4 # Copyright (C) 2006, 2007, 2008 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 job queue handling.
24 Locking: there's a single, large lock in the L{JobQueue} class. It's
25 used by all other classes in this module.
27 @var JOBQUEUE_THREADS: the number of worker threads we start for
40 from ganeti import constants
41 from ganeti import serializer
42 from ganeti import workerpool
43 from ganeti import opcodes
44 from ganeti import errors
45 from ganeti import mcpu
46 from ganeti import utils
47 from ganeti import jstore
48 from ganeti import rpc
52 JOBS_PER_ARCHIVE_DIRECTORY = 10000
55 class CancelJob(Exception):
56 """Special exception to cancel a job.
62 """Returns the current timestamp.
65 @return: the current time in the (seconds, microseconds) format
68 return utils.SplitTime(time.time())
71 class _QueuedOpCode(object):
72 """Encapsulates an opcode object.
74 @ivar log: holds the execution log and consists of tuples
75 of the form C{(log_serial, timestamp, level, message)}
76 @ivar input: the OpCode we encapsulate
77 @ivar status: the current status
78 @ivar result: the result of the LU execution
79 @ivar start_timestamp: timestamp for the start of the execution
80 @ivar stop_timestamp: timestamp for the end of the execution
83 __slots__ = ["input", "status", "result", "log",
84 "start_timestamp", "end_timestamp",
87 def __init__(self, op):
88 """Constructor for the _QuededOpCode.
90 @type op: L{opcodes.OpCode}
91 @param op: the opcode we encapsulate
95 self.status = constants.OP_STATUS_QUEUED
98 self.start_timestamp = None
99 self.end_timestamp = None
102 def Restore(cls, state):
103 """Restore the _QueuedOpCode from the serialized form.
106 @param state: the serialized state
107 @rtype: _QueuedOpCode
108 @return: a new _QueuedOpCode instance
111 obj = _QueuedOpCode.__new__(cls)
112 obj.input = opcodes.OpCode.LoadOpCode(state["input"])
113 obj.status = state["status"]
114 obj.result = state["result"]
115 obj.log = state["log"]
116 obj.start_timestamp = state.get("start_timestamp", None)
117 obj.end_timestamp = state.get("end_timestamp", None)
121 """Serializes this _QueuedOpCode.
124 @return: the dictionary holding the serialized state
128 "input": self.input.__getstate__(),
129 "status": self.status,
130 "result": self.result,
132 "start_timestamp": self.start_timestamp,
133 "end_timestamp": self.end_timestamp,
137 class _QueuedJob(object):
138 """In-memory job representation.
140 This is what we use to track the user-submitted jobs. Locking must
141 be taken care of by users of this class.
143 @type queue: L{JobQueue}
144 @ivar queue: the parent queue
147 @ivar ops: the list of _QueuedOpCode that constitute the job
148 @type run_op_index: int
149 @ivar run_op_index: the currently executing opcode, or -1 if
150 we didn't yet start executing
151 @type log_serial: int
152 @ivar log_serial: holds the index for the next log entry
153 @ivar received_timestamp: the timestamp for when the job was received
154 @ivar start_timestmap: the timestamp for start of execution
155 @ivar end_timestamp: the timestamp for end of execution
156 @ivar change: a Condition variable we use for waiting for job changes
159 __slots__ = ["queue", "id", "ops", "run_op_index", "log_serial",
160 "received_timestamp", "start_timestamp", "end_timestamp",
164 def __init__(self, queue, job_id, ops):
165 """Constructor for the _QueuedJob.
167 @type queue: L{JobQueue}
168 @param queue: our parent queue
170 @param job_id: our job id
172 @param ops: the list of opcodes we hold, which will be encapsulated
177 # TODO: use a better exception
178 raise Exception("No opcodes")
182 self.ops = [_QueuedOpCode(op) for op in ops]
183 self.run_op_index = -1
185 self.received_timestamp = TimeStampNow()
186 self.start_timestamp = None
187 self.end_timestamp = None
189 # Condition to wait for changes
190 self.change = threading.Condition(self.queue._lock)
193 def Restore(cls, queue, state):
194 """Restore a _QueuedJob from serialized state:
196 @type queue: L{JobQueue}
197 @param queue: to which queue the restored job belongs
199 @param state: the serialized state
201 @return: the restored _JobQueue instance
204 obj = _QueuedJob.__new__(cls)
207 obj.run_op_index = state["run_op_index"]
208 obj.received_timestamp = state.get("received_timestamp", None)
209 obj.start_timestamp = state.get("start_timestamp", None)
210 obj.end_timestamp = state.get("end_timestamp", None)
214 for op_state in state["ops"]:
215 op = _QueuedOpCode.Restore(op_state)
216 for log_entry in op.log:
217 obj.log_serial = max(obj.log_serial, log_entry[0])
220 # Condition to wait for changes
221 obj.change = threading.Condition(obj.queue._lock)
226 """Serialize the _JobQueue instance.
