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.OpQueryGroups: cmdlib.LUQueryGroups,
194 opcodes.OpDiagnoseOS: cmdlib.LUDiagnoseOS,
196 opcodes.OpQueryExports: cmdlib.LUQueryExports,
197 opcodes.OpPrepareExport: cmdlib.LUPrepareExport,
198 opcodes.OpExportInstance: cmdlib.LUExportInstance,
199 opcodes.OpRemoveExport: cmdlib.LURemoveExport,
201 opcodes.OpGetTags: cmdlib.LUGetTags,
202 opcodes.OpSearchTags: cmdlib.LUSearchTags,
203 opcodes.OpAddTags: cmdlib.LUAddTags,
204 opcodes.OpDelTags: cmdlib.LUDelTags,
206 opcodes.OpTestDelay: cmdlib.LUTestDelay,
207 opcodes.OpTestAllocator: cmdlib.LUTestAllocator,
208 opcodes.OpTestJobqueue: cmdlib.LUTestJobqueue,
211 def __init__(self, context, ec_id):
212 """Constructor for Processor
214 @type context: GanetiContext
215 @param context: global Ganeti context
217 @param ec_id: execution context identifier
220 self.context = context
223 self.rpc = rpc.RpcRunner(context.cfg)
224 self.hmclass = HooksMaster
226 def _AcquireLocks(self, level, names, shared, timeout, priority):
227 """Acquires locks via the Ganeti lock manager.
230 @param level: Lock level
231 @type names: list or string
232 @param names: Lock names
234 @param shared: Whether the locks should be acquired in shared mode
235 @type timeout: None or float
236 @param timeout: Timeout for acquiring the locks
237 @raise LockAcquireTimeout: In case locks couldn't be acquired in specified
242 self._cbs.CheckCancel()
244 acquired = self.context.glm.acquire(level, names, shared=shared,
245 timeout=timeout, priority=priority)
248 raise LockAcquireTimeout()
252 def _ExecLU(self, lu):
253 """Logical Unit execution sequence.
256 write_count = self.context.cfg.write_count
258 hm = HooksMaster(self.rpc.call_hooks_runner, lu)
259 h_results = hm.RunPhase(constants.HOOKS_PHASE_PRE)
260 lu.HooksCallBack(constants.HOOKS_PHASE_PRE, h_results,
263 if getattr(lu.op, "dry_run", False):
264 # in this mode, no post-hooks are run, and the config is not
265 # written (as it might have been modified by another LU, and we
266 # shouldn't do writeout on behalf of other threads
267 self.LogInfo("dry-run mode requested, not actually executing"
269 return lu.dry_run_result
272 result = lu.Exec(self.Log)
273 h_results = hm.RunPhase(constants.HOOKS_PHASE_POST)
274 result = lu.HooksCallBack(constants.HOOKS_PHASE_POST, h_results,
277 # FIXME: This needs locks if not lu_class.REQ_BGL
278 if write_count != self.context.cfg.write_count:
283 def _LockAndExecLU(self, lu, level, calc_timeout, priority):
284 """Execute a Logical Unit, with the needed locks.
286 This is a recursive function that starts locking the given level, and
287 proceeds up, till there are no more locks to acquire. Then it executes the
288 given LU and its opcodes.
291 adding_locks = level in lu.add_locks
292 acquiring_locks = level in lu.needed_locks
293 if level not in locking.LEVELS:
295 self._cbs.NotifyStart()
297 result = self._ExecLU(lu)
299 elif adding_locks and acquiring_locks:
300 # We could both acquire and add locks at the same level, but for now we
301 # don't need this, so we'll avoid the complicated code needed.
