def main():
logging.basicConfig()
##########################################
#Import necessary modules
##########################################
# This is assuming that worker.py is at a path ending in "/toil/worker.py".
sourcePath = os.path.dirname(os.path.dirname(__file__))
if sourcePath not in sys.path:
sys.path.append(sourcePath)
#Now we can import all the necessary functions
from toil.lib.bioio import setLogLevel
from toil.lib.bioio import getTotalCpuTime
from toil.lib.bioio import getTotalCpuTimeAndMemoryUsage
from toil.job import Job
try:
import boto
except ImportError:
pass
else:
# boto is installed, monkey patch it now
from bd2k.util.ec2.credentials import enable_metadata_credential_caching
enable_metadata_credential_caching()
##########################################
#Input args
##########################################
jobStoreLocator = sys.argv[1]
jobStoreID = sys.argv[2]
##########################################
#Load the jobStore/config file
##########################################
jobStore = Toil.resumeJobStore(jobStoreLocator)
config = jobStore.config
##########################################
#Create the worker killer, if requested
##########################################
if config.badWorker > 0 and random.random() < config.badWorker:
def badWorker():
#This will randomly kill the worker process at a random time
time.sleep(config.badWorkerFailInterval * random.random())
os.kill(os.getpid(), signal.SIGKILL) #signal.SIGINT)
#TODO: FIX OCCASIONAL DEADLOCK WITH SIGINT (tested on single machine)
t = Thread(target=badWorker)
# Ideally this would be a daemon thread but that causes an intermittent (but benign)
# exception similar to the one described here:
# http://stackoverflow.com/questions/20596918/python-exception-in-thread-thread-1-most-likely-raised-during-interpreter-shutd
# Our exception is:
# Exception in thread Thread-1 (most likely raised during interpreter shutdown):
# <type 'exceptions.AttributeError'>: 'NoneType' object has no attribute 'kill'
# This attribute error is caused by the call os.kill() and apparently unavoidable with a
# daemon
t.start()
##########################################
#Load the environment for the jobGraph
##########################################
#First load the environment for the jobGraph.
with jobStore.readSharedFileStream("environment.pickle") as fileHandle:
environment = cPickle.load(fileHandle)
for i in environment:
if i not in ("TMPDIR", "TMP", "HOSTNAME", "HOSTTYPE"):
os.environ[i] = environment[i]
# sys.path is used by __import__ to find modules
if "PYTHONPATH" in environment:
for e in environment["PYTHONPATH"].split(':'):
if e != '':
sys.path.append(e)
setLogLevel(config.logLevel)
toilWorkflowDir = Toil.getWorkflowDir(config.workflowID, config.workDir)
##########################################
#Setup the temporary directories.
##########################################
# Dir to put all this worker's temp files in.
localWorkerTempDir = tempfile.mkdtemp(dir=toilWorkflowDir)
os.chmod(localWorkerTempDir, 0755)
##########################################
#Setup the logging
##########################################
#This is mildly tricky because we don't just want to
#redirect stdout and stderr for this Python process; we want to redirect it
#for this process and all children. Consequently, we can't just replace
#sys.stdout and sys.stderr; we need to mess with the underlying OS-level
#file descriptors. See <http://stackoverflow.com/a/11632982/402891>
#When we start, standard input is file descriptor 0, standard output is
#file descriptor 1, and standard error is file descriptor 2.
#What file do we want to point FDs 1 and 2 to?
tempWorkerLogPath = os.path.join(localWorkerTempDir, "worker_log.txt")
#Save the original stdout and stderr (by opening new file descriptors to the
#same files)
origStdOut = os.dup(1)
origStdErr = os.dup(2)
#Open the file to send stdout/stderr to.
logFh = os.open(tempWorkerLogPath, os.O_WRONLY | os.O_CREAT | os.O_APPEND)
#Replace standard output with a descriptor for the log file
os.dup2(logFh, 1)
#Replace standard error with a descriptor for the log file
os.dup2(logFh, 2)
#Since we only opened the file once, all the descriptors duped from the
#original will share offset information, and won't clobber each others'
#writes. See <http://stackoverflow.com/a/5284108/402891>. This shouldn't
#matter, since O_APPEND seeks to the end of the file before every write, but
#maybe there's something odd going on...
