@ivar id: the job ID
@type ops: list
@ivar ops: the list of _QueuedOpCode that constitute the job
- @type run_op_index: int
- @ivar run_op_index: the currently executing opcode, or -1 if
- we didn't yet start executing
@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
+ @ivar lock_status: In-memory locking information for debugging
@ivar change: a Condition variable we use for waiting for job changes
"""
- __slots__ = ["queue", "id", "ops", "run_op_index", "log_serial",
+ # pylint: disable-msg=W0212
+ __slots__ = ["queue", "id", "ops", "log_serial",
"received_timestamp", "start_timestamp", "end_timestamp",
- "change",
+ "lock_status", "change",
"__weakref__"]
def __init__(self, queue, job_id, ops):
self.queue = queue
self.id = job_id
self.ops = [_QueuedOpCode(op) for op in ops]
- self.run_op_index = -1
self.log_serial = 0
self.received_timestamp = TimeStampNow()
self.start_timestamp = None
self.end_timestamp = None
+ # In-memory attributes
+ self.lock_status = None
+
# Condition to wait for changes
self.change = threading.Condition(self.queue._lock)
+ 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))
+
@classmethod
def Restore(cls, queue, state):
"""Restore a _QueuedJob from serialized state:
obj = _QueuedJob.__new__(cls)
obj.queue = queue
obj.id = state["id"]
- obj.run_op_index = state["run_op_index"]
obj.received_timestamp = state.get("received_timestamp", None)
obj.start_timestamp = state.get("start_timestamp", None)
obj.end_timestamp = state.get("end_timestamp", None)
+ # In-memory attributes
+ obj.lock_status = None
+
obj.ops = []
obj.log_serial = 0
for op_state in state["ops"]:
return {
"id": self.id,
"ops": [op.Serialize() for op in self.ops],
- "run_op_index": self.run_op_index,
"start_timestamp": self.start_timestamp,
"end_timestamp": self.end_timestamp,
"received_timestamp": self.received_timestamp,
not_marked = False
-class _JobQueueWorker(workerpool.BaseWorker):
- """The actual job workers.
+class _OpExecCallbacks(mcpu.OpExecCbBase):
+ def __init__(self, queue, job, op):
+ """Initializes this class.
- """
- def _NotifyStart(self):
+ @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
+
+ def NotifyStart(self):
"""Mark the opcode as running, not lock-waiting.
- 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.
+ 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.
"""
- assert self.queue, "Queue attribute is missing"
- assert self.opcode, "Opcode attribute is missing"
-
- self.queue.acquire()
+ self._queue.acquire()
try:
- assert self.opcode.status in (constants.OP_STATUS_WAITLOCK,
- constants.OP_STATUS_CANCELING)
+ assert self._op.status in (constants.OP_STATUS_WAITLOCK,
+ constants.OP_STATUS_CANCELING)
+
+ # All locks are acquired by now
+ self._job.lock_status = None
# Cancel here if we were asked to
- if self.opcode.status == constants.OP_STATUS_CANCELING:
+ if self._op.status == constants.OP_STATUS_CANCELING:
raise CancelJob()
- self.opcode.status = constants.OP_STATUS_RUNNING
+ self._op.status = constants.OP_STATUS_RUNNING
finally:
- self.queue.release()
+ self._queue.release()
- def RunTask(self, job):
+ 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())
+
+ self._queue.acquire()
+ try:
+ self._job.log_serial += 1
+ self._op.log.append((self._job.log_serial, timestamp, log_type, log_msg))
+
+ self._job.change.notifyAll()
+ finally:
+ self._queue.release()
+
+ def ReportLocks(self, msg):
+ """Write locking information to the job.
+
+ Called whenever the LU processor is waiting for a lock or has acquired one.
+
+ """
+ # Not getting the queue lock because this is a single assignment
+ self._job.lock_status = msg
+
+
+class _JobQueueWorker(workerpool.BaseWorker):
+ """The actual job workers.
+
+ """
+ def RunTask(self, job): # pylint: disable-msg=W0221
"""Job executor.
