import logging
import errno
-import re
import time
import weakref
import threading
+import itertools
try:
- # pylint: disable-msg=E0611
+ # pylint: disable=E0611
from pyinotify import pyinotify
except ImportError:
import pyinotify
@ivar writable: Whether the job is allowed to be modified
"""
- # pylint: disable-msg=W0212
+ # pylint: disable=W0212
__slots__ = ["queue", "id", "ops", "log_serial", "ops_iter", "cur_opctx",
"received_timestamp", "start_timestamp", "end_timestamp",
"__weakref__", "processor_lock", "writable"]
if op.status == constants.OP_STATUS_QUEUED:
pass
- elif op.status == constants.OP_STATUS_WAITLOCK:
- status = constants.JOB_STATUS_WAITLOCK
+ elif op.status == constants.OP_STATUS_WAITING:
+ status = constants.JOB_STATUS_WAITING
elif op.status == constants.OP_STATUS_RUNNING:
status = constants.JOB_STATUS_RUNNING
elif op.status == constants.OP_STATUS_CANCELING:
self.Finalize()
return (True, "Job %s canceled" % self.id)
- elif status == constants.JOB_STATUS_WAITLOCK:
+ elif status == constants.JOB_STATUS_WAITING:
# The worker will notice the new status and cancel the job
self.MarkUnfinishedOps(constants.OP_STATUS_CANCELING, None)
return (True, "Job %s will be canceled" % self.id)
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.
+ Processor.ExecOpCode) set to OP_STATUS_WAITING.
"""
assert self._op in self._job.ops
- assert self._op.status in (constants.OP_STATUS_WAITLOCK,
+ assert self._op.status in (constants.OP_STATUS_WAITING,
constants.OP_STATUS_CANCELING)
# Cancel here if we were asked to
"""Check whether job has been cancelled.
"""
- assert self._op.status in (constants.OP_STATUS_WAITLOCK,
+ assert self._op.status in (constants.OP_STATUS_WAITING,
constants.OP_STATUS_CANCELING)
# Cancel here if we were asked to
# no changes.
if (status not in (constants.JOB_STATUS_QUEUED,
constants.JOB_STATUS_RUNNING,
- constants.JOB_STATUS_WAITLOCK) or
+ constants.JOB_STATUS_WAITING) 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)
"""
@staticmethod
- def _CheckForChanges(job_load_fn, check_fn):
+ def _CheckForChanges(counter, job_load_fn, check_fn):
+ if counter.next() > 0:
+ # If this isn't the first check the job is given some more time to change
+ # again. This gives better performance for jobs generating many
+ # changes/messages.
+ time.sleep(0.1)
+
job = job_load_fn()
if not job:
raise errors.JobLost()
@param timeout: maximum time to wait in seconds
"""
+ counter = itertools.count()
try:
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),
+ counter, job_load_fn, check_fn),
utils.RETRY_REMAINING_TIME, timeout,
wait_fn=waiter.Wait)
finally:
"""
assert op in job.ops
assert op.status in (constants.OP_STATUS_QUEUED,
- constants.OP_STATUS_WAITLOCK)
+ constants.OP_STATUS_WAITING)
update = False
op.result = None
if op.status == constants.OP_STATUS_QUEUED:
- op.status = constants.OP_STATUS_WAITLOCK
+ op.status = constants.OP_STATUS_WAITING
update = True
if op.start_timestamp is None:
job.start_timestamp = op.start_timestamp
update = True
- assert op.status == constants.OP_STATUS_WAITLOCK
+ assert op.status == constants.OP_STATUS_WAITING
return update
"""
op = opctx.op
- assert op.status == constants.OP_STATUS_WAITLOCK
+ assert op.status == constants.OP_STATUS_WAITING
timeout = opctx.GetNextLockTimeout()
assert timeout is not None, "Received timeout for blocking acquire"
logging.debug("Couldn't acquire locks in %0.6fs", timeout)
- assert op.status in (constants.OP_STATUS_WAITLOCK,
+ assert op.status in (constants.OP_STATUS_WAITING,
constants.OP_STATUS_CANCELING)
