jqueue.py 66.5 KB
Newer Older
Iustin Pop's avatar
Iustin Pop committed
1
2
3
#
#

Michael Hanselmann's avatar
Michael Hanselmann committed
4
# Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011 Google Inc.
Iustin Pop's avatar
Iustin Pop committed
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 2 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
# General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
# 02110-1301, USA.


22
23
"""Module implementing the job queue handling.

24
25
26
27
28
Locking: there's a single, large lock in the L{JobQueue} class. It's
used by all other classes in this module.

@var JOBQUEUE_THREADS: the number of worker threads we start for
    processing jobs
29
30

"""
Iustin Pop's avatar
Iustin Pop committed
31

Michael Hanselmann's avatar
Michael Hanselmann committed
32
import logging
33
34
import errno
import re
35
import time
36
import weakref
Michael Hanselmann's avatar
Michael Hanselmann committed
37
import threading
Iustin Pop's avatar
Iustin Pop committed
38

Guido Trotter's avatar
Guido Trotter committed
39
40
41
42
43
44
45
try:
  # pylint: disable-msg=E0611
  from pyinotify import pyinotify
except ImportError:
  import pyinotify

from ganeti import asyncnotifier
Michael Hanselmann's avatar
Michael Hanselmann committed
46
from ganeti import constants
47
from ganeti import serializer
Michael Hanselmann's avatar
Michael Hanselmann committed
48
from ganeti import workerpool
49
from ganeti import locking
50
from ganeti import opcodes
Iustin Pop's avatar
Iustin Pop committed
51
from ganeti import errors
Michael Hanselmann's avatar
Michael Hanselmann committed
52
from ganeti import mcpu
53
from ganeti import utils
54
from ganeti import jstore
55
from ganeti import rpc
56
from ganeti import runtime
57
from ganeti import netutils
58
from ganeti import compat
Michael Hanselmann's avatar
Michael Hanselmann committed
59
from ganeti import ht
Michael Hanselmann's avatar
Michael Hanselmann committed
60

61

62
JOBQUEUE_THREADS = 25
63
JOBS_PER_ARCHIVE_DIRECTORY = 10000
Michael Hanselmann's avatar
Michael Hanselmann committed
64

65
66
67
# member lock names to be passed to @ssynchronized decorator
_LOCK = "_lock"
_QUEUE = "_queue"
68

Iustin Pop's avatar
Iustin Pop committed
69

70
class CancelJob(Exception):
71
72
73
74
75
  """Special exception to cancel a job.

  """


76
def TimeStampNow():
77
78
79
80
81
82
  """Returns the current timestamp.

  @rtype: tuple
  @return: the current time in the (seconds, microseconds) format

  """
83
84
85
  return utils.SplitTime(time.time())


Michael Hanselmann's avatar
Michael Hanselmann committed
86
class _QueuedOpCode(object):
Michael Hanselmann's avatar
Michael Hanselmann committed
87
  """Encapsulates an opcode object.
Michael Hanselmann's avatar
Michael Hanselmann committed
88

89
90
91
92
93
94
  @ivar log: holds the execution log and consists of tuples
  of the form C{(log_serial, timestamp, level, message)}
  @ivar input: the OpCode we encapsulate
  @ivar status: the current status
  @ivar result: the result of the LU execution
  @ivar start_timestamp: timestamp for the start of the execution
Iustin Pop's avatar
Iustin Pop committed
95
  @ivar exec_timestamp: timestamp for the actual LU Exec() function invocation
96
  @ivar stop_timestamp: timestamp for the end of the execution
97

Michael Hanselmann's avatar
Michael Hanselmann committed
98
  """
99
  __slots__ = ["input", "status", "result", "log", "priority",
Iustin Pop's avatar
Iustin Pop committed
100
               "start_timestamp", "exec_timestamp", "end_timestamp",
101
102
               "__weakref__"]

Michael Hanselmann's avatar
Michael Hanselmann committed
103
  def __init__(self, op):
104
105
106
107
108
109
    """Constructor for the _QuededOpCode.

    @type op: L{opcodes.OpCode}
    @param op: the opcode we encapsulate

    """
Michael Hanselmann's avatar
Michael Hanselmann committed
110
111
112
113
    self.input = op
    self.status = constants.OP_STATUS_QUEUED
    self.result = None
    self.log = []
114
    self.start_timestamp = None
Iustin Pop's avatar
Iustin Pop committed
115
    self.exec_timestamp = None
116
    self.end_timestamp = None
117

118
119
120
    # Get initial priority (it might change during the lifetime of this opcode)
    self.priority = getattr(op, "priority", constants.OP_PRIO_DEFAULT)

121
122
  @classmethod
  def Restore(cls, state):
123
124
125
126
127
128
129
130
    """Restore the _QueuedOpCode from the serialized form.

    @type state: dict
    @param state: the serialized state
    @rtype: _QueuedOpCode
    @return: a new _QueuedOpCode instance

    """
Michael Hanselmann's avatar
Michael Hanselmann committed
131
132
133
134
135
    obj = _QueuedOpCode.__new__(cls)
    obj.input = opcodes.OpCode.LoadOpCode(state["input"])
    obj.status = state["status"]
    obj.result = state["result"]
    obj.log = state["log"]
136
    obj.start_timestamp = state.get("start_timestamp", None)
Iustin Pop's avatar
Iustin Pop committed
137
    obj.exec_timestamp = state.get("exec_timestamp", None)
138
    obj.end_timestamp = state.get("end_timestamp", None)
139
    obj.priority = state.get("priority", constants.OP_PRIO_DEFAULT)
140
141
142
    return obj

  def Serialize(self):
143
144
145
146
147
148
    """Serializes this _QueuedOpCode.