#Close the descriptor we used to open the file
os.close(logFh)
for handler in list(logger.handlers): #Remove old handlers
logger.removeHandler(handler)
#Add the new handler. The sys.stderr stream has been redirected by swapping
#the file descriptor out from under it.
logger.addHandler(logging.StreamHandler(sys.stderr))
debugging = logging.getLogger().isEnabledFor(logging.DEBUG)
##########################################
#Worker log file trapped from here on in
##########################################
workerFailed = False
statsDict = MagicExpando()
statsDict.jobs = []
statsDict.workers.logsToMaster = []
blockFn = lambda: True
cleanCacheFn = lambda x: True
try:
#Put a message at the top of the log, just to make sure it's working.
print "---TOIL WORKER OUTPUT LOG---"
sys.stdout.flush()
#Log the number of open file descriptors so we can tell if we're leaking
#them.
logger.debug("Next available file descriptor: {}".format(
nextOpenDescriptor()))
logProcessContext(config, logger)
##########################################
#Load the jobGraph
##########################################
jobGraph = jobStore.load(jobStoreID)
logger.debug("Parsed jobGraph")
##########################################
#Cleanup from any earlier invocation of the jobGraph
##########################################
if jobGraph.command == None:
# Cleanup jobs already finished
f = lambda jobs: filter(
lambda x: len(x) > 0,
map(
lambda x: filter(lambda y: jobStore.exists(y.jobStoreID), x
), jobs))
jobGraph.stack = f(jobGraph.stack)
jobGraph.services = f(jobGraph.services)
logger.debug(
"Cleaned up any references to completed successor jobs")
#This cleans the old log file which may
#have been left if the job is being retried after a job failure.
oldLogFile = jobGraph.logJobStoreFileID
if oldLogFile != None:
jobGraph.logJobStoreFileID = None
jobStore.update(jobGraph) #Update first, before deleting any files
jobStore.deleteFile(oldLogFile)
##########################################
# If a checkpoint exists, restart from the checkpoint
##########################################
# The job is a checkpoint, and is being restarted after previously completing
if jobGraph.checkpoint != None:
logger.debug("Job is a checkpoint")
if len(jobGraph.stack) > 0 or len(
jobGraph.services) > 0 or jobGraph.command != None:
if jobGraph.command != None:
assert jobGraph.command == jobGraph.checkpoint
logger.debug(
"Checkpoint job already has command set to run")
else:
jobGraph.command = jobGraph.checkpoint
# Reduce the retry count
assert jobGraph.remainingRetryCount >= 0
jobGraph.remainingRetryCount = max(
0, jobGraph.remainingRetryCount - 1)
jobStore.update(
jobGraph) # Update immediately to ensure that checkpoint
# is made before deleting any remaining successors
if len(jobGraph.stack) > 0 or len(jobGraph.services) > 0:
# If the subtree of successors is not complete restart everything
logger.debug(
"Checkpoint job has unfinished successor jobs, deleting the jobs on the stack: %s, services: %s "
% (jobGraph.stack, jobGraph.services))
# Delete everything on the stack, as these represent successors to clean
# up as we restart the queue
def recursiveDelete(jobGraph2):
# Recursive walk the stack to delete all remaining jobs
for jobs in jobGraph2.stack + jobGraph2.services:
for jobNode in jobs:
if jobStore.exists(jobNode.jobStoreID):
recursiveDelete(
jobStore.load(jobNode.jobStoreID))
else:
logger.debug(
"Job %s has already been deleted",
jobNode)
if jobGraph2 != jobGraph:
logger.debug(
"Checkpoint is deleting old successor job: %s",
jobGraph2.jobStoreID)
jobStore.delete(jobGraph2.jobStoreID)
recursiveDelete(jobGraph)
jobGraph.stack = [
[], []
] # Initialise the job to mimic the state of a job
# that has been previously serialised but which as yet has no successors
jobGraph.services = [] # Empty the services
# Update the jobStore to avoid doing this twice on failure and make this clean.
jobStore.update(jobGraph)
# Otherwise, the job and successors are done, and we can cleanup stuff we couldn't clean
# because of the job being a checkpoint
else:
logger.debug(
"The checkpoint jobs seems to have completed okay, removing any checkpoint files to delete."
