def testCancelWhileInQueue(self):
queue = _FakeQueueForProc()
ops = [opcodes.OpTestDummy(result="Res%s" % i, fail=False) for i in range(5)]
# Create job
job_id = 17045
job = self._CreateJob(queue, job_id, ops)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
# Mark as cancelled
(success, _) = job.Cancel()
self.assert_(success)
self.assert_(compat.all(op.status == constants.OP_STATUS_CANCELED for op in job.ops))
opexec = _FakeExecOpCodeForProc(None, None)
jqueue._JobProcessor(queue, opexec, job)()
# Check result
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_CANCELED)
self.assertEqual(job.GetInfo(["status"]), [constants.JOB_STATUS_CANCELED])
self.assertFalse(job.start_timestamp)
self.assert_(job.end_timestamp)
self.assertFalse(compat.any(op.start_timestamp or op.end_timestamp for op in job.ops))
self.assertEqual(
job.GetInfo(["opstatus", "opresult"]),
[[constants.OP_STATUS_CANCELED for _ in job.ops], ["Job canceled by request" for _ in job.ops]],
)
def _TestPartial(self, job, successcount):
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
self.assertEqual(job.start_timestamp, job.ops[0].start_timestamp)
queue = _FakeQueueForProc()
opexec = _FakeExecOpCodeForProc(None, None)
for remaining in reversed(range(len(job.ops) - successcount)):
result = jqueue._JobProcessor(queue, opexec, job)()
if remaining == 0:
# Last opcode
self.assert_(result)
break
self.assertFalse(result)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_SUCCESS)
self.assertEqual(job.GetInfo(["status"]), [constants.JOB_STATUS_SUCCESS])
self.assertEqual(job.GetInfo(["opresult"]), [[op.input.result for op in job.ops]])
self.assertEqual(job.GetInfo(["opstatus"]), [[constants.OP_STATUS_SUCCESS for _ in job.ops]])
self.assert_(compat.all(op.start_timestamp and op.end_timestamp for op in job.ops))
self._GenericCheckJob(job)
# Finished jobs can't be processed any further
self.assertRaises(errors.ProgrammerError, jqueue._JobProcessor(queue, opexec, job))
# ... also after being restored
job2 = jqueue._QueuedJob.Restore(queue, job.Serialize())
self.assertRaises(errors.ProgrammerError, jqueue._JobProcessor(queue, opexec, job2))
def testOpcodeError(self):
queue = _FakeQueueForProc()
testdata = [(17077, 1, 0, 0), (1782, 5, 2, 2), (18179, 10, 9, 9), (4744, 10, 3, 8), (23816, 100, 39, 45)]
for (job_id, opcount, failfrom, failto) in testdata:
# Prepare opcodes
ops = [
opcodes.OpTestDummy(result="Res%s" % i, fail=(failfrom <= i and i <= failto)) for i in range(opcount)
]
# Create job
job = self._CreateJob(queue, job_id, ops)
opexec = _FakeExecOpCodeForProc(None, None)
for idx in range(len(ops)):
result = jqueue._JobProcessor(queue, opexec, job)()
if idx in (failfrom, len(ops) - 1):
# Last opcode
self.assert_(result)
break
self.assertFalse(result)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
# Check job status
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_ERROR)
self.assertEqual(job.GetInfo(["id"]), [job_id])
self.assertEqual(job.GetInfo(["status"]), [constants.JOB_STATUS_ERROR])
# Check opcode status
data = zip(job.ops, job.GetInfo(["opstatus"])[0], job.GetInfo(["opresult"])[0])
for idx, (op, opstatus, opresult) in enumerate(data):
if idx < failfrom:
assert not op.input.fail
self.assertEqual(opstatus, constants.OP_STATUS_SUCCESS)
self.assertEqual(opresult, op.input.result)
elif idx <= failto:
assert op.input.fail
self.assertEqual(opstatus, constants.OP_STATUS_ERROR)
self.assertRaises(errors.OpExecError, errors.MaybeRaise, opresult)
else:
assert not op.input.fail
self.assertEqual(opstatus, constants.OP_STATUS_ERROR)
self.assertRaises(errors.OpExecError, errors.MaybeRaise, opresult)
self.assert_(compat.all(op.start_timestamp and op.end_timestamp for op in job.ops[:failfrom]))
self._GenericCheckJob(job)
# Finished jobs can't be processed any further
self.assertRaises(errors.ProgrammerError, jqueue._JobProcessor(queue, opexec, job))
def testTimeout(self):
ops = [opcodes.OpTestDummy(result="Res%s" % i, fail=False) for i in range(10)]
# Create job
job_id = 15801
job = self._CreateJob(self.queue, job_id, ops)
self.job = job
self.opcounter = itertools.count(0)
opexec = _FakeExecOpCodeForProc(self._BeforeStart, self._AfterStart)
tsf = self._NewTimeoutStrategy
self.assertFalse(self.done_lock_before_blocking)
for i in itertools.count(0):
proc = jqueue._JobProcessor(self.queue, opexec, job, _timeout_strategy_factory=tsf)
result = proc(_nextop_fn=self._NextOpcode)
if result:
self.assertFalse(job.cur_opctx)
break
self.assertFalse(result)
if self.gave_lock:
self.assertFalse(job.cur_opctx)
else:
self.assert_(job.cur_opctx)
self.assertEqual(job.cur_opctx._timeout_strategy._fn, self.timeout_strategy.NextAttempt)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
self.assert_(job.start_timestamp)
self.assertFalse(job.end_timestamp)
self.assertEqual(self.curop, len(job.ops) - 1)
self.assertEqual(self.job, job)
self.assertEqual(self.opcounter.next(), len(job.ops))
self.assert_(self.done_lock_before_blocking)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_SUCCESS)
self.assertEqual(job.GetInfo(["status"]), [constants.JOB_STATUS_SUCCESS])
self.assertEqual(job.GetInfo(["opresult"]), [[op.input.result for op in job.ops]])
self.assertEqual(job.GetInfo(["opstatus"]), [len(job.ops) * [constants.OP_STATUS_SUCCESS]])
self.assert_(compat.all(op.start_timestamp and op.end_timestamp for op in job.ops))
# Finished jobs can't be processed any further
self.assertRaises(errors.ProgrammerError, jqueue._JobProcessor(self.queue, opexec, job))
def testPartiallyRun(self):
# Tests calling the processor on a job that's been partially run before the
# program was restarted
queue = _FakeQueueForProc()
opexec = _FakeExecOpCodeForProc(queue, None, None)
for job_id, successcount in [(30697, 1), (2552, 4), (12489, 9)]:
ops = [opcodes.OpTestDummy(result="Res%s" % i, fail=False)
for i in range(10)]
# Create job
job = self._CreateJob(queue, job_id, ops)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
for _ in range(successcount):
self.assertFalse(jqueue._JobProcessor(queue, opexec, job)())
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
self.assertEqual(job.GetInfo(["opstatus"]),
[[constants.OP_STATUS_SUCCESS
for _ in range(successcount)] +
[constants.OP_STATUS_QUEUED
for _ in range(len(ops) - successcount)]])
self.assert_(job.ops_iter)
# Serialize and restore (simulates program restart)
newjob = jqueue._QueuedJob.Restore(queue, job.Serialize())
self.assertFalse(newjob.ops_iter)
self._TestPartial(newjob, successcount)
def testCancelWhileRunning(self):
# Tests canceling a job with finished opcodes and more, unprocessed ones
queue = _FakeQueueForProc()
ops = [opcodes.OpTestDummy(result="Res%s" % i, fail=False) for i in range(3)]
# Create job
job_id = 28492
job = self._CreateJob(queue, job_id, ops)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
opexec = _FakeExecOpCodeForProc(None, None)
# Run one opcode
self.assertFalse(jqueue._JobProcessor(queue, opexec, job)())
# Job goes back to queued
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
self.assertEqual(
job.GetInfo(["opstatus", "opresult"]),
[
[constants.OP_STATUS_SUCCESS, constants.OP_STATUS_QUEUED, constants.OP_STATUS_QUEUED],
["Res0", None, None],
],
)
# Mark as cancelled
(success, _) = job.Cancel()
self.assert_(success)
# Try processing another opcode (this will actually cancel the job)
self.assert_(jqueue._JobProcessor(queue, opexec, job)())
# Check result
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_CANCELED)
self.assertEqual(job.GetInfo(["id"]), [job_id])
self.assertEqual(job.GetInfo(["status"]), [constants.JOB_STATUS_CANCELED])
self.assertEqual(
job.GetInfo(["opstatus", "opresult"]),
[
[constants.OP_STATUS_SUCCESS, constants.OP_STATUS_CANCELED, constants.OP_STATUS_CANCELED],
["Res0", "Job canceled by request", "Job canceled by request"],
],
)
def testSuccess(self):
queue = _FakeQueueForProc()
for (job_id, opcount) in [(25351, 1), (6637, 3), (24644, 10), (32207, 100)]:
ops = [opcodes.OpTestDummy(result="Res%s" % i, fail=False) for i in range(opcount)]
# Create job
job = self._CreateJob(queue, job_id, ops)
def _BeforeStart(timeout, priority):
self.assertFalse(queue.IsAcquired())
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_WAITLOCK)
def _AfterStart(op, cbs):
self.assertFalse(queue.IsAcquired())
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_RUNNING)
# Job is running, cancelling shouldn't be possible
(success, _) = job.Cancel()
self.assertFalse(success)
opexec = _FakeExecOpCodeForProc(_BeforeStart, _AfterStart)
for idx in range(len(ops)):
result = jqueue._JobProcessor(queue, opexec, job)()
if idx == len(ops) - 1:
# Last opcode
self.assert_(result)
else:
self.assertFalse(result)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
self.assert_(job.start_timestamp)
self.assertFalse(job.end_timestamp)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_SUCCESS)
self.assertEqual(job.GetInfo(["status"]), [constants.JOB_STATUS_SUCCESS])
self.assertEqual(job.GetInfo(["opresult"]), [[op.input.result for op in job.ops]])
self.assertEqual(job.GetInfo(["opstatus"]), [len(job.ops) * [constants.OP_STATUS_SUCCESS]])
self.assert_(compat.all(op.start_timestamp and op.end_timestamp for op in job.ops))
self._GenericCheckJob(job)
# Finished jobs can't be processed any further
self.assertRaises(errors.ProgrammerError, jqueue._JobProcessor(queue, opexec, job))
def testCancelWhileWaitlock(self):
queue = _FakeQueueForProc()
ops = [opcodes.OpTestDummy(result="Res%s" % i, fail=False) for i in range(5)]
# Create job
job_id = 11009
job = self._CreateJob(queue, job_id, ops)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
def _BeforeStart(timeout, priority):
self.assertFalse(queue.IsAcquired())
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_WAITLOCK)
# Mark as cancelled
(success, _) = job.Cancel()
self.assert_(success)
self.assert_(compat.all(op.status == constants.OP_STATUS_CANCELING for op in job.ops))
def _AfterStart(op, cbs):
self.assertFalse(queue.IsAcquired())
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_RUNNING)
opexec = _FakeExecOpCodeForProc(_BeforeStart, _AfterStart)
jqueue._JobProcessor(queue, opexec, job)()
# Check result
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_CANCELED)
self.assertEqual(job.GetInfo(["status"]), [constants.JOB_STATUS_CANCELED])
self.assert_(job.start_timestamp)
self.assert_(job.end_timestamp)
self.assertFalse(compat.all(op.start_timestamp and op.end_timestamp for op in job.ops))
self.assertEqual(
job.GetInfo(["opstatus", "opresult"]),
[[constants.OP_STATUS_CANCELED for _ in job.ops], ["Job canceled by request" for _ in job.ops]],
)
def testCancelWhileInQueue(self):
queue = _FakeQueueForProc()
ops = [opcodes.OpTestDummy(result="Res%s" % i, fail=False)
for i in range(5)]
# Create job
job_id = 17045
job = self._CreateJob(queue, job_id, ops)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
# Mark as cancelled
(success, _) = job.Cancel()
self.assert_(success)
self.assertRaises(IndexError, queue.GetNextUpdate)
self.assertFalse(job.start_timestamp)
self.assertTrue(job.end_timestamp)
self.assert_(compat.all(op.status == constants.OP_STATUS_CANCELED
for op in job.ops))
# Serialize to check for differences
before_proc = job.Serialize()
# Simulate processor called in workerpool
opexec = _FakeExecOpCodeForProc(queue, None, None)
self.assert_(jqueue._JobProcessor(queue, opexec, job)())
# Check result
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_CANCELED)
self.assertEqual(job.GetInfo(["status"]), [constants.JOB_STATUS_CANCELED])
self.assertFalse(job.start_timestamp)
self.assertTrue(job.end_timestamp)
self.assertFalse(compat.any(op.start_timestamp or op.end_timestamp
for op in job.ops))
self.assertEqual(job.GetInfo(["opstatus", "opresult"]),
[[constants.OP_STATUS_CANCELED for _ in job.ops],
["Job canceled by request" for _ in job.ops]])
# Must not have changed or written
self.assertEqual(before_proc, job.Serialize())
self.assertRaises(IndexError, queue.GetNextUpdate)
def testProcessorOnRunningJob(self):
ops = [opcodes.OpTestDummy(result="result", fail=False)]
queue = _FakeQueueForProc()
opexec = _FakeExecOpCodeForProc(None, None)
# Create job
job = self._CreateJob(queue, 9571, ops)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
job.ops[0].status = constants.OP_STATUS_RUNNING
assert len(job.ops) == 1
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_RUNNING)
# Calling on running job must fail
self.assertRaises(errors.ProgrammerError, jqueue._JobProcessor(queue, opexec, job))
def testCancelWhileWaitlockInQueue(self):
queue = _FakeQueueForProc()
ops = [opcodes.OpTestDummy(result="Res%s" % i, fail=False)
for i in range(5)]
# Create job
job_id = 8645
job = self._CreateJob(queue, job_id, ops)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
job.ops[0].status = constants.OP_STATUS_WAITLOCK
assert len(job.ops) == 5
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_WAITLOCK)
# Mark as cancelling
(success, _) = job.Cancel()
self.assert_(success)
self.assertRaises(IndexError, queue.GetNextUpdate)
self.assert_(compat.all(op.status == constants.OP_STATUS_CANCELING
for op in job.ops))
opexec = _FakeExecOpCodeForProc(queue, None, None)
self.assert_(jqueue._JobProcessor(queue, opexec, job)())
# Check result
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_CANCELED)
self.assertEqual(job.GetInfo(["status"]), [constants.JOB_STATUS_CANCELED])
self.assertFalse(job.start_timestamp)
self.assert_(job.end_timestamp)
self.assertFalse(compat.any(op.start_timestamp or op.end_timestamp
for op in job.ops))
self.assertEqual(job.GetInfo(["opstatus", "opresult"]),
[[constants.OP_STATUS_CANCELED for _ in job.ops],
["Job canceled by request" for _ in job.ops]])
def main():
debug = int(os.environ["GNT_DEBUG"])
logname = pathutils.GetLogFilename("jobs")
utils.SetupLogging(logname, "job-startup", debug=debug)
(job_id, llock, secret_params_serialized) = _SetupJob()
secret_params = ""
if secret_params_serialized:
secret_params_json = serializer.LoadJson(secret_params_serialized)
secret_params = RestorePrivateValueWrapping(secret_params_json)
utils.SetupLogging(logname, "job-%s" % (job_id,), debug=debug)
try:
logging.debug("Preparing the context and the configuration")
context = masterd.GanetiContext(llock)
logging.debug("Registering signal handlers")
cancel = [False]
prio_change = [False]
def _TermHandler(signum, _frame):
logging.info("Killed by signal %d", signum)
cancel[0] = True
signal.signal(signal.SIGTERM, _TermHandler)
def _HupHandler(signum, _frame):
logging.debug("Received signal %d, old flag was %s, will set to True",
signum, mcpu.sighupReceived)
mcpu.sighupReceived[0] = True
signal.signal(signal.SIGHUP, _HupHandler)
def _User1Handler(signum, _frame):
logging.info("Received signal %d, indicating priority change", signum)
prio_change[0] = True
signal.signal(signal.SIGUSR1, _User1Handler)
job = context.jobqueue.SafeLoadJobFromDisk(job_id, False)
job.SetPid(os.getpid())
if secret_params:
for i in range(0, len(secret_params)):
if hasattr(job.ops[i].input, "osparams_secret"):
job.ops[i].input.osparams_secret = secret_params[i]
execfun = mcpu.Processor(context, job_id, job_id).ExecOpCode
proc = _JobProcessor(context.jobqueue, execfun, job)
result = _JobProcessor.DEFER
while result != _JobProcessor.FINISHED:
result = proc()
if result == _JobProcessor.WAITDEP and not cancel[0]:
# Normally, the scheduler should avoid starting a job where the
# dependencies are not yet finalised. So warn, but wait an continue.
logging.warning("Got started despite a dependency not yet finished")
time.sleep(5)
if cancel[0]:
logging.debug("Got cancel request, cancelling job %d", job_id)
r = context.jobqueue.CancelJob(job_id)
job = context.jobqueue.SafeLoadJobFromDisk(job_id, False)
proc = _JobProcessor(context.jobqueue, execfun, job)
logging.debug("CancelJob result for job %d: %s", job_id, r)
cancel[0] = False
if prio_change[0]:
logging.debug("Received priority-change request")
try:
fname = os.path.join(pathutils.LUXID_MESSAGE_DIR, "%d.prio" % job_id)
new_prio = int(utils.ReadFile(fname))
utils.RemoveFile(fname)
logging.debug("Changing priority of job %d to %d", job_id, new_prio)
r = context.jobqueue.ChangeJobPriority(job_id, new_prio)
job = context.jobqueue.SafeLoadJobFromDisk(job_id, False)
proc = _JobProcessor(context.jobqueue, execfun, job)
logging.debug("Result of changing priority of %d to %d: %s", job_id,
new_prio, r)
except Exception: # pylint: disable=W0703
logging.warning("Informed of priority change, but could not"
" read new priority")
prio_change[0] = False
except Exception: # pylint: disable=W0703
logging.exception("Exception when trying to run job %d", job_id)
finally:
logging.debug("Job %d finalized", job_id)
logging.debug("Removing livelock file %s", llock.GetPath())
os.remove(llock.GetPath())
sys.exit(0)
def testLogMessages(self):
# Tests the "Feedback" callback function
queue = _FakeQueueForProc()
messages = {
1: [
(None, "Hello"),
(None, "World"),
(constants.ELOG_MESSAGE, "there"),
],
4: [
(constants.ELOG_JQUEUE_TEST, (1, 2, 3)),
(constants.ELOG_JQUEUE_TEST, ("other", "type")),
],
}
ops = [opcodes.OpTestDummy(result="Logtest%s" % i, fail=False,
messages=messages.get(i, []))
for i in range(5)]
# Create job
job = self._CreateJob(queue, 29386, ops)
def _BeforeStart(timeout, priority):
self.assertEqual(queue.GetNextUpdate(), (job, True))
self.assertRaises(IndexError, queue.GetNextUpdate)
self.assertFalse(queue.IsAcquired())
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_WAITLOCK)