    @rtype: dict
    @return: the dictionary holding the serialized state

    """
149
150
151
152
153
    return {
      "input": self.input.__getstate__(),
      "status": self.status,
      "result": self.result,
      "log": self.log,
154
      "start_timestamp": self.start_timestamp,
Iustin Pop's avatar
Iustin Pop committed
155
      "exec_timestamp": self.exec_timestamp,
156
      "end_timestamp": self.end_timestamp,
157
      "priority": self.priority,
158
      }
159

Michael Hanselmann's avatar
Michael Hanselmann committed
160
161
162
163

class _QueuedJob(object):
  """In-memory job representation.

164
165
166
167
168
169
170
171
172
173
174
175
176
  This is what we use to track the user-submitted jobs. Locking must
  be taken care of by users of this class.

  @type queue: L{JobQueue}
  @ivar queue: the parent queue
  @ivar id: the job ID
  @type ops: list
  @ivar ops: the list of _QueuedOpCode that constitute the job
  @type log_serial: int
  @ivar log_serial: holds the index for the next log entry
  @ivar received_timestamp: the timestamp for when the job was received
  @ivar start_timestmap: the timestamp for start of execution
  @ivar end_timestamp: the timestamp for end of execution
Michael Hanselmann's avatar
Michael Hanselmann committed
177
178

  """
Iustin Pop's avatar
Iustin Pop committed
179
  # pylint: disable-msg=W0212
180
  __slots__ = ["queue", "id", "ops", "log_serial", "ops_iter", "cur_opctx",
181
               "received_timestamp", "start_timestamp", "end_timestamp",
Michael Hanselmann's avatar
Michael Hanselmann committed
182
               "__weakref__", "processor_lock"]
183

Michael Hanselmann's avatar
Michael Hanselmann committed
184
  def __init__(self, queue, job_id, ops):
185
186
187
188
189
190
191
192
193
194
195
    """Constructor for the _QueuedJob.

    @type queue: L{JobQueue}
    @param queue: our parent queue
    @type job_id: job_id
    @param job_id: our job id
    @type ops: list
    @param ops: the list of opcodes we hold, which will be encapsulated
        in _QueuedOpCodes

    """
Michael Hanselmann's avatar
Michael Hanselmann committed
196
    if not ops:
Guido Trotter's avatar
Guido Trotter committed
197
      raise errors.GenericError("A job needs at least one opcode")
Michael Hanselmann's avatar
Michael Hanselmann committed
198

Michael Hanselmann's avatar
Michael Hanselmann committed
199
    self.queue = queue
200
    self.id = job_id
Michael Hanselmann's avatar
Michael Hanselmann committed
201
    self.ops = [_QueuedOpCode(op) for op in ops]
202
    self.log_serial = 0
203
204
205
    self.received_timestamp = TimeStampNow()
    self.start_timestamp = None
    self.end_timestamp = None
206

207
208
209
210
211
212
213
    self._InitInMemory(self)

  @staticmethod
  def _InitInMemory(obj):
    """Initializes in-memory variables.

    """
214
    obj.ops_iter = None
215
    obj.cur_opctx = None
Michael Hanselmann's avatar
Michael Hanselmann committed
216
    obj.processor_lock = threading.Lock()
217

218
219
220
221
222
223
224
  def __repr__(self):
    status = ["%s.%s" % (self.__class__.__module__, self.__class__.__name__),
              "id=%s" % self.id,
              "ops=%s" % ",".join([op.input.Summary() for op in self.ops])]

    return "<%s at %#x>" % (" ".join(status), id(self))

225
  @classmethod
Michael Hanselmann's avatar
Michael Hanselmann committed
226
  def Restore(cls, queue, state):
227
228
229
230
231
232
233
234
235
236
    """Restore a _QueuedJob from serialized state:

    @type queue: L{JobQueue}
    @param queue: to which queue the restored job belongs
    @type state: dict
    @param state: the serialized state
    @rtype: _JobQueue
    @return: the restored _JobQueue instance

    """
Michael Hanselmann's avatar
Michael Hanselmann committed
237
238
239
    obj = _QueuedJob.__new__(cls)
    obj.queue = queue
    obj.id = state["id"]
240
241
242
    obj.received_timestamp = state.get("received_timestamp", None)
    obj.start_timestamp = state.get("start_timestamp", None)
    obj.end_timestamp = state.get("end_timestamp", None)
243
244
245
246
247
248
249
250
251

    obj.ops = []
    obj.log_serial = 0
    for op_state in state["ops"]:
      op = _QueuedOpCode.Restore(op_state)
      for log_entry in op.log:
        obj.log_serial = max(obj.log_serial, log_entry[0])
      obj.ops.append(op)

252
    cls._InitInMemory(obj)
253

254
255
256
    return obj

  def Serialize(self):
257
258
259
260
261
262
    """Serialize the _JobQueue instance.

    @rtype: dict
    @return: the serialized state

    """
263
264
    return {
      "id": self.id,
Michael Hanselmann's avatar
Michael Hanselmann committed
265
      "ops": [op.Serialize() for op in self.ops],
266
267
268
      "start_timestamp": self.start_timestamp,
      "end_timestamp": self.end_timestamp,
      "received_timestamp": self.received_timestamp,
269
270
      }

Michael Hanselmann's avatar
Michael Hanselmann committed
271
  def CalcStatus(self):
272
273
274
275
276
277
278
279
280
281
    """Compute the status of this job.

    This function iterates over all the _QueuedOpCodes in the job and
    based on their status, computes the job status.