This functions processes a job. It is closely tied to the _QueuedJob and
@param job: the job to be processed
"""
- logging.info("Worker %s processing job %s",
- self.worker_id, job.id)
- proc = mcpu.Processor(self.pool.queue.context)
- self.queue = queue = job.queue
+ logging.info("Processing job %s", job.id)
+ proc = mcpu.Processor(self.pool.queue.context, job.id)
+ queue = job.queue
try:
try:
count = len(job.ops)
if op.status == constants.OP_STATUS_CANCELED:
raise CancelJob()
assert op.status == constants.OP_STATUS_QUEUED
- job.run_op_index = idx
op.status = constants.OP_STATUS_WAITLOCK
op.result = None
op.start_timestamp = TimeStampNow()
finally:
queue.release()
- def _Log(*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())
-
- queue.acquire()
- try:
- job.log_serial += 1
- op.log.append((job.log_serial, timestamp, log_type, log_msg))
-
- job.change.notifyAll()
- finally:
- queue.release()
-
- # Make sure not to hold lock while _Log is called
- self.opcode = op
- result = proc.ExecOpCode(input_opcode, _Log, self._NotifyStart)
+ # Make sure not to hold queue lock while calling ExecOpCode
+ result = proc.ExecOpCode(input_opcode,
+ _OpExecCallbacks(queue, job, op))
queue.acquire()
try:
queue.acquire()
try:
try:
- job.run_op_index = -1
+ job.lock_status = None
job.end_timestamp = TimeStampNow()
queue.UpdateJobUnlocked(job)
finally:
status = job.CalcStatus()
finally:
queue.release()
- logging.info("Worker %s finished job %s, status = %s",
- self.worker_id, job_id, status)
+
+ logging.info("Finished job %s, status = %s", job_id, status)
class _JobQueueWorkerPool(workerpool.WorkerPool):
"""
def __init__(self, queue):
- super(_JobQueueWorkerPool, self).__init__(JOBQUEUE_THREADS,
+ super(_JobQueueWorkerPool, self).__init__("JobQueue",
+ JOBQUEUE_THREADS,
_JobQueueWorker)
self.queue = queue
"""
def wrapper(self, *args, **kwargs):
+ # pylint: disable-msg=W0212
assert self._queue_lock is not None, "Queue should be open"
return fn(self, *args, **kwargs)
return wrapper
assert node_name != self._my_hostname
# Clean queue directory on added node
- rpc.RpcRunner.call_jobqueue_purge(node_name)
+ result = rpc.RpcRunner.call_jobqueue_purge(node_name)
+ msg = result.fail_msg
+ if msg:
+ logging.warning("Cannot cleanup queue directory on node %s: %s",
+ node_name, msg)
if not node.master_candidate:
# remove if existing, ignoring errors
for file_name in files:
# Read file content
- fd = open(file_name, "r")
- try:
- content = fd.read()
- finally:
- fd.close()
+ content = utils.ReadFile(file_name)
result = rpc.RpcRunner.call_jobqueue_update([node_name],
[node.primary_ip],
file_name, content)
- if not result[node_name]:
- logging.error("Failed to upload %s to %s", file_name, node_name)
+ msg = result[node_name].fail_msg
+ if msg:
+ logging.error("Failed to upload file %s to node %s: %s",
+ file_name, node_name, msg)
self._nodes[node_name] = node.primary_ip
except KeyError:
pass
- def _CheckRpcResult(self, result, nodes, failmsg):
+ @staticmethod
+ def _CheckRpcResult(result, nodes, failmsg):
"""Verifies the status of an RPC call.
Since we aim to keep consistency should this node (the current
success = []
for node in nodes:
- if result[node]:
- success.append(node)
- else:
+ msg = result[node].fail_msg
+ if msg:
failed.append(node)
-
- if failed:
- logging.error("%s failed on %s", failmsg, ", ".join(failed))
+ logging.error("RPC call %s (%s) failed on node %s: %s",
+ result[node].call, failmsg, node, msg)
+ else:
+ success.append(node)
# +1 for the master node
if (len(success) + 1) < len(failed):
result = rpc.RpcRunner.call_jobqueue_rename(names, addrs, rename)
self._CheckRpcResult(result, self._nodes, "Renaming files (%r)" % rename)
- def _FormatJobID(self, job_id):
+ @staticmethod
+ def _FormatJobID(job_id):
"""Convert a job ID to string format.