229 @return: the serialized state
234 "ops": [op.Serialize() for op in self.ops],
235 "run_op_index": self.run_op_index,
236 "start_timestamp": self.start_timestamp,
237 "end_timestamp": self.end_timestamp,
238 "received_timestamp": self.received_timestamp,
241 def CalcStatus(self):
242 """Compute the status of this job.
244 This function iterates over all the _QueuedOpCodes in the job and
245 based on their status, computes the job status.
248 - if we find a cancelled, or finished with error, the job
249 status will be the same
250 - otherwise, the last opcode with the status one of:
255 will determine the job status
257 - otherwise, it means either all opcodes are queued, or success,
258 and the job status will be the same
260 @return: the job status
263 status = constants.JOB_STATUS_QUEUED
267 if op.status == constants.OP_STATUS_SUCCESS:
272 if op.status == constants.OP_STATUS_QUEUED:
274 elif op.status == constants.OP_STATUS_WAITLOCK:
275 status = constants.JOB_STATUS_WAITLOCK
276 elif op.status == constants.OP_STATUS_RUNNING:
277 status = constants.JOB_STATUS_RUNNING
278 elif op.status == constants.OP_STATUS_CANCELING:
279 status = constants.JOB_STATUS_CANCELING
281 elif op.status == constants.OP_STATUS_ERROR:
282 status = constants.JOB_STATUS_ERROR
283 # The whole job fails if one opcode failed
285 elif op.status == constants.OP_STATUS_CANCELED:
286 status = constants.OP_STATUS_CANCELED
290 status = constants.JOB_STATUS_SUCCESS
294 def GetLogEntries(self, newer_than):
295 """Selectively returns the log entries.
297 @type newer_than: None or int
298 @param newer_than: if this is None, return all log entries,
299 otherwise return only the log entries with serial higher
302 @return: the list of the log entries selected
305 if newer_than is None:
312 entries.extend(filter(lambda entry: entry[0] > serial, op.log))
316 def MarkUnfinishedOps(self, status, result):
317 """Mark unfinished opcodes with a given status and result.
319 This is an utility function for marking all running or waiting to
320 be run opcodes with a given status. Opcodes which are already
321 finalised are not changed.
323 @param status: a given opcode status
324 @param result: the opcode result
329 if op.status in constants.OPS_FINALIZED:
330 assert not_marked, "Finalized opcodes found after non-finalized ones"
337 class _JobQueueWorker(workerpool.BaseWorker):
338 """The actual job workers.
341 def _NotifyStart(self):
342 """Mark the opcode as running, not lock-waiting.
344 This is called from the mcpu code as a notifier function, when the
345 LU is finally about to start the Exec() method. Of course, to have
346 end-user visible results, the opcode must be initially (before
347 calling into Processor.ExecOpCode) set to OP_STATUS_WAITLOCK.
350 assert self.queue, "Queue attribute is missing"
351 assert self.opcode, "Opcode attribute is missing"
355 assert self.opcode.status in (constants.OP_STATUS_WAITLOCK,
356 constants.OP_STATUS_CANCELING)
358 # Cancel here if we were asked to
359 if self.opcode.status == constants.OP_STATUS_CANCELING:
362 self.opcode.status = constants.OP_STATUS_RUNNING
366 def RunTask(self, job):
369 This functions processes a job. It is closely tied to the _QueuedJob and
370 _QueuedOpCode classes.