# Was job cancelled while we were waiting for the lock?
return (constants.OP_STATUS_CANCELING, None)
# Stay in waitlock while trying to re-acquire lock
- return (constants.OP_STATUS_WAITLOCK, None)
+ return (constants.OP_STATUS_WAITING, None)
except CancelJob:
logging.exception("%s: Canceling job", opctx.log_prefix)
assert op.status == constants.OP_STATUS_CANCELING
return (constants.OP_STATUS_CANCELING, None)
- except Exception, err: # pylint: disable-msg=W0703
+ except Exception, err: # pylint: disable=W0703
logging.exception("%s: Caught exception in %s",
opctx.log_prefix, opctx.summary)
return (constants.OP_STATUS_ERROR, _EncodeOpError(err))
for i in job.ops[opctx.index + 1:])
assert op.status in (constants.OP_STATUS_QUEUED,
- constants.OP_STATUS_WAITLOCK,
+ constants.OP_STATUS_WAITING,
constants.OP_STATUS_CANCELING)
assert (op.priority <= constants.OP_PRIO_LOWEST and
if op.status != constants.OP_STATUS_CANCELING:
assert op.status in (constants.OP_STATUS_QUEUED,
- constants.OP_STATUS_WAITLOCK)
+ constants.OP_STATUS_WAITING)
# Prepare to start opcode
if self._MarkWaitlock(job, op):
# Write to disk
queue.UpdateJobUnlocked(job)
- assert op.status == constants.OP_STATUS_WAITLOCK
- assert job.CalcStatus() == constants.JOB_STATUS_WAITLOCK
+ assert op.status == constants.OP_STATUS_WAITING
+ assert job.CalcStatus() == constants.JOB_STATUS_WAITING
assert job.start_timestamp and op.start_timestamp
assert waitjob is None
# Check if waiting for a job is necessary
waitjob = self._CheckDependencies(queue, job, opctx)
- assert op.status in (constants.OP_STATUS_WAITLOCK,
+ assert op.status in (constants.OP_STATUS_WAITING,
constants.OP_STATUS_CANCELING,
constants.OP_STATUS_ERROR)
assert not waitjob
- if op.status == constants.OP_STATUS_WAITLOCK:
+ if op.status == constants.OP_STATUS_WAITING:
# Couldn't get locks in time
assert not op.end_timestamp
else:
else:
assert op.status in constants.OPS_FINALIZED
- if op.status == constants.OP_STATUS_WAITLOCK or waitjob:
+ if op.status == constants.OP_STATUS_WAITING or waitjob:
finalize = False
if not waitjob and opctx.CheckPriorityIncrease():
op.priority >= constants.OP_PRIO_HIGHEST)
# In no case must the status be finalized here
- assert job.CalcStatus() == constants.JOB_STATUS_WAITLOCK
+ assert job.CalcStatus() == constants.JOB_STATUS_WAITING
else:
# Ensure all opcodes so far have been successful
"""The actual job workers.
"""
- def RunTask(self, job): # pylint: disable-msg=W0221
+ def RunTask(self, job): # pylint: disable=W0221
"""Job executor.
@type job: L{_QueuedJob}
CONTINUE,
WRONGSTATUS) = range(1, 6)
- # TODO: Export waiter information to lock monitor
-
def __init__(self, getstatus_fn, enqueue_fn):
"""Initializes this class.
self._lock = locking.SharedLock("JobDepMgr")
@locking.ssynchronized(_LOCK, shared=1)
+ def GetLockInfo(self, requested): # pylint: disable=W0613
+ """Retrieves information about waiting jobs.
+
+ @type requested: set
+ @param requested: Requested information, see C{query.LQ_*}
+
+ """
+ # No need to sort here, that's being done by the lock manager and query
+ # library. There are no priorities for notifying jobs, hence all show up as
+ # one item under "pending".
+ return [("job/%s" % job_id, None, None,
+ [("job", [job.id for job in waiters])])
+ for job_id, waiters in self._waiters.items()
+ if waiters]
+
+ @locking.ssynchronized(_LOCK, shared=1)
def JobWaiting(self, job):
"""Checks if a job is waiting.
"""
def wrapper(self, *args, **kwargs):
- # pylint: disable-msg=W0212
+ # pylint: disable=W0212
assert self._queue_filelock is not None, "Queue should be open"
return fn(self, *args, **kwargs)
return wrapper
class JobQueue(object):
"""Queue used to manage the jobs.
- @cvar _RE_JOB_FILE: regex matching the valid job file names
-
"""
- _RE_JOB_FILE = re.compile(r"^job-(%s)$" % constants.JOB_ID_TEMPLATE)
-
def __init__(self, context):
"""Constructor for JobQueue.
# Job dependencies
self.depmgr = _JobDependencyManager(self._GetJobStatusForDependencies,
self._EnqueueJobs)
+ self.context.glm.AddToLockMonitor(self.depmgr)
# Setup worker pool
self._wpool = _JobQueueWorkerPool(self)
restartjobs.append(job)
elif status in (constants.JOB_STATUS_RUNNING,
- constants.JOB_STATUS_WAITLOCK,
+ constants.JOB_STATUS_WAITING,
constants.JOB_STATUS_CANCELING):
logging.warning("Unfinished job %s found: %s", job.id, job)
- if status == constants.JOB_STATUS_WAITLOCK:
+ if status == constants.JOB_STATUS_WAITING:
# Restart job
job.MarkUnfinishedOps(constants.OP_STATUS_QUEUED, None)
restartjobs.append(job)
@return: a string representing the job identifier.
"""
- assert count > 0
+ assert ht.TPositiveInt(count)
+
# New number
serial = self._last_serial + count
return utils.PathJoin(constants.JOB_QUEUE_ARCHIVE_DIR,
cls._GetArchiveDirectory(job_id), "job-%s" % job_id)
- def _GetJobIDsUnlocked(self, sort=True):
+ @staticmethod
+ def _GetJobIDsUnlocked(sort=True):
"""Return all known job IDs.
The method only looks at disk because it's a requirement that all
"""
jlist = []
for filename in utils.ListVisibleFiles(constants.QUEUE_DIR):
- m = self._RE_JOB_FILE.match(filename)
+ m = constants.JOB_FILE_RE.match(filename)
if m:
jlist.append(m.group(1))
if sort:
try:
data = serializer.LoadJson(raw_data)
job = _QueuedJob.Restore(self, data, writable)
- except Exception, err: # pylint: disable-msg=W0703
+ except Exception, err: # pylint: disable=W0703
raise errors.JobFileCorrupted(err)
return job
# Try to load from disk
job = self.SafeLoadJobFromDisk(job_id, True, writable=False)
- assert not job.writable, "Got writable job"
+ assert not job.writable, "Got writable job" # pylint: disable=E1101
if job:
return job.CalcStatus()
projects / ganeti-local / blobdiff