)
#Delete any remnant files
map(
jobStore.deleteFile,
filter(jobStore.fileExists,
jobGraph.checkpointFilesToDelete))
##########################################
#Setup the stats, if requested
##########################################
if config.stats:
startTime = time.time()
startClock = getTotalCpuTime()
#Make a temporary file directory for the jobGraph
#localTempDir = makePublicDir(os.path.join(localWorkerTempDir, "localTempDir"))
startTime = time.time()
while True:
##########################################
#Run the jobGraph, if there is one
##########################################
if jobGraph.command is not None:
assert jobGraph.command.startswith("_toil ")
logger.debug("Got a command to run: %s" % jobGraph.command)
#Load the job
job = Job._loadJob(jobGraph.command, jobStore)
# If it is a checkpoint job, save the command
if job.checkpoint:
jobGraph.checkpoint = jobGraph.command
# Create a fileStore object for the job
fileStore = FileStore.createFileStore(
jobStore,
jobGraph,
localWorkerTempDir,
blockFn,
caching=not config.disableCaching)
with job._executor(jobGraph=jobGraph,
stats=statsDict if config.stats else None,
fileStore=fileStore):
with fileStore.open(job):
# Get the next block function and list that will contain any messages
blockFn = fileStore._blockFn
job._runner(jobGraph=jobGraph,
jobStore=jobStore,
fileStore=fileStore)
# Accumulate messages from this job & any subsequent chained jobs
statsDict.workers.logsToMaster += fileStore.loggingMessages
else:
#The command may be none, in which case
#the jobGraph is either a shell ready to be deleted or has
#been scheduled after a failure to cleanup
break
if FileStore._terminateEvent.isSet():
raise RuntimeError("The termination flag is set")
##########################################
#Establish if we can run another jobGraph within the worker
##########################################
#If no more jobs to run or services not finished, quit
if len(jobGraph.stack) == 0 or len(
jobGraph.services) > 0 or jobGraph.checkpoint != None:
logger.debug(
"Stopping running chain of jobs: length of stack: %s, services: %s, checkpoint: %s",
len(jobGraph.stack), len(jobGraph.services),
jobGraph.checkpoint != None)
break
#Get the next set of jobs to run
jobs = jobGraph.stack[-1]
assert len(jobs) > 0
#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)
break
#We check the requirements of the jobGraph to see if we can run it
#within the current worker
successorJobNode = jobs[0]
if successorJobNode.memory > jobGraph.memory:
logger.debug(
"We need more memory for the next job, so finishing")
break
if successorJobNode.cores > jobGraph.cores:
logger.debug(
"We need more cores for the next job, so finishing")
break
if successorJobNode.disk > jobGraph.disk:
logger.debug(
"We need more disk for the next job, so finishing")
break
if successorJobNode.predecessorNumber > 1:
logger.debug(
"The jobGraph has multiple predecessors, we must return to the leader."
)
break
# Load the successor jobGraph
successorJobGraph = jobStore.load(successorJobNode.jobStoreID)
# Somewhat ugly, but check if job is a checkpoint job and quit if
# so
if successorJobGraph.command.startswith("_toil "):
#Load the job
successorJob = Job._loadJob(successorJobGraph.command,
jobStore)
# Check it is not a checkpoint
if successorJob.checkpoint:
logger.debug("Next job is checkpoint, so finishing")
break
##########################################
#We have a single successor job that is not a checkpoint job.
#We transplant the successor jobGraph command and stack
#into the current jobGraph object so that it can be run
#as if it were a command that were part of the current jobGraph.
#We can then delete the successor jobGraph in the jobStore, as it is
#wholly incorporated into the current jobGraph.
##########################################
#Clone the jobGraph and its stack
jobGraph = copy.deepcopy(jobGraph)
#Remove the successor jobGraph
jobGraph.stack.pop()
#These should all match up
assert successorJobGraph.memory == successorJobNode.memory
assert successorJobGraph.cores == successorJobNode.cores
assert successorJobGraph.predecessorsFinished == set()
assert successorJobGraph.predecessorNumber == 1
assert successorJobGraph.command is not None
assert successorJobGraph.jobStoreID == successorJobNode.jobStoreID
#Transplant the command and stack to the current jobGraph
jobGraph.command = successorJobGraph.command
jobGraph.stack += successorJobGraph.stack
assert jobGraph.memory >= successorJobGraph.memory
assert jobGraph.cores >= successorJobGraph.cores
#Build a fileStore to update the job
fileStore = FileStore.createFileStore(
jobStore,
jobGraph,
localWorkerTempDir,
blockFn,
caching=not config.disableCaching)
#Update blockFn
blockFn = fileStore._blockFn
#Add successorJobGraph to those to be deleted
fileStore.jobsToDelete.add(successorJobGraph.jobStoreID)
#This will update the job once the previous job is done
fileStore._updateJobWhenDone()
#Clone the jobGraph and its stack again, so that updates to it do
#not interfere with this update
jobGraph = copy.deepcopy(jobGraph)
logger.debug("Starting the next job")
##########################################
#Finish up the stats
##########################################
if config.stats:
totalCPUTime, totalMemoryUsage = getTotalCpuTimeAndMemoryUsage()
statsDict.workers.time = str(time.time() - startTime)
statsDict.workers.clock = str(totalCPUTime - startClock)
statsDict.workers.memory = str(totalMemoryUsage)
# log the worker log path here so that if the file is truncated the path can still be found
logger.info(
"Worker log can be found at %s. Set --cleanWorkDir to retain this log",
localWorkerTempDir)
logger.info(
"Finished running the chain of jobs on this node, we ran for a total of %f seconds",
time.time() - startTime)
##########################################
#Trapping where worker goes wrong
##########################################
except: #Case that something goes wrong in worker
traceback.print_exc()
logger.error("Exiting the worker because of a failed job on host %s",
socket.gethostname())
FileStore._terminateEvent.set()
##########################################
#Wait for the asynchronous chain of writes/updates to finish
##########################################
blockFn()
##########################################
#All the asynchronous worker/update threads must be finished now,
#so safe to test if they completed okay
##########################################
if FileStore._terminateEvent.isSet():
jobGraph = jobStore.load(jobStoreID)
jobGraph.setupJobAfterFailure(config)
workerFailed = True
##########################################
#Cleanup
##########################################
#Close the worker logging
#Flush at the Python level
sys.stdout.flush()
sys.stderr.flush()
#Flush at the OS level
os.fsync(1)
os.fsync(2)
#Close redirected stdout and replace with the original standard output.
os.dup2(origStdOut, 1)
#Close redirected stderr and replace with the original standard error.
os.dup2(origStdErr, 2)
#sys.stdout and sys.stderr don't need to be modified at all. We don't need
#to call redirectLoggerStreamHandlers since they still log to sys.stderr
#Close our extra handles to the original standard output and standard error
#streams, so we don't leak file handles.
os.close(origStdOut)
os.close(origStdErr)
#Now our file handles are in exactly the state they were in before.
#Copy back the log file to the global dir, if needed
if workerFailed:
jobGraph.logJobStoreFileID = jobStore.getEmptyFileStoreID(
jobGraph.jobStoreID)
with jobStore.updateFileStream(jobGraph.logJobStoreFileID) as w:
with open(tempWorkerLogPath, "r") as f:
if os.path.getsize(tempWorkerLogPath) > logFileByteReportLimit:
f.seek(-logFileByteReportLimit,
2) # seek to last tooBig bytes of file
w.write(f.read())
jobStore.update(jobGraph)
elif debugging: # write log messages
with open(tempWorkerLogPath, 'r') as logFile:
if os.path.getsize(tempWorkerLogPath) > logFileByteReportLimit:
logFile.seek(-logFileByteReportLimit,
2) # seek to last tooBig bytes of file
logMessages = logFile.read().splitlines()
statsDict.logs = [
Expando(jobStoreID=jobStoreID, text=logMessage)
for logMessage in logMessages
]
if (debugging or config.stats or statsDict.workers.logsToMaster
) and not workerFailed: # We have stats/logging to report back
jobStore.writeStatsAndLogging(json.dumps(statsDict))
#Remove the temp dir
cleanUp = config.cleanWorkDir
if cleanUp == 'always' or (cleanUp == 'onSuccess'
and not workerFailed) or (cleanUp == 'onError'
and workerFailed):
shutil.rmtree(localWorkerTempDir)
#This must happen after the log file is done with, else there is no place to put the log
if (not workerFailed) and jobGraph.command == None and len(
jobGraph.stack) == 0 and len(jobGraph.services) == 0:
# We can now safely get rid of the jobGraph
jobStore.delete(jobGraph.jobStoreID)
def workerScript(jobStore, config, jobName, jobStoreID, redirectOutputToLogFile=True):
"""
Worker process script, runs a job.
:param str jobName: The "job name" (a user friendly name) of the job to be run
:param str jobStoreLocator: Specifies the job store to use
:param str jobStoreID: The job store ID of the job to be run
:param bool redirectOutputToLogFile: Redirect standard out and standard error to a log file
"""
logging.basicConfig()
setLogLevel(config.logLevel)
##########################################
#Create the worker killer, if requested
##########################################
logFileByteReportLimit = config.maxLogFileSize
if config.badWorker > 0 and random.random() < config.badWorker:
# We need to kill the process we are currently in, to simulate worker
# failure. We don't want to just send SIGKILL, because we can't tell
# that from a legitimate OOM on our CI runner. We're going to send
# SIGUSR1 so our terminations are distinctive, and then SIGKILL if that
# didn't stick. We definitely don't want to do this from *within* the
# process we are trying to kill, so we fork off. TODO: We can still
# leave the killing code running after the main Toil flow is done, but
# since it's now in a process instead of a thread, the main Python
# process won't wait around for its timeout to expire. I think this is
# better than the old thread-based way where all of Toil would wait
# around to be killed.
killTarget = os.getpid()
sleepTime = config.badWorkerFailInterval * random.random()
if os.fork() == 0:
# We are the child
# Let the parent run some amount of time
time.sleep(sleepTime)
# Kill it gently
os.kill(killTarget, signal.SIGUSR1)
# Wait for that to stick
time.sleep(0.01)
try:
# Kill it harder. Hope the PID hasn't already been reused.
# If we succeeded the first time, this will OSError
os.kill(killTarget, signal.SIGKILL)
except OSError:
pass
# Exit without doing any of Toil's cleanup
os._exit()
# We don't need to reap the child. Either it kills us, or we finish
# before it does. Either way, init will have to clean it up for us.
##########################################
#Load the environment for the jobGraph
##########################################
#First load the environment for the jobGraph.
with jobStore.readSharedFileStream("environment.pickle") as fileHandle:
environment = safeUnpickleFromStream(fileHandle)
for i in environment:
if i not in ("TMPDIR", "TMP", "HOSTNAME", "HOSTTYPE"):
os.environ[i] = environment[i]
# sys.path is used by __import__ to find modules
if "PYTHONPATH" in environment:
for e in environment["PYTHONPATH"].split(':'):
if e != '':
sys.path.append(e)
toilWorkflowDir = Toil.getWorkflowDir(config.workflowID, config.workDir)
##########################################
#Setup the temporary directories.
##########################################
# Dir to put all this worker's temp files in.
localWorkerTempDir = tempfile.mkdtemp(dir=toilWorkflowDir)
os.chmod(localWorkerTempDir, 0o755)
##########################################
#Setup the logging
##########################################
#This is mildly tricky because we don't just want to
#redirect stdout and stderr for this Python process; we want to redirect it
#for this process and all children. Consequently, we can't just replace
#sys.stdout and sys.stderr; we need to mess with the underlying OS-level
#file descriptors. See <http://stackoverflow.com/a/11632982/402891>
#When we start, standard input is file descriptor 0, standard output is
#file descriptor 1, and standard error is file descriptor 2.
#What file do we want to point FDs 1 and 2 to?
tempWorkerLogPath = os.path.join(localWorkerTempDir, "worker_log.txt")
if redirectOutputToLogFile:
# Announce that we are redirecting logging, and where it will now go.
# This is important if we are trying to manually trace a faulty worker invocation.
logger.info("Redirecting logging to %s", tempWorkerLogPath)
sys.stdout.flush()
sys.stderr.flush()
# Save the original stdout and stderr (by opening new file descriptors
# to the same files)
origStdOut = os.dup(1)
origStdErr = os.dup(2)
# Open the file to send stdout/stderr to.
logFh = os.open(tempWorkerLogPath, os.O_WRONLY | os.O_CREAT | os.O_APPEND)
# Replace standard output with a descriptor for the log file
os.dup2(logFh, 1)
# Replace standard error with a descriptor for the log file
os.dup2(logFh, 2)
# Since we only opened the file once, all the descriptors duped from
# the original will share offset information, and won't clobber each
# others' writes. See <http://stackoverflow.com/a/5284108/402891>. This
# shouldn't matter, since O_APPEND seeks to the end of the file before
# every write, but maybe there's something odd going on...
# Close the descriptor we used to open the file
os.close(logFh)
debugging = logging.getLogger().isEnabledFor(logging.DEBUG)
##########################################
#Worker log file trapped from here on in
##########################################
workerFailed = False
statsDict = MagicExpando()
statsDict.jobs = []
statsDict.workers.logsToMaster = []
blockFn = lambda : True
listOfJobs = [jobName]
job = None
try:
#Put a message at the top of the log, just to make sure it's working.
logger.info("---TOIL WORKER OUTPUT LOG---")
sys.stdout.flush()
logProcessContext(config)
##########################################
#Load the jobGraph
##########################################
jobGraph = jobStore.load(jobStoreID)
listOfJobs[0] = str(jobGraph)
logger.debug("Parsed job wrapper")
##########################################
#Cleanup from any earlier invocation of the jobGraph
##########################################
if jobGraph.command == None:
logger.debug("Wrapper has no user job to run.")
# Cleanup jobs already finished
f = lambda jobs : [z for z in [[y for y in x if jobStore.exists(y.jobStoreID)] for x in jobs] if len(z) > 0]
jobGraph.stack = f(jobGraph.stack)
jobGraph.services = f(jobGraph.services)
logger.debug("Cleaned up any references to completed successor jobs")
#This cleans the old log file which may
#have been left if the job is being retried after a job failure.
oldLogFile = jobGraph.logJobStoreFileID
if oldLogFile != None:
jobGraph.logJobStoreFileID = None
jobStore.update(jobGraph) #Update first, before deleting any files
jobStore.deleteFile(oldLogFile)
##########################################
# If a checkpoint exists, restart from the checkpoint
##########################################
# The job is a checkpoint, and is being restarted after previously completing
if jobGraph.checkpoint != None:
logger.debug("Job is a checkpoint")
# If the checkpoint still has extant jobs in its
# (flattened) stack and services, its subtree didn't
# complete properly. We handle the restart of the
# checkpoint here, removing its previous subtree.
if len([i for l in jobGraph.stack for i in l]) > 0 or len(jobGraph.services) > 0:
logger.debug("Checkpoint has failed.")
# Reduce the retry count
assert jobGraph.remainingRetryCount >= 0
jobGraph.remainingRetryCount = max(0, jobGraph.remainingRetryCount - 1)
jobGraph.restartCheckpoint(jobStore)
# Otherwise, the job and successors are done, and we can cleanup stuff we couldn't clean
# because of the job being a checkpoint
else:
logger.debug("The checkpoint jobs seems to have completed okay, removing any checkpoint files to delete.")
#Delete any remnant files
list(map(jobStore.deleteFile, list(filter(jobStore.fileExists, jobGraph.checkpointFilesToDelete))))
##########################################
#Setup the stats, if requested
##########################################
if config.stats:
startClock = getTotalCpuTime()
startTime = time.time()
while True:
##########################################
#Run the jobGraph, if there is one
##########################################
if jobGraph.command is not None:
assert jobGraph.command.startswith("_toil ")
logger.debug("Got a command to run: %s" % jobGraph.command)
#Load the job
job = Job._loadJob(jobGraph.command, jobStore)
# If it is a checkpoint job, save the command
if job.checkpoint:
jobGraph.checkpoint = jobGraph.command
# Create a fileStore object for the job
fileStore = FileStore.createFileStore(jobStore, jobGraph, localWorkerTempDir, blockFn,
caching=not config.disableCaching)
with job._executor(jobGraph=jobGraph,
stats=statsDict if config.stats else None,
fileStore=fileStore):
with fileStore.open(job):
# Get the next block function and list that will contain any messages
blockFn = fileStore._blockFn
job._runner(jobGraph=jobGraph, jobStore=jobStore, fileStore=fileStore)
# Accumulate messages from this job & any subsequent chained jobs
statsDict.workers.logsToMaster += fileStore.loggingMessages
else:
#The command may be none, in which case
#the jobGraph is either a shell ready to be deleted or has
#been scheduled after a failure to cleanup
logger.debug("No user job to run, so finishing")
break
if FileStore._terminateEvent.isSet():
raise RuntimeError("The termination flag is set")
##########################################
#Establish if we can run another jobGraph within the worker
##########################################
successorJobGraph = nextChainableJobGraph(jobGraph, jobStore)
if successorJobGraph is None or config.disableChaining:
# Can't chain any more jobs.
break
##########################################
#We have a single successor job that is not a checkpoint job.
#We transplant the successor jobGraph command and stack
#into the current jobGraph object so that it can be run
#as if it were a command that were part of the current jobGraph.
#We can then delete the successor jobGraph in the jobStore, as it is
#wholly incorporated into the current jobGraph.