372 @type job: L{_QueuedJob}
373 @param job: the job to be processed
376 logging.info("Worker %s processing job %s",
377 self.worker_id, job.id)
378 proc = mcpu.Processor(self.pool.queue.context)
379 self.queue = queue = job.queue
383 for idx, op in enumerate(job.ops):
384 op_summary = op.input.Summary()
385 if op.status == constants.OP_STATUS_SUCCESS:
386 # this is a job that was partially completed before master
387 # daemon shutdown, so it can be expected that some opcodes
388 # are already completed successfully (if any did error
389 # out, then the whole job should have been aborted and not
390 # resubmitted for processing)
391 logging.info("Op %s/%s: opcode %s already processed, skipping",
392 idx + 1, count, op_summary)
395 logging.info("Op %s/%s: Starting opcode %s", idx + 1, count,
400 if op.status == constants.OP_STATUS_CANCELED:
402 assert op.status == constants.OP_STATUS_QUEUED
403 job.run_op_index = idx
404 op.status = constants.OP_STATUS_WAITLOCK
406 op.start_timestamp = TimeStampNow()
407 if idx == 0: # first opcode
408 job.start_timestamp = op.start_timestamp
409 queue.UpdateJobUnlocked(job)
411 input_opcode = op.input
416 """Append a log entry.
422 log_type = constants.ELOG_MESSAGE
425 log_type, log_msg = args
427 # The time is split to make serialization easier and not lose
429 timestamp = utils.SplitTime(time.time())
434 op.log.append((job.log_serial, timestamp, log_type, log_msg))
436 job.change.notifyAll()
440 # Make sure not to hold lock while _Log is called
442 result = proc.ExecOpCode(input_opcode, _Log, self._NotifyStart)
446 op.status = constants.OP_STATUS_SUCCESS
448 op.end_timestamp = TimeStampNow()
449 queue.UpdateJobUnlocked(job)
453 logging.info("Op %s/%s: Successfully finished opcode %s",
454 idx + 1, count, op_summary)
456 # Will be handled further up
458 except Exception, err:
462 op.status = constants.OP_STATUS_ERROR
464 op.end_timestamp = TimeStampNow()
465 logging.info("Op %s/%s: Error in opcode %s: %s",
466 idx + 1, count, op_summary, err)
468 queue.UpdateJobUnlocked(job)
476 queue.CancelJobUnlocked(job)
479 except errors.GenericError, err:
480 logging.exception("Ganeti exception")
482 logging.exception("Unhandled exception")
487 job.run_op_index = -1
488 job.end_timestamp = TimeStampNow()
489 queue.UpdateJobUnlocked(job)
492 status = job.CalcStatus()
495 logging.info("Worker %s finished job %s, status = %s",
496 self.worker_id, job_id, status)
499 class _JobQueueWorkerPool(workerpool.WorkerPool):
500 """Simple class implementing a job-processing workerpool.
503 def __init__(self, queue):
504 super(_JobQueueWorkerPool, self).__init__(JOBQUEUE_THREADS,
509 class JobQueue(object):
510 """Queue used to manage the jobs.
512 @cvar _RE_JOB_FILE: regex matching the valid job file names
515 _RE_JOB_FILE = re.compile(r"^job-(%s)$" % constants.JOB_ID_TEMPLATE)
517 def _RequireOpenQueue(fn):
518 """Decorator for "public" functions.
520 This function should be used for all 'public' functions. That is,
521 functions usually called from other classes.
523 @warning: Use this decorator only after utils.LockedMethod!
532 def wrapper(self, *args, **kwargs):
533 assert self._queue_lock is not None, "Queue should be open"
534 return fn(self, *args, **kwargs)
537 def __init__(self, context):
538 """Constructor for JobQueue.
540 The constructor will initialize the job queue object and then
541 start loading the current jobs from disk, either for starting them
542 (if they were queue) or for aborting them (if they were already
545 @type context: GanetiContext
546 @param context: the context object for access to the configuration
547 data and other ganeti objects
550 self.context = context
551 self._memcache = weakref.WeakValueDictionary()
552 self._my_hostname = utils.HostInfo().name
555 self._lock = threading.Lock()
556 self.acquire = self._lock.acquire
557 self.release = self._lock.release
560 self._queue_lock = jstore.InitAndVerifyQueue(must_lock=True)
563 self._last_serial = jstore.ReadSerial()
564 assert self._last_serial is not None, ("Serial file was modified between"
565 " check in jstore and here")
567 # Get initial list of nodes
568 self._nodes = dict((n.name, n.primary_ip)
569 for n in self.context.cfg.GetAllNodesInfo().values()
570 if n.master_candidate)
574 del self._nodes[self._my_hostname]
578 # TODO: Check consistency across nodes
581 self._wpool = _JobQueueWorkerPool(self)
583 # We need to lock here because WorkerPool.AddTask() may start a job while
584 # we're still doing our work.