def _AfterStart(op, cbs):
self.assertEqual(queue.GetNextUpdate(), (job, True))
self.assertRaises(IndexError, queue.GetNextUpdate)
self.assertFalse(queue.IsAcquired())
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_RUNNING)
self.assertRaises(AssertionError, cbs.Feedback,
"too", "many", "arguments")
for (log_type, msg) in op.messages:
self.assertRaises(IndexError, queue.GetNextUpdate)
if log_type:
cbs.Feedback(log_type, msg)
else:
cbs.Feedback(msg)
# Check for job update without replication
self.assertEqual(queue.GetNextUpdate(), (job, False))
self.assertRaises(IndexError, queue.GetNextUpdate)
opexec = _FakeExecOpCodeForProc(queue, _BeforeStart, _AfterStart)
for remaining in reversed(range(len(job.ops))):
self.assertRaises(IndexError, queue.GetNextUpdate)
result = jqueue._JobProcessor(queue, opexec, job)()
self.assertEqual(queue.GetNextUpdate(), (job, True))
self.assertRaises(IndexError, queue.GetNextUpdate)
if remaining == 0:
# Last opcode
self.assert_(result)
break
self.assertFalse(result)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
self.assertRaises(IndexError, queue.GetNextUpdate)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_SUCCESS)
self.assertEqual(job.GetInfo(["opresult"]),
[[op.input.result for op in job.ops]])
logmsgcount = sum(len(m) for m in messages.values())
self._CheckLogMessages(job, logmsgcount)
# Serialize and restore (simulates program restart)
newjob = jqueue._QueuedJob.Restore(queue, job.Serialize())
self._CheckLogMessages(newjob, logmsgcount)
# Check each message
prevserial = -1
for idx, oplog in enumerate(job.GetInfo(["oplog"])[0]):
for (serial, timestamp, log_type, msg) in oplog:
(exptype, expmsg) = messages.get(idx).pop(0)
if exptype:
self.assertEqual(log_type, exptype)
else:
self.assertEqual(log_type, constants.ELOG_MESSAGE)
self.assertEqual(expmsg, msg)
self.assert_(serial > prevserial)
prevserial = serial
def testTimeout(self):
ops = [opcodes.OpTestDummy(result="Res%s" % i, fail=False)
for i in range(10)]
# Create job
job_id = 15801
job = self._CreateJob(self.queue, job_id, ops)
self.job = job
self.opcounter = itertools.count(0)
opexec = _FakeExecOpCodeForProc(self.queue, self._BeforeStart,
self._AfterStart)
tsf = self._NewTimeoutStrategy
self.assertFalse(self.done_lock_before_blocking)
while True:
proc = jqueue._JobProcessor(self.queue, opexec, job,
_timeout_strategy_factory=tsf)
self.assertRaises(IndexError, self.queue.GetNextUpdate)
if self.curop is not None:
self.prev_status = self.job.ops[self.curop].status
self.lock_acq_prio = None
result = proc(_nextop_fn=self._NextOpcode)
assert self.curop is not None
if result or self.gave_lock:
# Got lock and/or job is done, result must've been written
self.assertFalse(job.cur_opctx)
self.assertEqual(self.queue.GetNextUpdate(), (job, True))
self.assertRaises(IndexError, self.queue.GetNextUpdate)
self.assertEqual(self.lock_acq_prio, job.ops[self.curop].priority)
self.assert_(job.ops[self.curop].exec_timestamp)
if result:
self.assertFalse(job.cur_opctx)
break
self.assertFalse(result)
if self.curop == 0:
self.assertEqual(job.ops[self.curop].start_timestamp,
job.start_timestamp)
if self.gave_lock:
# Opcode finished, but job not yet done
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
else:
# Did not get locks
self.assert_(job.cur_opctx)
self.assertEqual(job.cur_opctx._timeout_strategy._fn,
self.timeout_strategy.NextAttempt)
self.assertFalse(job.ops[self.curop].exec_timestamp)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_WAITLOCK)