##########################################
# add the successor to the list of jobs run
listOfJobs.append(str(successorJobGraph))
#Clone the jobGraph and its stack
jobGraph = copy.deepcopy(jobGraph)
#Remove the successor jobGraph
jobGraph.stack.pop()
#Transplant the command and stack to the current jobGraph
jobGraph.command = successorJobGraph.command
jobGraph.stack += successorJobGraph.stack
# include some attributes for better identification of chained jobs in
# logging output
jobGraph.unitName = successorJobGraph.unitName
jobGraph.jobName = successorJobGraph.jobName
assert jobGraph.memory >= successorJobGraph.memory
assert jobGraph.cores >= successorJobGraph.cores
#Build a fileStore to update the job
fileStore = FileStore.createFileStore(jobStore, jobGraph, localWorkerTempDir, blockFn,
caching=not config.disableCaching)
#Update blockFn
blockFn = fileStore._blockFn
#Add successorJobGraph to those to be deleted
fileStore.jobsToDelete.add(successorJobGraph.jobStoreID)
#This will update the job once the previous job is done
fileStore._updateJobWhenDone()
#Clone the jobGraph and its stack again, so that updates to it do
#not interfere with this update
jobGraph = copy.deepcopy(jobGraph)
logger.debug("Starting the next job")
##########################################
#Finish up the stats
##########################################
if config.stats:
totalCPUTime, totalMemoryUsage = getTotalCpuTimeAndMemoryUsage()
statsDict.workers.time = str(time.time() - startTime)
statsDict.workers.clock = str(totalCPUTime - startClock)
statsDict.workers.memory = str(totalMemoryUsage)
# log the worker log path here so that if the file is truncated the path can still be found
if redirectOutputToLogFile:
logger.info("Worker log can be found at %s. Set --cleanWorkDir to retain this log", localWorkerTempDir)
logger.info("Finished running the chain of jobs on this node, we ran for a total of %f seconds", time.time() - startTime)
##########################################
#Trapping where worker goes wrong
##########################################
except: #Case that something goes wrong in worker
traceback.print_exc()
logger.error("Exiting the worker because of a failed job on host %s", socket.gethostname())
FileStore._terminateEvent.set()
##########################################
#Wait for the asynchronous chain of writes/updates to finish
##########################################
blockFn()
##########################################
#All the asynchronous worker/update threads must be finished now,
#so safe to test if they completed okay
##########################################
if FileStore._terminateEvent.isSet():
jobGraph = jobStore.load(jobStoreID)
jobGraph.setupJobAfterFailure(config)
workerFailed = True
if job and jobGraph.remainingRetryCount == 0:
job._succeeded = False
##########################################
#Cleanup
##########################################
# Close the worker logging
# Flush at the Python level
sys.stdout.flush()
sys.stderr.flush()
if redirectOutputToLogFile:
# Flush at the OS level
os.fsync(1)
os.fsync(2)
# Close redirected stdout and replace with the original standard output.
os.dup2(origStdOut, 1)
# Close redirected stderr and replace with the original standard error.
os.dup2(origStdErr, 2)
# sys.stdout and sys.stderr don't need to be modified at all. We don't
# need to call redirectLoggerStreamHandlers since they still log to
# sys.stderr
# Close our extra handles to the original standard output and standard
# error streams, so we don't leak file handles.
os.close(origStdOut)
os.close(origStdErr)
# Now our file handles are in exactly the state they were in before.
#Copy back the log file to the global dir, if needed
if workerFailed and redirectOutputToLogFile:
jobGraph.logJobStoreFileID = jobStore.getEmptyFileStoreID(jobGraph.jobStoreID)
jobGraph.chainedJobs = listOfJobs
with jobStore.updateFileStream(jobGraph.logJobStoreFileID) as w:
with open(tempWorkerLogPath, "r") as f:
if os.path.getsize(tempWorkerLogPath) > logFileByteReportLimit !=0:
if logFileByteReportLimit > 0:
f.seek(-logFileByteReportLimit, 2) # seek to last tooBig bytes of file
elif logFileByteReportLimit < 0:
f.seek(logFileByteReportLimit, 0) # seek to first tooBig bytes of file
w.write(f.read().encode('utf-8')) # TODO load file using a buffer
jobStore.update(jobGraph)
elif debugging and redirectOutputToLogFile: # write log messages
with open(tempWorkerLogPath, 'r') as logFile:
if os.path.getsize(tempWorkerLogPath) > logFileByteReportLimit != 0:
if logFileByteReportLimit > 0:
logFile.seek(-logFileByteReportLimit, 2) # seek to last tooBig bytes of file
elif logFileByteReportLimit < 0:
logFile.seek(logFileByteReportLimit, 0) # seek to first tooBig bytes of file
logMessages = logFile.read().splitlines()
statsDict.logs.names = listOfJobs
statsDict.logs.messages = logMessages
if (debugging or config.stats or statsDict.workers.logsToMaster) and not workerFailed: # We have stats/logging to report back
jobStore.writeStatsAndLogging(json.dumps(statsDict, ensure_ascii=True))
#Remove the temp dir
cleanUp = config.cleanWorkDir
if cleanUp == 'always' or (cleanUp == 'onSuccess' and not workerFailed) or (cleanUp == 'onError' and workerFailed):
shutil.rmtree(localWorkerTempDir)
#This must happen after the log file is done with, else there is no place to put the log
if (not workerFailed) and jobGraph.command == None and len(jobGraph.stack) == 0 and len(jobGraph.services) == 0:
# We can now safely get rid of the jobGraph
jobStore.delete(jobGraph.jobStoreID)
def main():
logging.basicConfig()
