks
354 acquiring_locks = level in lu.needed_locks
355 if level not in locking.LEVELS:
357 self._cbs.NotifyStart()
359 result = self._ExecLU(lu)
361 elif adding_locks and acquiring_locks:
362 # We could both acquire and add locks at the same level, but for now we
363 # don't need this, so we'll avoid the complicated code needed.
364 raise NotImplementedError("Can't declare locks to acquire when adding"
367 elif adding_locks or acquiring_locks:
368 lu.DeclareLocks(level)
369 share = lu.share_locks[level]
372 assert adding_locks ^ acquiring_locks, \
373 "Locks must be either added or acquired"
377 needed_locks = lu.needed_locks[level]
379 acquired = self._AcquireLocks(level, needed_locks, share,
383 raise _LockAcquireTimeout()
387 add_locks = lu.add_locks[level]
388 lu.remove_locks[level] = add_locks
391 self.context.glm.add(level, add_locks, acquired=1, shared=share)
392 except errors.LockError:
393 raise errors.OpPrereqError(
394 "Couldn't add locks (%s), probably because of a race condition"
395 " with another job, who added them first" % add_locks,
401 lu.acquired_locks[level] = acquired
403 result = self._LockAndExecLU(lu, level + 1, calc_timeout)
405 if level in lu.remove_locks:
406 self.context.glm.remove(level, lu.remove_locks[level])
408 if self.context.glm.is_owned(level):
409 self.context.glm.release(level)
412 result = self._LockAndExecLU(lu, level + 1, calc_timeout)
416 def ExecOpCode(self, op, cbs):
417 """Execute an opcode.
419 @type op: an OpCode instance
420 @param op: the opcode to be executed
421 @type cbs: L{OpExecCbBase}
422 @param cbs: Runtime callbacks
425 if not isinstance(op, opcodes.OpCode):
426 raise errors.ProgrammerError("Non-opcode instance passed"
431 lu_class = self.DISPATCH_TABLE.get(op.__class__, None)
433 raise errors.OpCodeUnknown("Unknown opcode")
435 timeout_strategy = _LockAttemptTimeoutStrategy()
439 acquire_timeout = timeout_strategy.CalcRemainingTimeout()
441 # Acquire the Big Ganeti Lock exclusively if this LU requires it,
442 # and in a shared fashion otherwise (to prevent concurrent run with
444 if self._AcquireLocks(locking.LEVEL_CLUSTER, locking.BGL,
445 not lu_class.REQ_BGL, acquire_timeout) is None:
446 raise _LockAcquireTimeout()
449 lu = lu_class(self, op, self.context, self.rpc)
451 assert lu.needed_locks is not None, "needed_locks not set by LU"
454 return self._LockAndExecLU(lu, locking.LEVEL_INSTANCE,
455 timeout_strategy.CalcRemainingTimeout)
458 self.context.cfg.DropECReservations(self._ec_id)
461 self.context.glm.release(locking.LEVEL_CLUSTER)
463 except _LockAcquireTimeout:
464 # Timeout while waiting for lock, try again
467 timeout_strategy = timeout_strategy.NextAttempt()
472 def _Feedback(self, *args):
473 """Forward call to feedback callback function.
477 self._cbs.Feedback(*args)
479 def LogStep(self, current, total, message):
480 """Log a change in LU execution progress.
483 logging.debug("Step %d/%d %s", current, total, message)
484 self._Feedback("STEP %d/%d %s" % (current, total, message))
486 def LogWarning(self, message, *args, **kwargs):
487 """Log a warning to the logs and the user.
489 The optional keyword argument is 'hint' and can be used to show a
490 hint to the user (presumably related to the warning). If the
491 message is empty, it will not be printed at all, allowing one to
495 assert not kwargs or (len(kwargs) == 1 and "hint" in kwargs), \
496 "Invalid keyword arguments for LogWarning (%s)" % str(kwargs)
498 message = message % tuple(args)
500 logging.warning(message)
501 self._Feedback(" - WARNING: %s" % message)
503 self._Feedback(" Hint: %s" % kwargs["hint"])
505 def LogInfo(self, message, *args):
506 """Log an informational message to the logs and the user.