587 logging.info("Inspecting job queue")
589 all_job_ids = self._GetJobIDsUnlocked()
590 jobs_count = len(all_job_ids)
591 lastinfo = time.time()
592 for idx, job_id in enumerate(all_job_ids):
593 # Give an update every 1000 jobs or 10 seconds
594 if (idx % 1000 == 0 or time.time() >= (lastinfo + 10.0) or
595 idx == (jobs_count - 1)):
596 logging.info("Job queue inspection: %d/%d (%0.1f %%)",
597 idx, jobs_count - 1, 100.0 * (idx + 1) / jobs_count)
598 lastinfo = time.time()
600 job = self._LoadJobUnlocked(job_id)
602 # a failure in loading the job can cause 'None' to be returned
606 status = job.CalcStatus()
608 if status in (constants.JOB_STATUS_QUEUED, ):
609 self._wpool.AddTask(job)
611 elif status in (constants.JOB_STATUS_RUNNING,
612 constants.JOB_STATUS_WAITLOCK,
613 constants.JOB_STATUS_CANCELING):
614 logging.warning("Unfinished job %s found: %s", job.id, job)
616 job.MarkUnfinishedOps(constants.OP_STATUS_ERROR,
617 "Unclean master daemon shutdown")
619 self.UpdateJobUnlocked(job)
621 logging.info("Job queue inspection finished")
625 self._wpool.TerminateWorkers()
630 def AddNode(self, node):
631 """Register a new node with the queue.
633 @type node: L{objects.Node}
634 @param node: the node object to be added
637 node_name = node.name
638 assert node_name != self._my_hostname
640 # Clean queue directory on added node
641 rpc.RpcRunner.call_jobqueue_purge(node_name)
643 if not node.master_candidate:
644 # remove if existing, ignoring errors
645 self._nodes.pop(node_name, None)
646 # and skip the replication of the job ids
649 # Upload the whole queue excluding archived jobs
650 files = [self._GetJobPath(job_id) for job_id in self._GetJobIDsUnlocked()]
652 # Upload current serial file
653 files.append(constants.JOB_QUEUE_SERIAL_FILE)
655 for file_name in files:
657 fd = open(file_name, "r")
663 result = rpc.RpcRunner.call_jobqueue_update([node_name],
666 if not result[node_name]:
667 logging.error("Failed to upload %s to %s", file_name, node_name)
669 self._nodes[node_name] = node.primary_ip
673 def RemoveNode(self, node_name):
674 """Callback called when removing nodes from the cluster.
677 @param node_name: the name of the node to remove
681 # The queue is removed by the "leave node" RPC call.
682 del self._nodes[node_name]
686 def _CheckRpcResult(self, result, nodes, failmsg):
687 """Verifies the status of an RPC call.
689 Since we aim to keep consistency should this node (the current
690 master) fail, we will log errors if our rpc fail, and especially
691 log the case when more than half of the nodes fails.
693 @param result: the data as returned from the rpc call
695 @param nodes: the list of nodes we made the call to
697 @param failmsg: the identifier to be used for logging
710 logging.error("%s failed on %s", failmsg, ", ".join(failed))
712 # +1 for the master node
713 if (len(success) + 1) < len(failed):
714 # TODO: Handle failing nodes
715 logging.error("More than half of the nodes failed")
717 def _GetNodeIp(self):
718 """Helper for returning the node name/ip list.
721 @return: a tuple of two lists, the first one with the node
722 names and the second one with the node addresses
725 name_list = self._nodes.keys()
726 addr_list = [self._nodes[name] for name in name_list]
727 return name_list, addr_list
729 def _WriteAndReplicateFileUnlocked(self, file_name, data):
730 """Writes a file locally and then replicates it to all nodes.
732 This function will replace the contents of a file on the local
733 node and then replicate it to all the other nodes we have.