Currently this just does C{str(job_id)} after performing some
@return: the list of job IDs
"""
+ # pylint: disable-msg=W0613
jlist = [self._ExtractJobID(name) for name in self._ListJobFiles()]
jlist = utils.NiceSort(jlist)
return jlist
filepath = self._GetJobPath(job_id)
logging.debug("Loading job from %s", filepath)
try:
- fd = open(filepath, "r")
+ raw_data = utils.ReadFile(filepath)
except IOError, err:
if err.errno in (errno.ENOENT, ):
return None
raise
- try:
- data = serializer.LoadJson(fd.read())
- finally:
- fd.close()
+
+ data = serializer.LoadJson(raw_data)
try:
job = _QueuedJob.Restore(self, data)
- except Exception, err:
+ except Exception, err: # pylint: disable-msg=W0703
new_path = self._GetArchivedJobPath(job_id)
if filepath == new_path:
# job already archived (future case)
queue, in order for it to be picked up by the queue processors.
@type job_id: job ID
- @param jod_id: the job ID for the new job
+ @param job_id: the job ID for the new job
@type ops: list
@param ops: The list of OpCodes that will become the new job.
@rtype: job ID
return results
-
@_RequireOpenQueue
def UpdateJobUnlocked(self, job):
"""Update a job's on disk storage.
as such by the clients
"""
- logging.debug("Waiting for changes in job %s", job_id)
-
- job_info = None
- log_entries = None
-
- end_time = time.time() + timeout
- while True:
- delta_time = end_time - time.time()
- if delta_time < 0:
- return constants.JOB_NOTCHANGED
+ job = self._LoadJobUnlocked(job_id)
+ if not job:
+ logging.debug("Job %s not found", job_id)
+ return None
- job = self._LoadJobUnlocked(job_id)
- if not job:
- logging.debug("Job %s not found", job_id)
- break
+ def _CheckForChanges():
+ logging.debug("Waiting for changes in job %s", job_id)
status = job.CalcStatus()
job_info = self._GetJobInfoUnlocked(job, fields)
job_info = serializer.LoadJson(serializer.DumpJson(job_info))
log_entries = serializer.LoadJson(serializer.DumpJson(log_entries))
- if status not in (constants.JOB_STATUS_QUEUED,
- constants.JOB_STATUS_RUNNING,
- constants.JOB_STATUS_WAITLOCK):
- # Don't even try to wait if the job is no longer running, there will be
- # no changes.
- break
-
- if (prev_job_info != job_info or
+ # Don't even try to wait if the job is no longer running, there will be
+ # no changes.
+ if (status not in (constants.JOB_STATUS_QUEUED,
+ constants.JOB_STATUS_RUNNING,
+ constants.JOB_STATUS_WAITLOCK) or
+ prev_job_info != job_info or
(log_entries and prev_log_serial != log_entries[0][0])):
- break
-
- logging.debug("Waiting again")
-
- # Release the queue lock while waiting
- job.change.wait(delta_time)
+ logging.debug("Job %s changed", job_id)
+ return (job_info, log_entries)
- logging.debug("Job %s changed", job_id)
+ raise utils.RetryAgain()
- if job_info is None and log_entries is None:
- return None
- else:
- return (job_info, log_entries)
+ try:
+ # Setting wait function to release the queue lock while waiting
+ return utils.Retry(_CheckForChanges, utils.RETRY_REMAINING_TIME, timeout,
+ wait_fn=job.change.wait)
+ except utils.RetryTimeout:
+ return constants.JOB_NOTCHANGED
@utils.LockedMethod
@_RequireOpenQueue
self._RenameFilesUnlocked(rename_files)
logging.debug("Successfully archived job(s) %s",
- ", ".join(job.id for job in archive_jobs))
+ utils.CommaJoin(job.id for job in archive_jobs))
return len(archive_jobs)
all_job_ids = self._GetJobIDsUnlocked(archived=False)
pending = []
for idx, job_id in enumerate(all_job_ids):
- last_touched = idx
+ last_touched = idx + 1
# Not optimal because jobs could be pending
# TODO: Measure average duration for job archival and take number of
if pending:
archived_count += self._ArchiveJobsUnlocked(pending)
- return (archived_count, len(all_job_ids) - last_touched - 1)
+ return (archived_count, len(all_job_ids) - last_touched)
- def _GetJobInfoUnlocked(self, job, fields):
+ @staticmethod
+ def _GetJobInfoUnlocked(job, fields):
"""Returns information about a job.
@type job: L{_QueuedJob}
row.append(job.start_timestamp)
elif fname == "end_ts":
row.append(job.end_timestamp)
+ elif fname == "lock_status":
+ row.append(job.lock_status)
elif fname == "summary":
row.append([op.input.Summary() for op in job.ops])
else: