def testJobDescription(self):
"""
Tests the public interface of a JobDescription.
"""
command = "by your command"
memory = 2^32
disk = 2^32
cores = "1"
preemptable = 1
j = JobDescription(command=command, requirements={"memory": memory, "cores": cores, "disk": disk, "preemptable": preemptable},
jobName='testJobGraph', unitName='noName')
#Check attributes
self.assertEqual(j.command, command)
self.assertEqual(j.memory, memory)
self.assertEqual(j.disk, disk)
self.assertEqual(j.cores, int(cores))
self.assertEqual(j.preemptable, bool(preemptable))
self.assertEqual(type(j.jobStoreID), TemporaryID)
self.assertEqual(list(j.successorsAndServiceHosts()), [])
self.assertEqual(list(j.allSuccessors()), [])
self.assertEqual(list(j.serviceHostIDsInBatches()), [])
self.assertEqual(list(j.services), [])
self.assertEqual(list(j.nextSuccessors()), [])
self.assertEqual(sum((len(level) for level in j.stack)), 0)
self.assertEqual(j.predecessorsFinished, set())
self.assertEqual(j.logJobStoreFileID, None)
#Check equals function (should be based on object identity and not contents)
j2 = JobDescription(command=command, requirements={"memory": memory, "cores": cores, "disk": disk, "preemptable": preemptable},
jobName='testJobGraph', unitName='noName')
self.assertNotEqual(j, j2)
def testJobDescriptionSequencing(self):
j = JobDescription(command='command', requirements={}, jobName='unimportant')
j.addChild('child')
j.addFollowOn('followOn')
# With a command, nothing should be ready to run
self.assertEqual(list(j.nextSuccessors()), [])
# With command cleared, child should be ready to run
j.command = None
self.assertEqual(list(j.nextSuccessors()), ['child'])
# Without the child, the follow-on should be ready to run
j.filterSuccessors(lambda jID: jID != 'child')
self.assertEqual(list(j.nextSuccessors()), ['followOn'])
# Without the follow-on, we should return None, to be distinct from an
# empty list. Nothing left to do!
j.filterSuccessors(lambda jID: jID != 'followOn')
self.assertEqual(j.nextSuccessors(), None)
def nextChainable(predecessor: JobDescription, jobStore: AbstractJobStore,
config: Config) -> Optional[JobDescription]:
"""
Returns the next chainable job's JobDescription after the given predecessor
JobDescription, if one exists, or None if the chain must terminate.
:param predecessor: The job to chain from
:param jobStore: The JobStore to fetch JobDescriptions from.
:param config: The configuration for the current run.
"""
#If no more jobs to run or services not finished, quit
if len(predecessor.stack) == 0 or len(predecessor.services) > 0 or (
isinstance(predecessor, CheckpointJobDescription)
and predecessor.checkpoint != None):
logger.debug(
"Stopping running chain of jobs: length of stack: %s, services: %s, checkpoint: %s",
len(predecessor.stack), len(predecessor.services),
(isinstance(predecessor, CheckpointJobDescription)
and predecessor.checkpoint != None))
return None
if len(predecessor.stack) > 1 and len(predecessor.stack[-1]) > 0 and len(
predecessor.stack[-2]) > 0:
# TODO: Without a real stack list we can freely mutate, we can't chain
# to a child, which may branch, and then go back and do the follow-ons
# of the original job.
# TODO: Go back to a free-form stack list and require some kind of
# stack build phase?
logger.debug(
"Stopping running chain of jobs because job has both children and follow-ons"
)
return None
#Get the next set of jobs to run
jobs = predecessor.nextSuccessors()
if len(jobs) == 0:
# If there are no jobs, we might just not have any children.
logger.debug(
"Stopping running chain of jobs because job has no ready children or follow-ons"
)
return None
#If there are 2 or more jobs to run in parallel we quit
if len(jobs) >= 2:
logger.debug(
"No more jobs can run in series by this worker,"
" it's got %i children",
len(jobs) - 1)
return None
# Grab the only job that should be there.
successorID = next(iter(jobs))
# Load the successor JobDescription
successor = jobStore.load(successorID)
#We check the requirements of the successor to see if we can run it
#within the current worker
if successor.memory > predecessor.memory:
logger.debug("We need more memory for the next job, so finishing")
return None
if successor.cores > predecessor.cores:
logger.debug("We need more cores for the next job, so finishing")
return None
if successor.disk > predecessor.disk:
logger.debug("We need more disk for the next job, so finishing")
return None
if successor.preemptable != predecessor.preemptable:
logger.debug(
"Preemptability is different for the next job, returning to the leader"
)
return None
if successor.predecessorNumber > 1:
logger.debug(
"The next job has multiple predecessors; we must return to the leader."
)
return None
if len(successor.services) > 0:
logger.debug(
"The next job requires services that will not yet be started; we must return to the leader."
)
return None
if isinstance(successor, CheckpointJobDescription):
# Check if job is a checkpoint job and quit if so
logger.debug("Next job is checkpoint, so finishing")
return None
# Made it through! This job is chainable.
return successor
def _buildToilState(self, jobDesc: JobDescription) -> None:
"""
Traverses tree of jobs down from the subtree root JobDescription
(jobDesc), building the ToilState class.
:param jobDesc: The description for the root job of the workflow being run.
"""
# If the job description has a command, is a checkpoint, has services
# or is ready to be deleted it is ready to be processed (i.e. it is updated)
if (jobDesc.command is not None
or (isinstance(jobDesc, CheckpointJobDescription)
and jobDesc.checkpoint is not None)
or len(jobDesc.services) > 0
or jobDesc.nextSuccessors() is None):
logger.debug(
"Found job to run: %s, with command: %s, with checkpoint: %s, with "
"services: %s, with no next successors: %s",
jobDesc.jobStoreID,
jobDesc.command is not None,
isinstance(jobDesc, CheckpointJobDescription)
and jobDesc.checkpoint is not None,
len(jobDesc.services) > 0,
jobDesc.nextSuccessors() is None,
)
# Set the job updated because we should be able to make progress on it.
self.bus.put(JobUpdatedMessage(str(jobDesc.jobStoreID), 0))
if isinstance(jobDesc, CheckpointJobDescription
) and jobDesc.checkpoint is not None:
jobDesc.command = jobDesc.checkpoint
else: # There exist successors
logger.debug(
"Adding job: %s to the state with %s successors",
jobDesc.jobStoreID,
len(jobDesc.nextSuccessors()),
)
# Record the number of successors
self.successorCounts[str(jobDesc.jobStoreID)] = len(
jobDesc.nextSuccessors())
def processSuccessorWithMultiplePredecessors(
successor: JobDescription) -> None:
# If jobDesc is not reported as complete by the successor
if jobDesc.jobStoreID not in successor.predecessorsFinished:
# Update the successor's status to mark the predecessor complete
successor.predecessorsFinished.add(jobDesc.jobStoreID)
# If the successor has no predecessors to finish
assert len(successor.predecessorsFinished
) <= successor.predecessorNumber
if len(successor.predecessorsFinished
) == successor.predecessorNumber:
# It is ready to be run, so remove it from the set of waiting jobs
self.jobsToBeScheduledWithMultiplePredecessors.remove(
successorJobStoreID)
# Recursively consider the successor
self._buildToilState(successor)
# For each successor
for successorJobStoreID in jobDesc.nextSuccessors():
# If the successor does not yet point back at a
# predecessor we have not yet considered it
if successorJobStoreID not in self.successor_to_predecessors:
# Add the job as a predecessor
self.successor_to_predecessors[successorJobStoreID] = {
str(jobDesc.jobStoreID)
}
# We load the successor job
successor = self.get_job(successorJobStoreID)
# If predecessor number > 1 then the successor has multiple predecessors
if successor.predecessorNumber > 1:
# We put the successor job in the set of waiting successor
# jobs with multiple predecessors
assert successorJobStoreID not in self.jobsToBeScheduledWithMultiplePredecessors
self.jobsToBeScheduledWithMultiplePredecessors.add(
successorJobStoreID)
# Process successor
processSuccessorWithMultiplePredecessors(successor)
else:
# The successor has only this job as a predecessor so
# recursively consider the successor
self._buildToilState(successor)
else:
# We've already seen the successor
# Add the job as a predecessor
assert (jobDesc.jobStoreID not in self.
successor_to_predecessors[successorJobStoreID])
self.successor_to_predecessors[successorJobStoreID].add(
str(jobDesc.jobStoreID))
# If the successor has multiple predecessors
if successorJobStoreID in self.jobsToBeScheduledWithMultiplePredecessors:
# Get the successor from cache
successor = self.get_job(successorJobStoreID)
# Process successor
processSuccessorWithMultiplePredecessors(successor)