736 @param file_name: the path of the file to be replicated
738 @param data: the new contents of the file
741 utils.WriteFile(file_name, data=data)
743 names, addrs = self._GetNodeIp()
744 result = rpc.RpcRunner.call_jobqueue_update(names, addrs, file_name, data)
745 self._CheckRpcResult(result, self._nodes,
746 "Updating %s" % file_name)
748 def _RenameFilesUnlocked(self, rename):
749 """Renames a file locally and then replicate the change.
751 This function will rename a file in the local queue directory
752 and then replicate this rename to all the other nodes we have.
754 @type rename: list of (old, new)
755 @param rename: List containing tuples mapping old to new names
758 # Rename them locally
759 for old, new in rename:
760 utils.RenameFile(old, new, mkdir=True)
762 # ... and on all nodes
763 names, addrs = self._GetNodeIp()
764 result = rpc.RpcRunner.call_jobqueue_rename(names, addrs, rename)
765 self._CheckRpcResult(result, self._nodes, "Renaming files (%r)" % rename)
767 def _FormatJobID(self, job_id):
768 """Convert a job ID to string format.
770 Currently this just does C{str(job_id)} after performing some
771 checks, but if we want to change the job id format this will
772 abstract this change.
774 @type job_id: int or long
775 @param job_id: the numeric job id
777 @return: the formatted job id
780 if not isinstance(job_id, (int, long)):
781 raise errors.ProgrammerError("Job ID '%s' not numeric" % job_id)
783 raise errors.ProgrammerError("Job ID %s is negative" % job_id)
788 def _GetArchiveDirectory(cls, job_id):
789 """Returns the archive directory for a job.
792 @param job_id: Job identifier
794 @return: Directory name
797 return str(int(job_id) / JOBS_PER_ARCHIVE_DIRECTORY)
799 def _NewSerialUnlocked(self):
800 """Generates a new job identifier.
802 Job identifiers are unique during the lifetime of a cluster.
805 @return: a string representing the job identifier.
809 serial = self._last_serial + 1
812 self._WriteAndReplicateFileUnlocked(constants.JOB_QUEUE_SERIAL_FILE,
815 # Keep it only if we were able to write the file
816 self._last_serial = serial
818 return self._FormatJobID(serial)
821 def _GetJobPath(job_id):
822 """Returns the job file for a given job id.
825 @param job_id: the job identifier
827 @return: the path to the job file
830 return os.path.join(constants.QUEUE_DIR, "job-%s" % job_id)
833 def _GetArchivedJobPath(cls, job_id):
834 """Returns the archived job file for a give job id.
837 @param job_id: the job identifier
839 @return: the path to the archived job file
842 path = "%s/job-%s" % (cls._GetArchiveDirectory(job_id), job_id)
843 return os.path.join(constants.JOB_QUEUE_ARCHIVE_DIR, path)
846 def _ExtractJobID(cls, name):
847 """Extract the job id from a filename.
850 @param name: the job filename
851 @rtype: job id or None
852 @return: the job id corresponding to the given filename,
853 or None if the filename does not represent a valid
857 m = cls._RE_JOB_FILE.match(name)
863 def _GetJobIDsUnlocked(self, archived=False):
864 """Return all known job IDs.
866 If the parameter archived is True, archived jobs IDs will be
867 included. Currently this argument is unused.
869 The method only looks at disk because it's a requirement that all
870 jobs are present on disk (so in the _memcache we don't have any
874 @return: the list of job IDs
877 jlist = [self._ExtractJobID(name) for name in self._ListJobFiles()]
878 jlist = utils.NiceSort(jlist)
881 def _ListJobFiles(self):
882 """Returns the list of current job files.
885 @return: the list of job file names
888 return [name for name in utils.ListVisibleFiles(constants.QUEUE_DIR)
889 if self._RE_JOB_FILE.match(name)]
891 def _LoadJobUnlocked(self, job_id):
892 """Loads a job from the disk or memory.
894 Given a job id, this will return the cached job object if
895 existing, or try to load the job from the disk. If loading from
896 disk, it will also add the job to the cache.