    The algorithm is:
      - if we find a cancelled, or finished with error, the job
        status will be the same
      - otherwise, the last opcode with the status one of:
          - waitlock
282
          - canceling
283
284
285
286
287
288
289
290
291
292
          - running

        will determine the job status

      - otherwise, it means either all opcodes are queued, or success,
        and the job status will be the same

    @return: the job status

    """
Michael Hanselmann's avatar
Michael Hanselmann committed
293
294
295
    status = constants.JOB_STATUS_QUEUED

    all_success = True
Michael Hanselmann's avatar
Michael Hanselmann committed
296
297
    for op in self.ops:
      if op.status == constants.OP_STATUS_SUCCESS:
Michael Hanselmann's avatar
Michael Hanselmann committed
298
299
300
301
        continue

      all_success = False

Michael Hanselmann's avatar
Michael Hanselmann committed
302
      if op.status == constants.OP_STATUS_QUEUED:
Michael Hanselmann's avatar
Michael Hanselmann committed
303
        pass
Iustin Pop's avatar
Iustin Pop committed
304
305
      elif op.status == constants.OP_STATUS_WAITLOCK:
        status = constants.JOB_STATUS_WAITLOCK
Michael Hanselmann's avatar
Michael Hanselmann committed
306
      elif op.status == constants.OP_STATUS_RUNNING:
Michael Hanselmann's avatar
Michael Hanselmann committed
307
        status = constants.JOB_STATUS_RUNNING
308
309
310
      elif op.status == constants.OP_STATUS_CANCELING:
        status = constants.JOB_STATUS_CANCELING
        break
Michael Hanselmann's avatar
Michael Hanselmann committed
311
      elif op.status == constants.OP_STATUS_ERROR:
312
313
314
        status = constants.JOB_STATUS_ERROR
        # The whole job fails if one opcode failed
        break
Michael Hanselmann's avatar
Michael Hanselmann committed
315
      elif op.status == constants.OP_STATUS_CANCELED:
316
317
        status = constants.OP_STATUS_CANCELED
        break
Michael Hanselmann's avatar
Michael Hanselmann committed
318
319
320
321
322
323

    if all_success:
      status = constants.JOB_STATUS_SUCCESS

    return status

324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
  def CalcPriority(self):
    """Gets the current priority for this job.

    Only unfinished opcodes are considered. When all are done, the default
    priority is used.

    @rtype: int

    """
    priorities = [op.priority for op in self.ops
                  if op.status not in constants.OPS_FINALIZED]

    if not priorities:
      # All opcodes are done, assume default priority
      return constants.OP_PRIO_DEFAULT

    return min(priorities)

342
  def GetLogEntries(self, newer_than):
343
344
345
    """Selectively returns the log entries.

    @type newer_than: None or int
Michael Hanselmann's avatar
Michael Hanselmann committed
346
    @param newer_than: if this is None, return all log entries,
347
348
349
350
351
352
        otherwise return only the log entries with serial higher
        than this value
    @rtype: list
    @return: the list of the log entries selected

    """
353
354
355
356
357
358
359
    if newer_than is None:
      serial = -1
    else:
      serial = newer_than

    entries = []
    for op in self.ops:
360
      entries.extend(filter(lambda entry: entry[0] > serial, op.log))
361
362
363

    return entries

364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
  def GetInfo(self, fields):
    """Returns information about a job.

    @type fields: list
    @param fields: names of fields to return
    @rtype: list
    @return: list with one element for each field
    @raise errors.OpExecError: when an invalid field
        has been passed

    """
    row = []
    for fname in fields:
      if fname == "id":
        row.append(self.id)
      elif fname == "status":
        row.append(self.CalcStatus())
381
382
      elif fname == "priority":
        row.append(self.CalcPriority())
383
384
385
386
387
388
389
390
391
392
393
394
395
396
      elif fname == "ops":
        row.append([op.input.__getstate__() for op in self.ops])
      elif fname == "opresult":
        row.append([op.result for op in self.ops])
      elif fname == "opstatus":
        row.append([op.status for op in self.ops])
      elif fname == "oplog":
        row.append([op.log for op in self.ops])
      elif fname == "opstart":
        row.append([op.start_timestamp for op in self.ops])
      elif fname == "opexec":
        row.append([op.exec_timestamp for op in self.ops])
      elif fname == "opend":
        row.append([op.end_timestamp for op in self.ops])
397
398
      elif fname == "oppriority":
        row.append([op.priority for op in self.ops])
399
400
401
402
403
404
405
406
407
408
409
410
      elif fname == "received_ts":
        row.append(self.received_timestamp)
      elif fname == "start_ts":
        row.append(self.start_timestamp)
      elif fname == "end_ts":
        row.append(self.end_timestamp)
      elif fname == "summary":
        row.append([op.input.Summary() for op in self.ops])
      else:
        raise errors.OpExecError("Invalid self query field '%s'" % fname)
    return row

411
412
413
414
415
416
417
418
419
420
421
  def MarkUnfinishedOps(self, status, result):
    """Mark unfinished opcodes with a given status and result.

    This is an utility function for marking all running or waiting to
    be run opcodes with a given status. Opcodes which are already
    finalised are not changed.

    @param status: a given opcode status
    @param result: the opcode result

    """
422
423
424
425
426
427
428
429
    not_marked = True
    for op in self.ops:
      if op.status in constants.OPS_FINALIZED:
        assert not_marked, "Finalized opcodes found after non-finalized ones"
        continue
      op.status = status
      op.result = result
      not_marked = False
430

431
432
433
434
435
436
  def Finalize(self):
    """Marks the job as finalized.

    """
    self.end_timestamp = TimeStampNow()

437
  def Cancel(self):
438
439
440
441
442
443
444
    """Marks job as canceled/-ing if possible.

    @rtype: tuple; (bool, string)
    @return: Boolean describing whether job was successfully canceled or marked
      as canceling and a text message

    """
445
446
447
448
449
    status = self.CalcStatus()

    if status == constants.JOB_STATUS_QUEUED:
      self.MarkUnfinishedOps(constants.OP_STATUS_CANCELED,
                             "Job canceled by request")
450
      self.Finalize()
451
      return (True, "Job %s canceled" % self.id)
452
453
454
455

    elif status == constants.JOB_STATUS_WAITLOCK:
      # The worker will notice the new status and cancel the job
      self.MarkUnfinishedOps(constants.OP_STATUS_CANCELING, None)
456
      return (True, "Job %s will be canceled" % self.id)
457

458
459
460
    else:
      logging.debug("Job %s is no longer waiting in the queue", self.id)
      return (False, "Job %s is no longer waiting in the queue" % self.id)
461

462

463
class _OpExecCallbacks(mcpu.OpExecCbBase):
464
465
  def __init__(self, queue, job, op):
    """Initializes this class.
466

467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
    @type queue: L{JobQueue}
    @param queue: Job queue
    @type job: L{_QueuedJob}
    @param job: Job object
    @type op: L{_QueuedOpCode}
    @param op: OpCode

    """
    assert queue, "Queue is missing"
    assert job, "Job is missing"
    assert op, "Opcode is missing"

    self._queue = queue
    self._job = job
    self._op = op

483
484
485
486
487
488
489
490
491
  def _CheckCancel(self):
    """Raises an exception to cancel the job if asked to.