302 raise NotImplementedError("Can't declare locks to acquire when adding"
305 elif adding_locks or acquiring_locks:
306 lu.DeclareLocks(level)
307 share = lu.share_locks[level]
310 assert adding_locks ^ acquiring_locks, \
311 "Locks must be either added or acquired"
315 needed_locks = lu.needed_locks[level]
317 acquired = self._AcquireLocks(level, needed_locks, share,
318 calc_timeout(), priority)
321 add_locks = lu.add_locks[level]
322 lu.remove_locks[level] = add_locks
325 self.context.glm.add(level, add_locks, acquired=1, shared=share)
326 except errors.LockError:
327 raise errors.OpPrereqError(
328 "Couldn't add locks (%s), probably because of a race condition"
329 " with another job, who added them first" % add_locks,
335 lu.acquired_locks[level] = acquired
337 result = self._LockAndExecLU(lu, level + 1, calc_timeout, priority)
339 if level in lu.remove_locks:
340 self.context.glm.remove(level, lu.remove_locks[level])
342 if self.context.glm.is_owned(level):
343 self.context.glm.release(level)
346 result = self._LockAndExecLU(lu, level + 1, calc_timeout, priority)
350 def ExecOpCode(self, op, cbs, timeout=None, priority=None):
351 """Execute an opcode.
353 @type op: an OpCode instance
354 @param op: the opcode to be executed
355 @type cbs: L{OpExecCbBase}
356 @param cbs: Runtime callbacks
357 @type timeout: float or None
358 @param timeout: Maximum time to acquire all locks, None for no timeout
359 @type priority: number or None
360 @param priority: Priority for acquiring lock(s)
361 @raise LockAcquireTimeout: In case locks couldn't be acquired in specified
365 if not isinstance(op, opcodes.OpCode):
366 raise errors.ProgrammerError("Non-opcode instance passed"
369 lu_class = self.DISPATCH_TABLE.get(op.__class__, None)
371 raise errors.OpCodeUnknown("Unknown opcode")
374 calc_timeout = lambda: None
376 calc_timeout = utils.RunningTimeout(timeout, False).Remaining
380 # Acquire the Big Ganeti Lock exclusively if this LU requires it,
381 # and in a shared fashion otherwise (to prevent concurrent run with
383 self._AcquireLocks(locking.LEVEL_CLUSTER, locking.BGL,
384 not lu_class.REQ_BGL, calc_timeout(),
387 lu = lu_class(self, op, self.context, self.rpc)
389 assert lu.needed_locks is not None, "needed_locks not set by LU"
392 return self._LockAndExecLU(lu, locking.LEVEL_INSTANCE, calc_timeout,
396 self.context.cfg.DropECReservations(self._ec_id)
398 self.context.glm.release(locking.LEVEL_CLUSTER)
402 def Log(self, *args):
403 """Forward call to feedback callback function.
407 self._cbs.Feedback(*args)
409 def LogStep(self, current, total, message):
410 """Log a change in LU execution progress.
413 logging.debug("Step %d/%d %s", current, total, message)
414 self.Log("STEP %d/%d %s" % (current, total, message))
416 def LogWarning(self, message, *args, **kwargs):
417 """Log a warning to the logs and the user.
419 The optional keyword argument is 'hint' and can be used to show a
420 hint to the user (presumably related to the warning). If the
421 message is empty, it will not be printed at all, allowing one to
425 assert not kwargs or (len(kwargs) == 1 and "hint" in kwargs), \
426 "Invalid keyword arguments for LogWarning (%s)" % str(kwargs)
428 message = message % tuple(args)
430 logging.warning(message)
431 self.Log(" - WARNING: %s" % message)
433 self.Log(" Hint: %s" % kwargs["hint"])
435 def LogInfo(self, message, *args):
436 """Log an informational message to the logs and the user.
440 message = message % tuple(args)
441 logging.info(message)
442 self.Log(" - INFO: %s" % message)
446 errors.ProgrammerError("Tried to use execution context id when not set")
450 class HooksMaster(object):
453 This class distributes the run commands to the nodes based on the
456 In order to remove the direct dependency on the rpc module, the
457 constructor needs a function which actually does the remote
458 call. This will usually be rpc.call_hooks_runner, but any function
459 which behaves the same works.
