+ def _OnInotify(self, notifier_enabled):
+ """Callback for inotify.
+
+ """
+ if not notifier_enabled:
+ 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
+
+ def Wait(self, timeout):
+ """Waits for a job to change.
+
+ @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
+
+ """
+ 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),
+ utils.RETRY_REMAINING_TIME, timeout,
+ wait_fn=waiter.Wait)
+ finally:
+ waiter.Close()
+ except (errors.InotifyError, errors.JobLost):
+ return None
+ except utils.RetryTimeout:
+ return constants.JOB_NOTCHANGED
+
+
+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)
+
+
+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
+
+
+class _OpExecContext:
+ def __init__(self, op, index, log_prefix, timeout_strategy_factory):
+ """Initializes this class.
+
+ """
+ self.op = op
+ self.index = index
+ self.log_prefix = log_prefix
+ self.summary = op.input.Summary()
+
+ 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()
+
+
+class _JobProcessor(object):
+ def __init__(self, queue, opexec_fn, job,
+ _timeout_strategy_factory=mcpu.LockAttemptTimeoutStrategy):
+ """Initializes this class.
+
+ """
+ self.queue = queue
+ self.opexec_fn = opexec_fn
+ self.job = job
+ self._timeout_strategy_factory = _timeout_strategy_factory
+
+ @staticmethod
+ def _FindNextOpcode(job, timeout_strategy_factory):
+ """Locates the next opcode to run.
+
+ @type job: L{_QueuedJob}
+ @param job: Job object
+ @param timeout_strategy_factory: Callable to create new timeout strategy
+
+ """
+ # 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.
+ if job.ops_iter is None:
+ job.ops_iter = enumerate(job.ops)
+
+ # Find next opcode to run
+ while True:
+ try:
+ (idx, op) = job.ops_iter.next()
+ 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")
+
+ opctx = _OpExecContext(op, idx, "Op %s/%s" % (idx + 1, len(job.ops)),
+ timeout_strategy_factory)
+
+ if op.status == constants.OP_STATUS_CANCELED:
+ # Cancelled jobs are handled by the caller
+ assert not compat.any(i.status != constants.OP_STATUS_CANCELED
+ for i in job.ops[idx:])
+
+ elif op.status in constants.OPS_FINALIZED:
+ # 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)
+ continue
+
+ return opctx
+
+ @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
+ @type op: L{_QueuedOpCode}
+ @param op: Opcode object
+
+ """
+ assert op in job.ops
+ assert op.status in (constants.OP_STATUS_QUEUED,
+ constants.OP_STATUS_WAITLOCK)
+
+ update = False
+
+ op.result = None
+
+ 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
+
+ if job.start_timestamp is None:
+ job.start_timestamp = op.start_timestamp
+ update = True
+
+ assert op.status == constants.OP_STATUS_WAITLOCK
+
+ return update
+
+ def _ExecOpCodeUnlocked(self, opctx):
+ """Processes one opcode and returns the result.
+
+ """
+ op = opctx.op
+
+ assert op.status == constants.OP_STATUS_WAITLOCK
+
+ timeout = opctx.GetNextLockTimeout()
projects / ganeti-local / blobdiff