    """
    # Cancel here if we were asked to
    if self._op.status == constants.OP_STATUS_CANCELING:
      logging.debug("Canceling opcode")
      raise CancelJob()

492
  @locking.ssynchronized(_QUEUE, shared=1)
493
  def NotifyStart(self):
Iustin Pop's avatar
Iustin Pop committed
494
495
    """Mark the opcode as running, not lock-waiting.

496
497
498
499
    This is called from the mcpu code as a notifier function, when the LU is
    finally about to start the Exec() method. Of course, to have end-user
    visible results, the opcode must be initially (before calling into
    Processor.ExecOpCode) set to OP_STATUS_WAITLOCK.
Iustin Pop's avatar
Iustin Pop committed
500
501

    """
502
    assert self._op in self._job.ops
503
504
    assert self._op.status in (constants.OP_STATUS_WAITLOCK,
                               constants.OP_STATUS_CANCELING)
505

506
    # Cancel here if we were asked to
507
    self._CheckCancel()
508

509
    logging.debug("Opcode is now running")
510

511
512
513
514
515
    self._op.status = constants.OP_STATUS_RUNNING
    self._op.exec_timestamp = TimeStampNow()

    # And finally replicate the job status
    self._queue.UpdateJobUnlocked(self._job)
516

517
  @locking.ssynchronized(_QUEUE, shared=1)
518
519
520
521
522
523
524
525
  def _AppendFeedback(self, timestamp, log_type, log_msg):
    """Internal feedback append function, with locks

    """
    self._job.log_serial += 1
    self._op.log.append((self._job.log_serial, timestamp, log_type, log_msg))
    self._queue.UpdateJobUnlocked(self._job, replicate=False)

526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
  def Feedback(self, *args):
    """Append a log entry.

    """
    assert len(args) < 3

    if len(args) == 1:
      log_type = constants.ELOG_MESSAGE
      log_msg = args[0]
    else:
      (log_type, log_msg) = args

    # The time is split to make serialization easier and not lose
    # precision.
    timestamp = utils.SplitTime(time.time())
541
    self._AppendFeedback(timestamp, log_type, log_msg)
542

543
544
  def CheckCancel(self):
    """Check whether job has been cancelled.
545
546

    """
547
548
549
550
551
552
    assert self._op.status in (constants.OP_STATUS_WAITLOCK,
                               constants.OP_STATUS_CANCELING)

    # Cancel here if we were asked to
    self._CheckCancel()

553
554
555
556
557
558
559
560
561
  def SubmitManyJobs(self, jobs):
    """Submits jobs for processing.

    See L{JobQueue.SubmitManyJobs}.

    """
    # Locking is done in job queue
    return self._queue.SubmitManyJobs(jobs)

562

563
564
565
class _JobChangesChecker(object):
  def __init__(self, fields, prev_job_info, prev_log_serial):
    """Initializes this class.
Guido Trotter's avatar
Guido Trotter committed
566

567
568
569
570
571
572
    @type fields: list of strings
    @param fields: Fields requested by LUXI client
    @type prev_job_info: string
    @param prev_job_info: previous job info, as passed by the LUXI client
    @type prev_log_serial: string
    @param prev_log_serial: previous job serial, as passed by the LUXI client
Guido Trotter's avatar
Guido Trotter committed
573

574
575
576
577
    """
    self._fields = fields
    self._prev_job_info = prev_job_info
    self._prev_log_serial = prev_log_serial
Guido Trotter's avatar
Guido Trotter committed
578

579
580
  def __call__(self, job):
    """Checks whether job has changed.
Guido Trotter's avatar
Guido Trotter committed
581

582
583
    @type job: L{_QueuedJob}
    @param job: Job object
Guido Trotter's avatar
Guido Trotter committed
584
585

    """
586
587
588
    status = job.CalcStatus()
    job_info = job.GetInfo(self._fields)
    log_entries = job.GetLogEntries(self._prev_log_serial)
Guido Trotter's avatar
Guido Trotter committed
589
590
591
592
593
594
595
596

    # Serializing and deserializing data can cause type changes (e.g. from
    # tuple to list) or precision loss. We're doing it here so that we get
    # the same modifications as the data received from the client. Without
    # this, the comparison afterwards might fail without the data being
    # significantly different.
    # TODO: we just deserialized from disk, investigate how to make sure that
    # the job info and log entries are compatible to avoid this further step.
597
598
599
600
    # TODO: Doing something like in testutils.py:UnifyValueType might be more
    # efficient, though floats will be tricky
    job_info = serializer.LoadJson(serializer.DumpJson(job_info))
    log_entries = serializer.LoadJson(serializer.DumpJson(log_entries))
Guido Trotter's avatar
Guido Trotter committed
601
602
603

    # Don't even try to wait if the job is no longer running, there will be
    # no changes.
604
605
606
607
608
609
610
    if (status not in (constants.JOB_STATUS_QUEUED,
                       constants.JOB_STATUS_RUNNING,
                       constants.JOB_STATUS_WAITLOCK) or
        job_info != self._prev_job_info or
        (log_entries and self._prev_log_serial != log_entries[0][0])):
      logging.debug("Job %s changed", job.id)
      return (job_info, log_entries)
Guido Trotter's avatar
Guido Trotter committed
611

612
613
614
615
616
617
618
619
620
621
    return None


class _JobFileChangesWaiter(object):
  def __init__(self, filename):
    """Initializes this class.