898 @param job_id: the job id
899 @rtype: L{_QueuedJob} or None
900 @return: either None or the job object
903 job = self._memcache.get(job_id, None)
905 logging.debug("Found job %s in memcache", job_id)
908 filepath = self._GetJobPath(job_id)
909 logging.debug("Loading job from %s", filepath)
911 fd = open(filepath, "r")
913 if err.errno in (errno.ENOENT, ):
917 data = serializer.LoadJson(fd.read())
922 job = _QueuedJob.Restore(self, data)
923 except Exception, err:
924 new_path = self._GetArchivedJobPath(job_id)
925 if filepath == new_path:
926 # job already archived (future case)
927 logging.exception("Can't parse job %s", job_id)
930 logging.exception("Can't parse job %s, will archive.", job_id)
931 self._RenameFilesUnlocked([(filepath, new_path)])
934 self._memcache[job_id] = job
935 logging.debug("Added job %s to the cache", job_id)
938 def _GetJobsUnlocked(self, job_ids):
939 """Return a list of jobs based on their IDs.
942 @param job_ids: either an empty list (meaning all jobs),
945 @return: the list of job objects
949 job_ids = self._GetJobIDsUnlocked()
951 return [self._LoadJobUnlocked(job_id) for job_id in job_ids]
954 def _IsQueueMarkedDrain():
955 """Check if the queue is marked from drain.
957 This currently uses the queue drain file, which makes it a
958 per-node flag. In the future this can be moved to the config file.
961 @return: True of the job queue is marked for draining
964 return os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
967 def SetDrainFlag(drain_flag):
968 """Sets the drain flag for the queue.
970 This is similar to the function L{backend.JobQueueSetDrainFlag},
971 and in the future we might merge them.
973 @type drain_flag: boolean
974 @param drain_flag: Whether to set or unset the drain flag
978 utils.WriteFile(constants.JOB_QUEUE_DRAIN_FILE, data="", close=True)
980 utils.RemoveFile(constants.JOB_QUEUE_DRAIN_FILE)
984 def _SubmitJobUnlocked(self, ops):
985 """Create and store a new job.
987 This enters the job into our job queue and also puts it on the new
988 queue, in order for it to be picked up by the queue processors.
991 @param ops: The list of OpCodes that will become the new job.
993 @return: the job ID of the newly created job
994 @raise errors.JobQueueDrainError: if the job is marked for draining
997 if self._IsQueueMarkedDrain():
998 raise errors.JobQueueDrainError("Job queue is drained, refusing job")
1000 # Check job queue size
1001 size = len(self._ListJobFiles())
1002 if size >= constants.JOB_QUEUE_SIZE_SOFT_LIMIT:
1003 # TODO: Autoarchive jobs. Make sure it's not done on every job
1004 # submission, though.
1008 if size >= constants.JOB_QUEUE_SIZE_HARD_LIMIT:
1009 raise errors.JobQueueFull()
1011 # Get job identifier
1012 job_id = self._NewSerialUnlocked()
1013 job = _QueuedJob(self, job_id, ops)
1016 self.UpdateJobUnlocked(job)
1018 logging.debug("Adding new job %s to the cache", job_id)
1019 self._memcache[job_id] = job
1021 # Add to worker pool
1022 self._wpool.AddTask(job)
1028 def SubmitJob(self, ops):
1029 """Create and store a new job.
1031 @see: L{_SubmitJobUnlocked}
1034 return self._SubmitJobUnlocked(ops)
1038 def SubmitManyJobs(self, jobs):
1039 """Create and store multiple jobs.
1041 @see: L{_SubmitJobUnlocked}
1047 data = self._SubmitJobUnlocked(ops)
1049 except errors.GenericError, err:
1052 results.append((status, data))
1058 def UpdateJobUnlocked(self, job):
1059 """Update a job's on disk storage.
1061 After a job has been modified, this function needs to be called in
1062 order to write the changes to disk and replicate them to the other
1065 @type job: L{_QueuedJob}
1066 @param job: the changed job
1069 filename = self._GetJobPath(job.id)
1070 data = serializer.DumpJson(job.Serialize(), indent=False)
1071 logging.debug("Writing job %s to %s", job.id, filename)
1072 self._WriteAndReplicateFileUnlocked(filename, data)
1074 # Notify waiters about potential changes
1075 job.change.notifyAll()
1079 def WaitForJobChanges(self, job_id, fields, prev_job_info, prev_log_serial,
1081 """Waits for changes in a job.
