Statistics
| Branch: | Tag: | Revision:

root / lib / mcpu.py @ a0d2fe2c

History | View | Annotate | Download (18.8 kB)

1
#
2
#
3

    
4
# Copyright (C) 2006, 2007 Google Inc.
5
#
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.
10
#
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.
15
#
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
19
# 02110-1301, USA.
20

    
21

    
22
"""Module implementing the logic behind the cluster operations
23

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
28

29
"""
30

    
31
import logging
32
import random
33
import time
34

    
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

    
42

    
43
class _LockAcquireTimeout(Exception):
44
  """Internal exception to report timeouts on acquiring locks.
45

46
  """
47

    
48

    
49
def _CalculateLockAttemptTimeouts():
50
  """Calculate timeouts for lock attempts.
51

52
  """
53
  result = [1.0]
54

    
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
58

    
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.
62
    if timeout > 10.0:
63
      timeout = 10.0
64

    
65
    elif timeout < 0.1:
66
      # Lower boundary for safety
67
      timeout = 0.1
68

    
69
    result.append(timeout)
70

    
71
  return result
72

    
73

    
74
class _LockAttemptTimeoutStrategy(object):
75
  """Class with lock acquire timeout strategy.
76

77
  """
78
  __slots__ = [
79
    "_attempt",
80
    "_random_fn",
81
    "_start_time",
82
    "_time_fn",
83
    "_running_timeout",
84
    ]
85

    
86
  _TIMEOUT_PER_ATTEMPT = _CalculateLockAttemptTimeouts()
87

    
88
  def __init__(self, attempt=0, _time_fn=time.time, _random_fn=random.random):
89
    """Initializes this class.
90

91
    @type attempt: int
92
    @param attempt: Current attempt number
93
    @param _time_fn: Time function for unittests
94
    @param _random_fn: Random number generator for unittests
95

96
    """
97
    object.__init__(self)
98

    
99
    if attempt < 0:
100
      raise ValueError("Attempt must be zero or positive")
101

    
102
    self._attempt = attempt
103
    self._time_fn = _time_fn
104
    self._random_fn = _random_fn
105

    
106
    try:
107
      timeout = self._TIMEOUT_PER_ATTEMPT[attempt]
108
    except IndexError:
109
      # No more timeouts, do blocking acquire
110
      timeout = None
111

    
112
    self._running_timeout = locking.RunningTimeout(timeout, False,
113
                                                   _time_fn=_time_fn)
114

    
115
  def NextAttempt(self):
116
    """Returns the strategy for the next attempt.
117

118
    """
119
    return _LockAttemptTimeoutStrategy(attempt=self._attempt + 1,
120
                                       _time_fn=self._time_fn,
121
                                       _random_fn=self._random_fn)
122

    
123
  def CalcRemainingTimeout(self):
124
    """Returns the remaining timeout.
125

126
    """
127
    timeout = self._running_timeout.Remaining()
128

    
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))
135

    
136
    return timeout
137

    
138

    
139
class OpExecCbBase: # pylint: disable-msg=W0232
140
  """Base class for OpCode execution callbacks.
141

142
  """
143
  def NotifyStart(self):
144
    """Called when we are about to execute the LU.
145

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.
148

149
    """
150

    
151
  def Feedback(self, *args):
152
    """Sends feedback from the LU code to the end-user.
153

154
    """
155

    
156
  def CheckCancel(self):
157
    """Check whether job has been cancelled.
158

159
    """
160

    
161

    
162
class Processor(object):
163
  """Object which runs OpCodes"""
164
  DISPATCH_TABLE = {
165
    # Cluster
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,
176
    # node lu
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,
188
    # instance lu
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,
208
    # os lu
209
    opcodes.OpDiagnoseOS: cmdlib.LUDiagnoseOS,
210
    # exports lu
211
    opcodes.OpQueryExports: cmdlib.LUQueryExports,
212
    opcodes.OpPrepareExport: cmdlib.LUPrepareExport,
213
    opcodes.OpExportInstance: cmdlib.LUExportInstance,
214
    opcodes.OpRemoveExport: cmdlib.LURemoveExport,
215
    # tags lu
216
    opcodes.OpGetTags: cmdlib.LUGetTags,
217
    opcodes.OpSearchTags: cmdlib.LUSearchTags,
218
    opcodes.OpAddTags: cmdlib.LUAddTags,
219
    opcodes.OpDelTags: cmdlib.LUDelTags,
220
    # test lu
221
    opcodes.OpTestDelay: cmdlib.LUTestDelay,
222
    opcodes.OpTestAllocator: cmdlib.LUTestAllocator,
223
    opcodes.OpTestJobqueue: cmdlib.LUTestJobqueue,
224
    }
225