# If priority has changed since acquiring locks, the job must've been
# updated
if self.lock_acq_prio != job.ops[self.curop].priority:
self.assertEqual(self.queue.GetNextUpdate(), (job, True))
self.assertRaises(IndexError, self.queue.GetNextUpdate)
self.assert_(job.start_timestamp)
self.assertFalse(job.end_timestamp)
self.assertEqual(self.curop, len(job.ops) - 1)
self.assertEqual(self.job, job)
self.assertEqual(self.opcounter.next(), len(job.ops))
self.assert_(self.done_lock_before_blocking)
self.assertRaises(IndexError, self.queue.GetNextUpdate)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_SUCCESS)
self.assertEqual(job.GetInfo(["status"]), [constants.JOB_STATUS_SUCCESS])
self.assertEqual(job.GetInfo(["opresult"]),
[[op.input.result for op in job.ops]])
self.assertEqual(job.GetInfo(["opstatus"]),
[len(job.ops) * [constants.OP_STATUS_SUCCESS]])
self.assert_(compat.all(op.start_timestamp and op.end_timestamp
for op in job.ops))
# Calling the processor on a finished job should be a no-op
self.assertTrue(jqueue._JobProcessor(self.queue, opexec, job)())
self.assertRaises(IndexError, self.queue.GetNextUpdate)
def testSubmitManyJobs(self):
queue = _FakeQueueForProc()
job_id = 15656
ops = [
opcodes.OpTestDummy(result="Res0", fail=False,
submit_jobs=[]),
opcodes.OpTestDummy(result="Res1", fail=False,
submit_jobs=[
[opcodes.OpTestDummy(result="r1j0", fail=False)],
]),
opcodes.OpTestDummy(result="Res2", fail=False,
submit_jobs=[
[opcodes.OpTestDummy(result="r2j0o0", fail=False),
opcodes.OpTestDummy(result="r2j0o1", fail=False),
opcodes.OpTestDummy(result="r2j0o2", fail=False),
opcodes.OpTestDummy(result="r2j0o3", fail=False)],
[opcodes.OpTestDummy(result="r2j1", fail=False)],
[opcodes.OpTestDummy(result="r2j3o0", fail=False),
opcodes.OpTestDummy(result="r2j3o1", fail=False)],
]),
]
# Create job
job = self._CreateJob(queue, job_id, ops)
def _BeforeStart(timeout, priority):
self.assertEqual(queue.GetNextUpdate(), (job, True))
self.assertRaises(IndexError, queue.GetNextUpdate)
self.assertFalse(queue.IsAcquired())
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_WAITLOCK)
self.assertFalse(job.cur_opctx)
def _AfterStart(op, cbs):
self.assertEqual(queue.GetNextUpdate(), (job, True))
self.assertRaises(IndexError, queue.GetNextUpdate)
self.assertFalse(queue.IsAcquired())
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_RUNNING)
self.assertFalse(job.cur_opctx)
# Job is running, cancelling shouldn't be possible
(success, _) = job.Cancel()
self.assertFalse(success)
opexec = _FakeExecOpCodeForProc(queue, _BeforeStart, _AfterStart)
for idx in range(len(ops)):
self.assertRaises(IndexError, queue.GetNextUpdate)
result = jqueue._JobProcessor(queue, opexec, job)()
self.assertEqual(queue.GetNextUpdate(), (job, True))
self.assertRaises(IndexError, queue.GetNextUpdate)
if idx == len(ops) - 1:
# Last opcode
self.assert_(result)
else:
self.assertFalse(result)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_QUEUED)
self.assert_(job.start_timestamp)
self.assertFalse(job.end_timestamp)
self.assertRaises(IndexError, queue.GetNextUpdate)
for idx, submitted_ops in enumerate(job_ops
for op in ops
for job_ops in op.submit_jobs):
self.assertEqual(queue.GetNextSubmittedJob(),
(1000 + idx, submitted_ops))
self.assertRaises(IndexError, queue.GetNextSubmittedJob)
self.assertEqual(job.CalcStatus(), constants.JOB_STATUS_SUCCESS)
self.assertEqual(job.GetInfo(["status"]), [constants.JOB_STATUS_SUCCESS])