    @type filename: string
    @param filename: Path to job file
    @raises errors.InotifyError: if the notifier cannot be setup
Guido Trotter's avatar
Guido Trotter committed
622

623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
    """
    self._wm = pyinotify.WatchManager()
    self._inotify_handler = \
      asyncnotifier.SingleFileEventHandler(self._wm, self._OnInotify, filename)
    self._notifier = \
      pyinotify.Notifier(self._wm, default_proc_fun=self._inotify_handler)
    try:
      self._inotify_handler.enable()
    except Exception:
      # pyinotify doesn't close file descriptors automatically
      self._notifier.stop()
      raise

  def _OnInotify(self, notifier_enabled):
    """Callback for inotify.

    """
Guido Trotter's avatar
Guido Trotter committed
640
    if not notifier_enabled:
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
      self._inotify_handler.enable()

  def Wait(self, timeout):
    """Waits for the job file to change.

    @type timeout: float
    @param timeout: Timeout in seconds
    @return: Whether there have been events

    """
    assert timeout >= 0
    have_events = self._notifier.check_events(timeout * 1000)
    if have_events:
      self._notifier.read_events()
    self._notifier.process_events()
    return have_events

  def Close(self):
    """Closes underlying notifier and its file descriptor.

    """
    self._notifier.stop()


class _JobChangesWaiter(object):
  def __init__(self, filename):
    """Initializes this class.

    @type filename: string
    @param filename: Path to job file

    """
    self._filewaiter = None
    self._filename = filename
Guido Trotter's avatar
Guido Trotter committed
675

676
677
  def Wait(self, timeout):
    """Waits for a job to change.
Guido Trotter's avatar
Guido Trotter committed
678

679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
    @type timeout: float
    @param timeout: Timeout in seconds
    @return: Whether there have been events

    """
    if self._filewaiter:
      return self._filewaiter.Wait(timeout)

    # Lazy setup: Avoid inotify setup cost when job file has already changed.
    # If this point is reached, return immediately and let caller check the job
    # file again in case there were changes since the last check. This avoids a
    # race condition.
    self._filewaiter = _JobFileChangesWaiter(self._filename)

    return True

  def Close(self):
    """Closes underlying waiter.

    """
    if self._filewaiter:
      self._filewaiter.Close()


class _WaitForJobChangesHelper(object):
  """Helper class using inotify to wait for changes in a job file.

  This class takes a previous job status and serial, and alerts the client when
  the current job status has changed.

  """
  @staticmethod
  def _CheckForChanges(job_load_fn, check_fn):
    job = job_load_fn()
    if not job:
      raise errors.JobLost()

    result = check_fn(job)
    if result is None:
      raise utils.RetryAgain()

    return result

  def __call__(self, filename, job_load_fn,
               fields, prev_job_info, prev_log_serial, timeout):
    """Waits for changes on a job.

    @type filename: string
    @param filename: File on which to wait for changes
    @type job_load_fn: callable
    @param job_load_fn: Function to load job
    @type fields: list of strings
    @param fields: Which fields to check for changes
    @type prev_job_info: list or None
    @param prev_job_info: Last job information returned
    @type prev_log_serial: int
    @param prev_log_serial: Last job message serial number
    @type timeout: float
    @param timeout: maximum time to wait in seconds

    """
Guido Trotter's avatar
Guido Trotter committed
740
    try:
741
742
743
744
745
746
747
748
749
      check_fn = _JobChangesChecker(fields, prev_job_info, prev_log_serial)
      waiter = _JobChangesWaiter(filename)
      try:
        return utils.Retry(compat.partial(self._CheckForChanges,
                                          job_load_fn, check_fn),
                           utils.RETRY_REMAINING_TIME, timeout,
                           wait_fn=waiter.Wait)
      finally:
        waiter.Close()
Guido Trotter's avatar
Guido Trotter committed
750
751
752
753
754
755
    except (errors.InotifyError, errors.JobLost):
      return None
    except utils.RetryTimeout:
      return constants.JOB_NOTCHANGED


756
757
758
759
760
761
762
763
764
765
766
767
def _EncodeOpError(err):
  """Encodes an error which occurred while processing an opcode.

  """
  if isinstance(err, errors.GenericError):
    to_encode = err
  else:
    to_encode = errors.OpExecError(str(err))

  return errors.EncodeException(to_encode)


768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
class _TimeoutStrategyWrapper:
  def __init__(self, fn):
    """Initializes this class.

    """
    self._fn = fn
    self._next = None

  def _Advance(self):
    """Gets the next timeout if necessary.

    """
    if self._next is None:
      self._next = self._fn()

  def Peek(self):
    """Returns the next timeout.

    """
    self._Advance()
    return self._next

  def Next(self):
    """Returns the current timeout and advances the internal state.

    """
    self._Advance()
    result = self._next
    self._next = None
    return result


800
class _OpExecContext:
801
  def __init__(self, op, index, log_prefix, timeout_strategy_factory):
802
803
804
805
806
807
808
809
    """Initializes this class.

    """
    self.op = op
    self.index = index
    self.log_prefix = log_prefix
    self.summary = op.input.Summary()

Michael Hanselmann's avatar
Michael Hanselmann committed
810
811
812
813
814
815
    # Create local copy to modify
    if getattr(op.input, opcodes.DEPEND_ATTR, None):
      self.jobdeps = op.input.depends[:]
    else:
      self.jobdeps = None

816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
    self._timeout_strategy_factory = timeout_strategy_factory
    self._ResetTimeoutStrategy()

  def _ResetTimeoutStrategy(self):
    """Creates a new timeout strategy.

    """
    self._timeout_strategy = \
      _TimeoutStrategyWrapper(self._timeout_strategy_factory().NextAttempt)

  def CheckPriorityIncrease(self):
    """Checks whether priority can and should be increased.

    Called when locks couldn't be acquired.