    
226
  def __init__(self, context, ec_id):
227
    """Constructor for Processor
228

229
    @type context: GanetiContext
230
    @param context: global Ganeti context
231
    @type ec_id: string
232
    @param ec_id: execution context identifier
233

234
    """
235
    self.context = context
236
    self._ec_id = ec_id
237
    self._cbs = None
238
    self.rpc = rpc.RpcRunner(context.cfg)
239
    self.hmclass = HooksMaster
240

    
241
  def _AcquireLocks(self, level, names, shared, timeout):
242
    """Acquires locks via the Ganeti lock manager.
243

244
    @type level: int
245
    @param level: Lock level
246
    @type names: list or string
247
    @param names: Lock names
248
    @type shared: bool
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
252

253
    """
254
    if self._cbs:
255
      self._cbs.CheckCancel()
256

    
257
    acquired = self.context.glm.acquire(level, names, shared=shared,
258
                                        timeout=timeout)
259

    
260
    return acquired
261

    
262
  def _ExecLU(self, lu):
263
    """Logical Unit execution sequence.
264

265
    """
266
    write_count = self.context.cfg.write_count
267
    lu.CheckPrereq()
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,
271
                     self.Log, None)
272

    
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"
278
                   " the operation")
279
      return lu.dry_run_result
280

    
281
    try:
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,
285
                                self.Log, result)
286
    finally:
287
      # FIXME: This needs locks if not lu_class.REQ_BGL
288
      if write_count != self.context.cfg.write_count:
289
        hm.RunConfigUpdate()
290

    
291
    return result
292

    
293
  def _LockAndExecLU(self, lu, level, calc_timeout):
294
    """Execute a Logical Unit, with the needed locks.
295

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.
299

300
    """
301
    adding_locks = level in lu.add_locks
302
    acquiring_locks = level in lu.needed_locks
303
    if level not in locking.LEVELS:
304
      if self._cbs:
305
        self._cbs.NotifyStart()
306

    
307
      result = self._ExecLU(lu)
308

    
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"
313
                                " others")
314

    
315
    elif adding_locks or acquiring_locks:
316
      lu.DeclareLocks(level)
317
      share = lu.share_locks[level]
318

    
319
      try:
320
        assert adding_locks ^ acquiring_locks, \
321
          "Locks must be either added or acquired"
322

    
323
        if acquiring_locks:
324
          # Acquiring locks
325
          needed_locks = lu.needed_locks[level]
326

    
327
          acquired = self._AcquireLocks(level, needed_locks, share,
328
                                        calc_timeout())
329

    
330
          if acquired is None:
331
            raise _LockAcquireTimeout()
332

    
333
        else:
334
          # Adding locks
335
          add_locks = lu.add_locks[level]
336
          lu.remove_locks[level] = add_locks
337

    
338
          try:
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,
344
              errors.ECODE_FAULT)
345

    
346
          acquired = add_locks
347

    
348
        try:
349
          lu.acquired_locks[level] = acquired
350

    
351
          result = self._LockAndExecLU(lu, level + 1, calc_timeout)
352
        finally:
353
          if level in lu.remove_locks:
354
            self.context.glm.remove(level, lu.remove_locks[level])
355
      finally:
356
        if self.context.glm.is_owned(level):
357
          self.context.glm.release(level)
358

    
359
    else:
360
      result = self._LockAndExecLU(lu, level + 1, calc_timeout)
361

    
362
    return result
363

    
364
  def ExecOpCode(self, op, cbs):
365
    """Execute an opcode.
366

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

372
    """
373
    if not isinstance(op, opcodes.OpCode):
374
      raise errors.ProgrammerError("Non-opcode instance passed"
375
                                   " to ExecOpcode")
376