510 message = message % tuple(args)
511 logging.info(message)
512 self._Feedback(" - INFO: %s" % message)
516 errors.ProgrammerError("Tried to use execution context id when not set")
520 class HooksMaster(object):
523 This class distributes the run commands to the nodes based on the
526 In order to remove the direct dependency on the rpc module, the
527 constructor needs a function which actually does the remote
528 call. This will usually be rpc.call_hooks_runner, but any function
529 which behaves the same works.
532 def __init__(self, callfn, lu):
536 self.env, node_list_pre, node_list_post = self._BuildEnv()
537 self.node_list = {constants.HOOKS_PHASE_PRE: node_list_pre,
538 constants.HOOKS_PHASE_POST: node_list_post}
541 """Compute the environment and the target nodes.
543 Based on the opcode and the current node list, this builds the
544 environment for the hooks and the target node list for the run.
548 "PATH": "/sbin:/bin:/usr/sbin:/usr/bin",
549 "GANETI_HOOKS_VERSION": constants.HOOKS_VERSION,
550 "GANETI_OP_CODE": self.op.OP_ID,
551 "GANETI_OBJECT_TYPE": self.lu.HTYPE,
552 "GANETI_DATA_DIR": constants.DATA_DIR,
555 if self.lu.HPATH is not None:
556 lu_env, lu_nodes_pre, lu_nodes_post = self.lu.BuildHooksEnv()
559 env["GANETI_" + key] = lu_env[key]
561 lu_nodes_pre = lu_nodes_post = []
563 return env, frozenset(lu_nodes_pre), frozenset(lu_nodes_post)
565 def _RunWrapper(self, node_list, hpath, phase):
566 """Simple wrapper over self.callfn.
568 This method fixes the environment before doing the rpc call.
571 env = self.env.copy()
572 env["GANETI_HOOKS_PHASE"] = phase
573 env["GANETI_HOOKS_PATH"] = hpath
574 if self.lu.cfg is not None:
575 env["GANETI_CLUSTER"] = self.lu.cfg.GetClusterName()
576 env["GANETI_MASTER"] = self.lu.cfg.GetMasterNode()
578 env = dict([(str(key), str(val)) for key, val in env.iteritems()])
580 return self.callfn(node_list, hpath, phase, env)
582 def RunPhase(self, phase, nodes=None):
583 """Run all the scripts for a phase.
585 This is the main function of the HookMaster.
587 @param phase: one of L{constants.HOOKS_PHASE_POST} or
588 L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
589 @param nodes: overrides the predefined list of nodes for the given phase
590 @return: the processed results of the hooks multi-node rpc call
591 @raise errors.HooksFailure: on communication failure to the nodes
592 @raise errors.HooksAbort: on failure of one of the hooks
595 if not self.node_list[phase] and not nodes:
596 # empty node list, we should not attempt to run this as either
597 # we're in the cluster init phase and the rpc client part can't
598 # even attempt to run, or this LU doesn't do hooks at all
600 hpath = self.lu.HPATH
601 if nodes is not None:
602 results = self._RunWrapper(nodes, hpath, phase)
604 results = self._RunWrapper(self.node_list[phase], hpath, phase)
607 msg = "Communication Failure"
608 if phase == constants.HOOKS_PHASE_PRE:
609 raise errors.HooksFailure(msg)
611 self.lu.LogWarning(msg)
613 for node_name in results:
614 res = results[node_name]
619 self.lu.LogWarning("Communication failure to node %s: %s",
622 for script, hkr, output in res.payload:
623 if hkr == constants.HKR_FAIL:
624 if phase == constants.HOOKS_PHASE_PRE:
625 errs.append((node_name, script, output))
628 output = "(no output)"
629 self.lu.LogWarning("On %s script %s failed, output: %s" %
630 (node_name, script, output))
631 if errs and phase == constants.HOOKS_PHASE_PRE:
632 raise errors.HooksAbort(errs)
635 def RunConfigUpdate(self):
636 """Run the special configuration update hook
638 This is a special hook that runs only on the master after each
639 top-level LI if the configuration has been updated.
642 phase = constants.HOOKS_PHASE_POST
643 hpath = constants.HOOKS_NAME_CFGUPDATE
644 nodes = [self.lu.cfg.GetMasterNode()]
645 self._RunWrapper(nodes, hpath, phase)