    """
    op = self.op

    # Exhausted all retries and next round should not use blocking acquire
    # for locks?
    if (self._timeout_strategy.Peek() is None and
        op.priority > constants.OP_PRIO_HIGHEST):
      logging.debug("Increasing priority")
      op.priority -= 1
      self._ResetTimeoutStrategy()
      return True

    return False

  def GetNextLockTimeout(self):
    """Returns the next lock acquire timeout.

    """
    return self._timeout_strategy.Next()

851

852
class _JobProcessor(object):
853
854
  def __init__(self, queue, opexec_fn, job,
               _timeout_strategy_factory=mcpu.LockAttemptTimeoutStrategy):
855
856
857
858
859
860
    """Initializes this class.

    """
    self.queue = queue
    self.opexec_fn = opexec_fn
    self.job = job
861
    self._timeout_strategy_factory = _timeout_strategy_factory
862
863

  @staticmethod
864
  def _FindNextOpcode(job, timeout_strategy_factory):
865
866
867
868
    """Locates the next opcode to run.

    @type job: L{_QueuedJob}
    @param job: Job object
869
    @param timeout_strategy_factory: Callable to create new timeout strategy
870
871
872
873
874
875

    """
    # Create some sort of a cache to speed up locating next opcode for future
    # lookups
    # TODO: Consider splitting _QueuedJob.ops into two separate lists, one for
    # pending and one for processed ops.
876
877
    if job.ops_iter is None:
      job.ops_iter = enumerate(job.ops)
878
879
880
881

    # Find next opcode to run
    while True:
      try:
882
        (idx, op) = job.ops_iter.next()
883
884
885
886
887
888
889
      except StopIteration:
        raise errors.ProgrammerError("Called for a finished job")

      if op.status == constants.OP_STATUS_RUNNING:
        # Found an opcode already marked as running
        raise errors.ProgrammerError("Called for job marked as running")

890
891
      opctx = _OpExecContext(op, idx, "Op %s/%s" % (idx + 1, len(job.ops)),
                             timeout_strategy_factory)
892

893
894
      if op.status not in constants.OPS_FINALIZED:
        return opctx
895

896
897
898
899
900
901
      # This is a job that was partially completed before master daemon
      # shutdown, so it can be expected that some opcodes are already
      # completed successfully (if any did error out, then the whole job
      # should have been aborted and not resubmitted for processing).
      logging.info("%s: opcode %s already processed, skipping",
                   opctx.log_prefix, opctx.summary)
902
903
904
905
906
907
908
909
910

  @staticmethod
  def _MarkWaitlock(job, op):
    """Marks an opcode as waiting for locks.

    The job's start timestamp is also set if necessary.

    @type job: L{_QueuedJob}
    @param job: Job object
911
912
    @type op: L{_QueuedOpCode}
    @param op: Opcode object
913
914
915

    """
    assert op in job.ops
916
917
918
919
    assert op.status in (constants.OP_STATUS_QUEUED,
                         constants.OP_STATUS_WAITLOCK)

    update = False
920
921

    op.result = None
922
923
924
925
926
927
928
929

    if op.status == constants.OP_STATUS_QUEUED:
      op.status = constants.OP_STATUS_WAITLOCK
      update = True

    if op.start_timestamp is None:
      op.start_timestamp = TimeStampNow()
      update = True
930
931
932

    if job.start_timestamp is None:
      job.start_timestamp = op.start_timestamp
933
934
935
936
937
      update = True

    assert op.status == constants.OP_STATUS_WAITLOCK

    return update
938

Michael Hanselmann's avatar
Michael Hanselmann committed
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
  @staticmethod
  def _CheckDependencies(queue, job, opctx):
    """Checks if an opcode has dependencies and if so, processes them.

    @type queue: L{JobQueue}
    @param queue: Queue object
    @type job: L{_QueuedJob}
    @param job: Job object
    @type opctx: L{_OpExecContext}
    @param opctx: Opcode execution context
    @rtype: bool
    @return: Whether opcode will be re-scheduled by dependency tracker

    """
    op = opctx.op

    result = False

    while opctx.jobdeps:
      (dep_job_id, dep_status) = opctx.jobdeps[0]

      (depresult, depmsg) = queue.depmgr.CheckAndRegister(job, dep_job_id,
                                                          dep_status)
      assert ht.TNonEmptyString(depmsg), "No dependency message"

      logging.info("%s: %s", opctx.log_prefix, depmsg)

      if depresult == _JobDependencyManager.CONTINUE:
        # Remove dependency and continue
        opctx.jobdeps.pop(0)

      elif depresult == _JobDependencyManager.WAIT:
        # Need to wait for notification, dependency tracker will re-add job
        # to workerpool
        result = True
        break

      elif depresult == _JobDependencyManager.CANCEL:
        # Job was cancelled, cancel this job as well
        job.Cancel()
        assert op.status == constants.OP_STATUS_CANCELING
        break

      elif depresult in (_JobDependencyManager.WRONGSTATUS,
                         _JobDependencyManager.ERROR):
        # Job failed or there was an error, this job must fail
        op.status = constants.OP_STATUS_ERROR
        op.result = _EncodeOpError(errors.OpExecError(depmsg))
        break

      else:
        raise errors.ProgrammerError("Unknown dependency result '%s'" %
                                     depresult)

    return result

995
  def _ExecOpCodeUnlocked(self, opctx):
996
997
998
    """Processes one opcode and returns the result.