1083 @type job_id: string
1084 @param job_id: Job identifier
1085 @type fields: list of strings
1086 @param fields: Which fields to check for changes
1087 @type prev_job_info: list or None
1088 @param prev_job_info: Last job information returned
1089 @type prev_log_serial: int
1090 @param prev_log_serial: Last job message serial number
1091 @type timeout: float
1092 @param timeout: maximum time to wait
1093 @rtype: tuple (job info, log entries)
1094 @return: a tuple of the job information as required via
1095 the fields parameter, and the log entries as a list
1097 if the job has not changed and the timeout has expired,
1098 we instead return a special value,
1099 L{constants.JOB_NOTCHANGED}, which should be interpreted
1100 as such by the clients
1103 logging.debug("Waiting for changes in job %s", job_id)
1104 end_time = time.time() + timeout
1106 delta_time = end_time - time.time()
1108 return constants.JOB_NOTCHANGED
1110 job = self._LoadJobUnlocked(job_id)
1112 logging.debug("Job %s not found", job_id)
1115 status = job.CalcStatus()
1116 job_info = self._GetJobInfoUnlocked(job, fields)
1117 log_entries = job.GetLogEntries(prev_log_serial)
1119 # Serializing and deserializing data can cause type changes (e.g. from
1120 # tuple to list) or precision loss. We're doing it here so that we get
1121 # the same modifications as the data received from the client. Without
1122 # this, the comparison afterwards might fail without the data being
1123 # significantly different.
1124 job_info = serializer.LoadJson(serializer.DumpJson(job_info))
1125 log_entries = serializer.LoadJson(serializer.DumpJson(log_entries))
1127 if status not in (constants.JOB_STATUS_QUEUED,
1128 constants.JOB_STATUS_RUNNING,
1129 constants.JOB_STATUS_WAITLOCK):
1130 # Don't even try to wait if the job is no longer running, there will be
1134 if (prev_job_info != job_info or
1135 (log_entries and prev_log_serial != log_entries[0][0])):
1138 logging.debug("Waiting again")
1140 # Release the queue lock while waiting
1141 job.change.wait(delta_time)
1143 logging.debug("Job %s changed", job_id)
1145 return (job_info, log_entries)
1149 def CancelJob(self, job_id):
1152 This will only succeed if the job has not started yet.
1154 @type job_id: string
1155 @param job_id: job ID of job to be cancelled.
1158 logging.info("Cancelling job %s", job_id)
1160 job = self._LoadJobUnlocked(job_id)
1162 logging.debug("Job %s not found", job_id)
1163 return (False, "Job %s not found" % job_id)
1165 job_status = job.CalcStatus()
1167 if job_status not in (constants.JOB_STATUS_QUEUED,
1168 constants.JOB_STATUS_WAITLOCK):
1169 logging.debug("Job %s is no longer in the queue", job.id)
1170 return (False, "Job %s is no longer in the queue" % job.id)
1172 if job_status == constants.JOB_STATUS_QUEUED:
1173 self.CancelJobUnlocked(job)
1174 return (True, "Job %s canceled" % job.id)
1176 elif job_status == constants.JOB_STATUS_WAITLOCK:
1177 # The worker will notice the new status and cancel the job
1179 job.MarkUnfinishedOps(constants.OP_STATUS_CANCELING, None)
1181 self.UpdateJobUnlocked(job)
1182 return (True, "Job %s will be canceled" % job.id)
1185 def CancelJobUnlocked(self, job):
1186 """Marks a job as canceled.
1190 job.MarkUnfinishedOps(constants.OP_STATUS_CANCELED,
1191 "Job canceled by request")
1193 self.UpdateJobUnlocked(job)
1196 def _ArchiveJobsUnlocked(self, jobs):
1199 @type jobs: list of L{_QueuedJob}
1200 @param jobs: Job objects
1202 @return: Number of archived jobs
1208 if job.CalcStatus() not in (constants.JOB_STATUS_CANCELED,
1209 constants.JOB_STATUS_SUCCESS,
1210 constants.JOB_STATUS_ERROR):
1211 logging.debug("Job %s is not yet done", job.id)
1214 archive_jobs.append(job)
1216 old = self._GetJobPath(job.id)
1217 new = self._GetArchivedJobPath(job.id)
1218 rename_files.append((old, new))
1220 # TODO: What if 1..n files fail to rename?