self.assertEqual(job.GetInfo(["opresult"]),
[[[], [1000], [1001, 1002, 1003]]])
self.assertEqual(job.GetInfo(["opstatus"]),
[len(job.ops) * [constants.OP_STATUS_SUCCESS]])
self._GenericCheckJob(job)
# Calling the processor on a finished job should be a no-op
self.assertTrue(jqueue._JobProcessor(queue, opexec, job)())
self.assertRaises(IndexError, queue.GetNextUpdate)
def main():
debug = int(os.environ["GNT_DEBUG"])
logname = pathutils.GetLogFilename("jobs")
utils.SetupLogging(logname, "job-startup", debug=debug)
(job_id, livelock_name, secret_params_serialized) = _GetMasterInfo()
secret_params = ""
if secret_params_serialized:
secret_params_json = serializer.LoadJson(secret_params_serialized)
secret_params = RestorePrivateValueWrapping(secret_params_json)
utils.SetupLogging(logname, "job-%s" % (job_id,), debug=debug)
try:
logging.debug("Preparing the context and the configuration")
context = masterd.GanetiContext(livelock_name)
logging.debug("Registering signal handlers")
cancel = [False]
prio_change = [False]
def _TermHandler(signum, _frame):
logging.info("Killed by signal %d", signum)
cancel[0] = True
signal.signal(signal.SIGTERM, _TermHandler)
def _HupHandler(signum, _frame):
logging.debug("Received signal %d, old flag was %s, will set to True",
signum, mcpu.sighupReceived)
mcpu.sighupReceived[0] = True
signal.signal(signal.SIGHUP, _HupHandler)
def _User1Handler(signum, _frame):
logging.info("Received signal %d, indicating priority change", signum)
prio_change[0] = True
signal.signal(signal.SIGUSR1, _User1Handler)
job = context.jobqueue.SafeLoadJobFromDisk(job_id, False)
job.SetPid(os.getpid())
if secret_params:
for i in range(0, len(secret_params)):
if hasattr(job.ops[i].input, "osparams_secret"):
job.ops[i].input.osparams_secret = secret_params[i]
execfun = mcpu.Processor(context, job_id, job_id).ExecOpCode
proc = _JobProcessor(context.jobqueue, execfun, job)
result = _JobProcessor.DEFER
while result != _JobProcessor.FINISHED:
result = proc()
if result == _JobProcessor.WAITDEP and not cancel[0]:
# Normally, the scheduler should avoid starting a job where the
# dependencies are not yet finalised. So warn, but wait an continue.
logging.warning("Got started despite a dependency not yet finished")
time.sleep(5)
if cancel[0]:
logging.debug("Got cancel request, cancelling job %d", job_id)
r = context.jobqueue.CancelJob(job_id)
job = context.jobqueue.SafeLoadJobFromDisk(job_id, False)
proc = _JobProcessor(context.jobqueue, execfun, job)
logging.debug("CancelJob result for job %d: %s", job_id, r)
cancel[0] = False
if prio_change[0]:
logging.debug("Received priority-change request")
try:
fname = os.path.join(pathutils.LUXID_MESSAGE_DIR, "%d.prio" % job_id)
new_prio = int(utils.ReadFile(fname))
utils.RemoveFile(fname)
logging.debug("Changing priority of job %d to %d", job_id, new_prio)
r = context.jobqueue.ChangeJobPriority(job_id, new_prio)
job = context.jobqueue.SafeLoadJobFromDisk(job_id, False)
proc = _JobProcessor(context.jobqueue, execfun, job)
logging.debug("Result of changing priority of %d to %d: %s", job_id,
new_prio, r)
except Exception: # pylint: disable=W0703
logging.warning("Informed of priority change, but could not"
" read new priority")
prio_change[0] = False
except Exception: # pylint: disable=W0703
logging.exception("Exception when trying to run job %d", job_id)
finally:
logging.debug("Job %d finalized", job_id)
logging.debug("Removing livelock file %s", livelock_name.GetPath())
os.remove(livelock_name.GetPath())
sys.exit(0)