    """
999
1000
    op = opctx.op

1001
1002
    assert op.status == constants.OP_STATUS_WAITLOCK

1003
1004
    timeout = opctx.GetNextLockTimeout()

1005
1006
1007
    try:
      # Make sure not to hold queue lock while calling ExecOpCode
      result = self.opexec_fn(op.input,
1008
                              _OpExecCallbacks(self.queue, self.job, op),
1009
                              timeout=timeout, priority=op.priority)
1010
1011
1012
    except mcpu.LockAcquireTimeout:
      assert timeout is not None, "Received timeout for blocking acquire"
      logging.debug("Couldn't acquire locks in %0.6fs", timeout)
1013
1014
1015
1016
1017
1018
1019
1020

      assert op.status in (constants.OP_STATUS_WAITLOCK,
                           constants.OP_STATUS_CANCELING)

      # Was job cancelled while we were waiting for the lock?
      if op.status == constants.OP_STATUS_CANCELING:
        return (constants.OP_STATUS_CANCELING, None)

1021
1022
      # Stay in waitlock while trying to re-acquire lock
      return (constants.OP_STATUS_WAITLOCK, None)
1023
    except CancelJob:
1024
      logging.exception("%s: Canceling job", opctx.log_prefix)
1025
1026
1027
      assert op.status == constants.OP_STATUS_CANCELING
      return (constants.OP_STATUS_CANCELING, None)
    except Exception, err: # pylint: disable-msg=W0703
1028
1029
      logging.exception("%s: Caught exception in %s",
                        opctx.log_prefix, opctx.summary)
1030
1031
      return (constants.OP_STATUS_ERROR, _EncodeOpError(err))
    else:
1032
1033
      logging.debug("%s: %s successful",
                    opctx.log_prefix, opctx.summary)
1034
1035
      return (constants.OP_STATUS_SUCCESS, result)

1036
  def __call__(self, _nextop_fn=None):
1037
1038
    """Continues execution of a job.

1039
    @param _nextop_fn: Callback function for tests
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
    @rtype: bool
    @return: True if job is finished, False if processor needs to be called
             again

    """
    queue = self.queue
    job = self.job

    logging.debug("Processing job %s", job.id)

    queue.acquire(shared=1)
    try:
      opcount = len(job.ops)

1054
1055
1056
1057
      # Don't do anything for finalized jobs
      if job.CalcStatus() in constants.JOBS_FINALIZED:
        return True

1058
1059
1060
      # Is a previous opcode still pending?
      if job.cur_opctx:
        opctx = job.cur_opctx
1061
        job.cur_opctx = None
1062
1063
1064
1065
1066
      else:
        if __debug__ and _nextop_fn:
          _nextop_fn()
        opctx = self._FindNextOpcode(job, self._timeout_strategy_factory)

1067
      op = opctx.op
1068
1069
1070

      # Consistency check
      assert compat.all(i.status in (constants.OP_STATUS_QUEUED,
1071
                                     constants.OP_STATUS_CANCELING)
1072
                        for i in job.ops[opctx.index + 1:])
1073
1074
1075

      assert op.status in (constants.OP_STATUS_QUEUED,
                           constants.OP_STATUS_WAITLOCK,
1076
                           constants.OP_STATUS_CANCELING)
1077

1078
1079
1080
      assert (op.priority <= constants.OP_PRIO_LOWEST and
              op.priority >= constants.OP_PRIO_HIGHEST)

Michael Hanselmann's avatar
Michael Hanselmann committed
1081
1082
      waitjob = None

1083
      if op.status != constants.OP_STATUS_CANCELING:
1084
1085
1086
        assert op.status in (constants.OP_STATUS_QUEUED,
                             constants.OP_STATUS_WAITLOCK)

1087
        # Prepare to start opcode
1088
1089
1090
        if self._MarkWaitlock(job, op):
          # Write to disk
          queue.UpdateJobUnlocked(job)
1091
1092
1093

        assert op.status == constants.OP_STATUS_WAITLOCK
        assert job.CalcStatus() == constants.JOB_STATUS_WAITLOCK
1094
        assert job.start_timestamp and op.start_timestamp
Michael Hanselmann's avatar
Michael Hanselmann committed
1095
1096
1097
1098
        assert waitjob is None

        # Check if waiting for a job is necessary
        waitjob = self._CheckDependencies(queue, job, opctx)
1099

Michael Hanselmann's avatar
Michael Hanselmann committed
1100
1101
1102
        assert op.status in (constants.OP_STATUS_WAITLOCK,
                             constants.OP_STATUS_CANCELING,
                             constants.OP_STATUS_ERROR)
1103

Michael Hanselmann's avatar
Michael Hanselmann committed
1104
1105
1106
1107
        if not (waitjob or op.status in (constants.OP_STATUS_CANCELING,
                                         constants.OP_STATUS_ERROR)):
          logging.info("%s: opcode %s waiting for locks",
                       opctx.log_prefix, opctx.summary)
1108

Michael Hanselmann's avatar
Michael Hanselmann committed
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
          assert not opctx.jobdeps, "Not all dependencies were removed"

          queue.release()
          try:
            (op_status, op_result) = self._ExecOpCodeUnlocked(opctx)
          finally:
            queue.acquire(shared=1)

          op.status = op_status
          op.result = op_result

          assert not waitjob
1121

1122
        if op.status == constants.OP_STATUS_WAITLOCK:
1123
1124
          # Couldn't get locks in time
          assert not op.end_timestamp
1125
        else:
1126
1127
          # Finalize opcode
          op.end_timestamp = TimeStampNow()
1128

1129
1130
1131
1132
1133
          if op.status == constants.OP_STATUS_CANCELING:
            assert not compat.any(i.status != constants.OP_STATUS_CANCELING
                                  for i in job.ops[opctx.index:])
          else:
            assert op.status in constants.OPS_FINALIZED
1134

Michael Hanselmann's avatar
Michael Hanselmann committed
1135
      if op.status == constants.OP_STATUS_WAITLOCK or waitjob:
1136
1137
        finalize = False

Michael Hanselmann's avatar
Michael Hanselmann committed
1138
        if not waitjob and opctx.CheckPriorityIncrease():
1139
1140
          # Priority was changed, need to update on-disk file
          queue.UpdateJobUnlocked(job)
1141

1142
1143
        # Keep around for another round
        job.cur_opctx = opctx
1144

1145
1146
        assert (op.priority <= constants.OP_PRIO_LOWEST and
                op.priority >= constants.OP_PRIO_HIGHEST)
1147

1148
        # In no case must the status be finalized here
1149
        assert job.CalcStatus() == constants.JOB_STATUS_WAITLOCK
1150
1151

      else:
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
        # Ensure all opcodes so far have been successful
        assert (opctx.index == 0 or
                compat.all(i.status == constants.OP_STATUS_SUCCESS
                           for i in job.ops[:opctx.index]))