    
377
    self._cbs = cbs
378
    try:
379
      lu_class = self.DISPATCH_TABLE.get(op.__class__, None)
380
      if lu_class is None:
381
        raise errors.OpCodeUnknown("Unknown opcode")
382

    
383
      timeout_strategy = _LockAttemptTimeoutStrategy()
384

    
385
      while True:
386
        try:
387
          acquire_timeout = timeout_strategy.CalcRemainingTimeout()
388

    
389
          # Acquire the Big Ganeti Lock exclusively if this LU requires it,
390
          # and in a shared fashion otherwise (to prevent concurrent run with
391
          # an exclusive LU.
392
          if self._AcquireLocks(locking.LEVEL_CLUSTER, locking.BGL,
393
                                not lu_class.REQ_BGL, acquire_timeout) is None:
394
            raise _LockAcquireTimeout()
395

    
396
          try:
397
            lu = lu_class(self, op, self.context, self.rpc)
398
            lu.ExpandNames()
399
            assert lu.needed_locks is not None, "needed_locks not set by LU"
400

    
401
            try:
402
              return self._LockAndExecLU(lu, locking.LEVEL_INSTANCE,
403
                                         timeout_strategy.CalcRemainingTimeout)
404
            finally:
405
              if self._ec_id:
406
                self.context.cfg.DropECReservations(self._ec_id)
407

    
408
          finally:
409
            self.context.glm.release(locking.LEVEL_CLUSTER)
410

    
411
        except _LockAcquireTimeout:
412
          # Timeout while waiting for lock, try again
413
          pass
414

    
415
        timeout_strategy = timeout_strategy.NextAttempt()
416

    
417
    finally:
418
      self._cbs = None
419

    
420
  def Log(self, *args):
421
    """Forward call to feedback callback function.
422

423
    """
424
    if self._cbs:
425
      self._cbs.Feedback(*args)
426

    
427
  def LogStep(self, current, total, message):
428
    """Log a change in LU execution progress.
429

430
    """
431
    logging.debug("Step %d/%d %s", current, total, message)
432
    self.Log("STEP %d/%d %s" % (current, total, message))
433

    
434
  def LogWarning(self, message, *args, **kwargs):
435
    """Log a warning to the logs and the user.
436

437
    The optional keyword argument is 'hint' and can be used to show a
438
    hint to the user (presumably related to the warning). If the
439
    message is empty, it will not be printed at all, allowing one to
440
    show only a hint.
441

442
    """
443
    assert not kwargs or (len(kwargs) == 1 and "hint" in kwargs), \
444
           "Invalid keyword arguments for LogWarning (%s)" % str(kwargs)
445
    if args:
446
      message = message % tuple(args)
447
    if message:
448
      logging.warning(message)
449
      self.Log(" - WARNING: %s" % message)
450
    if "hint" in kwargs:
451
      self.Log("      Hint: %s" % kwargs["hint"])
452

    
453
  def LogInfo(self, message, *args):
454
    """Log an informational message to the logs and the user.
455

456
    """
457
    if args:
458
      message = message % tuple(args)
459
    logging.info(message)
460
    self.Log(" - INFO: %s" % message)
461

    
462
  def GetECId(self):
463
    if not self._ec_id:
464
      errors.ProgrammerError("Tried to use execution context id when not set")
465
    return self._ec_id
466

    
467

    
468
class HooksMaster(object):
469
  """Hooks master.
470

471
  This class distributes the run commands to the nodes based on the
472
  specific LU class.
473

474
  In order to remove the direct dependency on the rpc module, the
475
  constructor needs a function which actually does the remote
476
  call. This will usually be rpc.call_hooks_runner, but any function
477
  which behaves the same works.
478

479
  """
480
  def __init__(self, callfn, lu):
481
    self.callfn = callfn
482
    self.lu = lu
483
    self.op = lu.op
484
    self.env, node_list_pre, node_list_post = self._BuildEnv()
485
    self.node_list = {constants.HOOKS_PHASE_PRE: node_list_pre,
486
                      constants.HOOKS_PHASE_POST: node_list_post}
487

    
488
  def _BuildEnv(self):
489
    """Compute the environment and the target nodes.
490