462 def __init__(self, callfn, lu):
466 self.env, node_list_pre, node_list_post = self._BuildEnv()
467 self.node_list = {constants.HOOKS_PHASE_PRE: node_list_pre,
468 constants.HOOKS_PHASE_POST: node_list_post}
471 """Compute the environment and the target nodes.
473 Based on the opcode and the current node list, this builds the
474 environment for the hooks and the target node list for the run.
478 "PATH": "/sbin:/bin:/usr/sbin:/usr/bin",
479 "GANETI_HOOKS_VERSION": constants.HOOKS_VERSION,
480 "GANETI_OP_CODE": self.op.OP_ID,
481 "GANETI_OBJECT_TYPE": self.lu.HTYPE,
482 "GANETI_DATA_DIR": constants.DATA_DIR,
485 if self.lu.HPATH is not None:
486 lu_env, lu_nodes_pre, lu_nodes_post = self.lu.BuildHooksEnv()
489 env["GANETI_" + key] = lu_env[key]
491 lu_nodes_pre = lu_nodes_post = []
493 return env, frozenset(lu_nodes_pre), frozenset(lu_nodes_post)
495 def _RunWrapper(self, node_list, hpath, phase):
496 """Simple wrapper over self.callfn.
498 This method fixes the environment before doing the rpc call.
501 env = self.env.copy()
502 env["GANETI_HOOKS_PHASE"] = phase
503 env["GANETI_HOOKS_PATH"] = hpath
504 if self.lu.cfg is not None:
505 env["GANETI_CLUSTER"] = self.lu.cfg.GetClusterName()
506 env["GANETI_MASTER"] = self.lu.cfg.GetMasterNode()
508 env = dict([(str(key), str(val)) for key, val in env.iteritems()])
510 return self.callfn(node_list, hpath, phase, env)
512 def RunPhase(self, phase, nodes=None):
513 """Run all the scripts for a phase.
515 This is the main function of the HookMaster.
517 @param phase: one of L{constants.HOOKS_PHASE_POST} or
518 L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
519 @param nodes: overrides the predefined list of nodes for the given phase
520 @return: the processed results of the hooks multi-node rpc call
521 @raise errors.HooksFailure: on communication failure to the nodes
522 @raise errors.HooksAbort: on failure of one of the hooks
525 if not self.node_list[phase] and not nodes:
526 # empty node list, we should not attempt to run this as either
527 # we're in the cluster init phase and the rpc client part can't
528 # even attempt to run, or this LU doesn't do hooks at all
530 hpath = self.lu.HPATH
531 if nodes is not None:
532 results = self._RunWrapper(nodes, hpath, phase)
534 results = self._RunWrapper(self.node_list[phase], hpath, phase)
537 msg = "Communication Failure"
538 if phase == constants.HOOKS_PHASE_PRE:
539 raise errors.HooksFailure(msg)
541 self.lu.LogWarning(msg)
543 for node_name in results:
544 res = results[node_name]
549 self.lu.LogWarning("Communication failure to node %s: %s",
552 for script, hkr, output in res.payload:
553 if hkr == constants.HKR_FAIL:
554 if phase == constants.HOOKS_PHASE_PRE:
555 errs.append((node_name, script, output))
558 output = "(no output)"
559 self.lu.LogWarning("On %s script %s failed, output: %s" %
560 (node_name, script, output))
561 if errs and phase == constants.HOOKS_PHASE_PRE:
562 raise errors.HooksAbort(errs)
565 def RunConfigUpdate(self):
566 """Run the special configuration update hook
568 This is a special hook that runs only on the master after each
569 top-level LI if the configuration has been updated.
572 phase = constants.HOOKS_PHASE_POST
573 hpath = constants.HOOKS_NAME_CFGUPDATE
574 nodes = [self.lu.cfg.GetMasterNode()]
575 self._RunWrapper(nodes, hpath, phase)