        # Reset context
        job.cur_opctx = None

        if op.status == constants.OP_STATUS_SUCCESS:
          finalize = False

        elif op.status == constants.OP_STATUS_ERROR:
          # Ensure failed opcode has an exception as its result
          assert errors.GetEncodedError(job.ops[opctx.index].result)

          to_encode = errors.OpExecError("Preceding opcode failed")
          job.MarkUnfinishedOps(constants.OP_STATUS_ERROR,
                                _EncodeOpError(to_encode))
          finalize = True
1171

1172
1173
1174
1175
          # Consistency check
          assert compat.all(i.status == constants.OP_STATUS_ERROR and
                            errors.GetEncodedError(i.result)
                            for i in job.ops[opctx.index:])
1176

1177
1178
1179
1180
1181
1182
1183
1184
        elif op.status == constants.OP_STATUS_CANCELING:
          job.MarkUnfinishedOps(constants.OP_STATUS_CANCELED,
                                "Job canceled by request")
          finalize = True

        else:
          raise errors.ProgrammerError("Unknown status '%s'" % op.status)

1185
1186
1187
1188
1189
        if opctx.index == (opcount - 1):
          # Finalize on last opcode
          finalize = True

        if finalize:
1190
          # All opcodes have been run, finalize job
1191
          job.Finalize()
1192
1193
1194
1195

        # Write to disk. If the job status is final, this is the final write
        # allowed. Once the file has been written, it can be archived anytime.
        queue.UpdateJobUnlocked(job)
1196

Michael Hanselmann's avatar
Michael Hanselmann committed
1197
1198
        assert not waitjob

1199
        if finalize:
1200
          logging.info("Finished job %s, status = %s", job.id, job.CalcStatus())
Michael Hanselmann's avatar
Michael Hanselmann committed
1201
1202
          # TODO: Check locking
          queue.depmgr.NotifyWaiters(job.id)
1203
          return True
1204

Michael Hanselmann's avatar
Michael Hanselmann committed
1205
1206
1207
      assert not waitjob or queue.depmgr.JobWaiting(job)

      return bool(waitjob)
1208
1209
1210
1211
    finally:
      queue.release()


1212
1213
1214
1215
class _JobQueueWorker(workerpool.BaseWorker):
  """The actual job workers.

  """
Iustin Pop's avatar
Iustin Pop committed
1216
  def RunTask(self, job): # pylint: disable-msg=W0221
Michael Hanselmann's avatar
Michael Hanselmann committed
1217
1218
    """Job executor.

1219
1220
1221
    @type job: L{_QueuedJob}
    @param job: the job to be processed

Michael Hanselmann's avatar
Michael Hanselmann committed
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
    """
    # Ensure only one worker is active on a single job. If a job registers for
    # a dependency job, and the other job notifies before the first worker is
    # done, the job can end up in the tasklist more than once.
    job.processor_lock.acquire()
    try:
      return self._RunTaskInner(job)
    finally:
      job.processor_lock.release()

  def _RunTaskInner(self, job):
    """Executes a job.

    Must be called with per-job lock acquired.

Michael Hanselmann's avatar
Michael Hanselmann committed
1237
    """
1238
1239
1240
    queue = job.queue
    assert queue == self.pool.queue

1241
1242
    setname_fn = lambda op: self.SetTaskName(self._GetWorkerName(job, op))
    setname_fn(None)
1243

1244
1245
    proc = mcpu.Processor(queue.context, job.id)

1246
1247
1248
1249
1250
    # Create wrapper for setting thread name
    wrap_execop_fn = compat.partial(self._WrapExecOpCode, setname_fn,
                                    proc.ExecOpCode)

    if not _JobProcessor(queue, wrap_execop_fn, job)():
1251
      # Schedule again
1252
      raise workerpool.DeferTask(priority=job.CalcPriority())
Michael Hanselmann's avatar
Michael Hanselmann committed
1253

1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
  @staticmethod
  def _WrapExecOpCode(setname_fn, execop_fn, op, *args, **kwargs):
    """Updates the worker thread name to include a short summary of the opcode.

    @param setname_fn: Callable setting worker thread name
    @param execop_fn: Callable for executing opcode (usually
                      L{mcpu.Processor.ExecOpCode})

    """
    setname_fn(op)
    try:
      return execop_fn(op, *args, **kwargs)
    finally:
      setname_fn(None)

  @staticmethod
  def _GetWorkerName(job, op):
    """Sets the worker thread name.

    @type job: L{_QueuedJob}
    @type op: L{opcodes.OpCode}

    """
    parts = ["Job%s" % job.id]

    if op:
      parts.append(op.TinySummary())

    return "/".join(parts)

Michael Hanselmann's avatar
Michael Hanselmann committed
1284
1285

class _JobQueueWorkerPool(workerpool.WorkerPool):
1286
1287
1288
  """Simple class implementing a job-processing workerpool.

  """
1289
  def __init__(self, queue):
1290
    super(_JobQueueWorkerPool, self).__init__("Jq",
1291
                                              JOBQUEUE_THREADS,
Michael Hanselmann's avatar
Michael Hanselmann committed
1292
                                              _JobQueueWorker)
1293
    self.queue = queue
Michael Hanselmann's avatar
Michael Hanselmann committed
1294
1295


Michael Hanselmann's avatar
Michael Hanselmann committed
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
class _JobDependencyManager:
  """Keeps track of job dependencies.

  """
  (WAIT,
   ERROR,
   CANCEL,
   CONTINUE,
   WRONGSTATUS) = range(1, 6)

  # TODO: Export waiter information to lock monitor

  def __init__(self, getstatus_fn, enqueue_fn):
    """Initializes this class.

    """
    self._getstatus_fn = getstatus_fn
    self._enqueue_fn = enqueue_fn

    self._waiters = {}
    self._lock = locking.SharedLock(