1221 self._RenameFilesUnlocked(rename_files)
1223 logging.debug("Successfully archived job(s) %s",
1224 ", ".join(job.id for job in archive_jobs))
1226 return len(archive_jobs)
1230 def ArchiveJob(self, job_id):
1233 This is just a wrapper over L{_ArchiveJobsUnlocked}.
1235 @type job_id: string
1236 @param job_id: Job ID of job to be archived.
1238 @return: Whether job was archived
1241 logging.info("Archiving job %s", job_id)
1243 job = self._LoadJobUnlocked(job_id)
1245 logging.debug("Job %s not found", job_id)
1248 return self._ArchiveJobsUnlocked([job]) == 1
1252 def AutoArchiveJobs(self, age, timeout):
1253 """Archives all jobs based on age.
1255 The method will archive all jobs which are older than the age
1256 parameter. For jobs that don't have an end timestamp, the start
1257 timestamp will be considered. The special '-1' age will cause
1258 archival of all jobs (that are not running or queued).
1261 @param age: the minimum age in seconds
1264 logging.info("Archiving jobs with age more than %s seconds", age)
1267 end_time = now + timeout
1271 all_job_ids = self._GetJobIDsUnlocked(archived=False)
1273 for idx, job_id in enumerate(all_job_ids):
1276 # Not optimal because jobs could be pending
1277 # TODO: Measure average duration for job archival and take number of
1278 # pending jobs into account.
1279 if time.time() > end_time:
1282 # Returns None if the job failed to load
1283 job = self._LoadJobUnlocked(job_id)
1285 if job.end_timestamp is None:
1286 if job.start_timestamp is None:
1287 job_age = job.received_timestamp
1289 job_age = job.start_timestamp
1291 job_age = job.end_timestamp
1293 if age == -1 or now - job_age[0] > age:
1296 # Archive 10 jobs at a time
1297 if len(pending) >= 10:
1298 archived_count += self._ArchiveJobsUnlocked(pending)
1302 archived_count += self._ArchiveJobsUnlocked(pending)
1304 return (archived_count, len(all_job_ids) - last_touched - 1)
1306 def _GetJobInfoUnlocked(self, job, fields):
1307 """Returns information about a job.
1309 @type job: L{_QueuedJob}
1310 @param job: the job which we query
1312 @param fields: names of fields to return
1314 @return: list with one element for each field
1315 @raise errors.OpExecError: when an invalid field
1320 for fname in fields:
1323 elif fname == "status":
1324 row.append(job.CalcStatus())
1325 elif fname == "ops":
1326 row.append([op.input.__getstate__() for op in job.ops])
1327 elif fname == "opresult":
1328 row.append([op.result for op in job.ops])
1329 elif fname == "opstatus":
1330 row.append([op.status for op in job.ops])
1331 elif fname == "oplog":
1332 row.append([op.log for op in job.ops])
1333 elif fname == "opstart":
1334 row.append([op.start_timestamp for op in job.ops])
1335 elif fname == "opend":
1336 row.append([op.end_timestamp for op in job.ops])
1337 elif fname == "received_ts":
1338 row.append(job.received_timestamp)
1339 elif fname == "start_ts":
1340 row.append(job.start_timestamp)
1341 elif fname == "end_ts":
1342 row.append(job.end_timestamp)
1343 elif fname == "summary":
1344 row.append([op.input.Summary() for op in job.ops])
1346 raise errors.OpExecError("Invalid job query field '%s'" % fname)
1351 def QueryJobs(self, job_ids, fields):
1352 """Returns a list of jobs in queue.
1354 This is a wrapper of L{_GetJobsUnlocked}, which actually does the
1355 processing for each job.
1358 @param job_ids: sequence of job identifiers or None for all
1360 @param fields: names of fields to return
1362 @return: list one element per job, each element being list with
1363 the requested fields
1368 for job in self._GetJobsUnlocked(job_ids):
1372 jobs.append(self._GetJobInfoUnlocked(job, fields))
1379 """Stops the job queue.
1381 This shutdowns all the worker threads an closes the queue.
1384 self._wpool.TerminateWorkers()
1386 self._queue_lock.Close()
1387 self._queue_lock = None