491
    Based on the opcode and the current node list, this builds the
492
    environment for the hooks and the target node list for the run.
493

494
    """
495
    env = {
496
      "PATH": "/sbin:/bin:/usr/sbin:/usr/bin",
497
      "GANETI_HOOKS_VERSION": constants.HOOKS_VERSION,
498
      "GANETI_OP_CODE": self.op.OP_ID,
499
      "GANETI_OBJECT_TYPE": self.lu.HTYPE,
500
      "GANETI_DATA_DIR": constants.DATA_DIR,
501
      }
502

    
503
    if self.lu.HPATH is not None:
504
      lu_env, lu_nodes_pre, lu_nodes_post = self.lu.BuildHooksEnv()
505
      if lu_env:
506
        for key in lu_env:
507
          env["GANETI_" + key] = lu_env[key]
508
    else:
509
      lu_nodes_pre = lu_nodes_post = []
510

    
511
    return env, frozenset(lu_nodes_pre), frozenset(lu_nodes_post)
512

    
513
  def _RunWrapper(self, node_list, hpath, phase):
514
    """Simple wrapper over self.callfn.
515

516
    This method fixes the environment before doing the rpc call.
517

518
    """
519
    env = self.env.copy()
520
    env["GANETI_HOOKS_PHASE"] = phase
521
    env["GANETI_HOOKS_PATH"] = hpath
522
    if self.lu.cfg is not None:
523
      env["GANETI_CLUSTER"] = self.lu.cfg.GetClusterName()
524
      env["GANETI_MASTER"] = self.lu.cfg.GetMasterNode()
525

    
526
    env = dict([(str(key), str(val)) for key, val in env.iteritems()])
527

    
528
    return self.callfn(node_list, hpath, phase, env)
529

    
530
  def RunPhase(self, phase, nodes=None):
531
    """Run all the scripts for a phase.
532

533
    This is the main function of the HookMaster.
534

535
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
536
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
537
    @param nodes: overrides the predefined list of nodes for the given phase
538
    @return: the processed results of the hooks multi-node rpc call
539
    @raise errors.HooksFailure: on communication failure to the nodes
540
    @raise errors.HooksAbort: on failure of one of the hooks
541

542
    """
543
    if not self.node_list[phase] and not nodes:
544
      # empty node list, we should not attempt to run this as either
545
      # we're in the cluster init phase and the rpc client part can't
546
      # even attempt to run, or this LU doesn't do hooks at all
547
      return
548
    hpath = self.lu.HPATH
549
    if nodes is not None:
550
      results = self._RunWrapper(nodes, hpath, phase)
551
    else:
552
      results = self._RunWrapper(self.node_list[phase], hpath, phase)
553
    errs = []
554
    if not results:
555
      msg = "Communication Failure"
556
      if phase == constants.HOOKS_PHASE_PRE:
557
        raise errors.HooksFailure(msg)
558
      else:
559
        self.lu.LogWarning(msg)
560
        return results
561
    for node_name in results:
562
      res = results[node_name]
563
      if res.offline:
564
        continue
565
      msg = res.fail_msg
566
      if msg:
567
        self.lu.LogWarning("Communication failure to node %s: %s",
568
                           node_name, msg)
569
        continue
570
      for script, hkr, output in res.payload:
571
        if hkr == constants.HKR_FAIL:
572
          if phase == constants.HOOKS_PHASE_PRE:
573
            errs.append((node_name, script, output))
574
          else:
575
            if not output:
576
              output = "(no output)"
577
            self.lu.LogWarning("On %s script %s failed, output: %s" %
578
                               (node_name, script, output))
579
    if errs and phase == constants.HOOKS_PHASE_PRE:
580
      raise errors.HooksAbort(errs)
581
    return results
582

    
583
  def RunConfigUpdate(self):
584
    """Run the special configuration update hook
585

586
    This is a special hook that runs only on the master after each
587
    top-level LI if the configuration has been updated.
588

589
    """
590
    phase = constants.HOOKS_PHASE_POST
591
    hpath = constants.HOOKS_NAME_CFGUPDATE
592
    nodes = [self.lu.cfg.GetMasterNode()]
593
    self._RunWrapper(nodes, hpath, phase)