##########################################
#Import necessary modules
##########################################
# This is assuming that worker.py is at a path ending in "/toil/worker.py".
sourcePath = os.path.dirname(os.path.dirname(__file__))
if sourcePath not in sys.path:
sys.path.append(sourcePath)
#Now we can import all the necessary functions
from toil.lib.bioio import setLogLevel
from toil.lib.bioio import getTotalCpuTime
from toil.lib.bioio import getTotalCpuTimeAndMemoryUsage
from toil.job import Job
try:
import boto
except ImportError:
pass
else:
# boto is installed, monkey patch it now
from bd2k.util.ec2.credentials import enable_metadata_credential_caching
enable_metadata_credential_caching()
##########################################
#Input args
##########################################
jobStoreLocator = sys.argv[1]
jobStoreID = sys.argv[2]
# we really want a list of job names but the ID will suffice if the job graph can't
# be loaded. If we can discover the name, we will replace this initial entry
listOfJobs = [jobStoreID]
##########################################
#Load the jobStore/config file
##########################################
jobStore = Toil.resumeJobStore(jobStoreLocator)
config = jobStore.config
##########################################
#Create the worker killer, if requested
##########################################
logFileByteReportLimit = config.maxLogFileSize
if config.badWorker > 0 and random.random() < config.badWorker:
def badWorker():
#This will randomly kill the worker process at a random time
time.sleep(config.badWorkerFailInterval * random.random())
os.kill(os.getpid(), signal.SIGKILL) #signal.SIGINT)
#TODO: FIX OCCASIONAL DEADLOCK WITH SIGINT (tested on single machine)
t = Thread(target=badWorker)
# Ideally this would be a daemon thread but that causes an intermittent (but benign)
# exception similar to the one described here:
# http://stackoverflow.com/questions/20596918/python-exception-in-thread-thread-1-most-likely-raised-during-interpreter-shutd
# Our exception is:
# Exception in thread Thread-1 (most likely raised during interpreter shutdown):
# <type 'exceptions.AttributeError'>: 'NoneType' object has no attribute 'kill'
# This attribute error is caused by the call os.kill() and apparently unavoidable with a
# daemon
t.start()
##########################################
#Load the environment for the jobGraph
##########################################
#First load the environment for the jobGraph.
with jobStore.readSharedFileStream("environment.pickle") as fileHandle:
environment = cPickle.load(fileHandle)
for i in environment:
if i not in ("TMPDIR", "TMP", "HOSTNAME", "HOSTTYPE"):
os.environ[i] = environment[i]
# sys.path is used by __import__ to find modules
if "PYTHONPATH" in environment:
for e in environment["PYTHONPATH"].split(':'):
if e != '':
sys.path.append(e)
setLogLevel(config.logLevel)
toilWorkflowDir = Toil.getWorkflowDir(config.workflowID, config.workDir)
##########################################
#Setup the temporary directories.
##########################################
# Dir to put all this worker's temp files in.
localWorkerTempDir = tempfile.mkdtemp(dir=toilWorkflowDir)
os.chmod(localWorkerTempDir, 0o755)
##########################################
#Setup the logging
##########################################
#This is mildly tricky because we don't just want to
#redirect stdout and stderr for this Python process; we want to redirect it
#for this process and all children. Consequently, we can't just replace
#sys.stdout and sys.stderr; we need to mess with the underlying OS-level
#file descriptors. See <http://stackoverflow.com/a/11632982/402891>
#When we start, standard input is file descriptor 0, standard output is
#file descriptor 1, and standard error is file descriptor 2.
#What file do we want to point FDs 1 and 2 to?
tempWorkerLogPath = os.path.join(localWorkerTempDir, "worker_log.txt")
#Save the original stdout and stderr (by opening new file descriptors to the
#same files)
origStdOut = os.dup(1)
origStdErr = os.dup(2)
#Open the file to send stdout/stderr to.
logFh = os.open(tempWorkerLogPath, os.O_WRONLY | os.O_CREAT | os.O_APPEND)
#Replace standard output with a descriptor for the log file
os.dup2(logFh, 1)
#Replace standard error with a descriptor for the log file
os.dup2(logFh, 2)
#Since we only opened the file once, all the descriptors duped from the
#original will share offset information, and won't clobber each others'
#writes. See <http://stackoverflow.com/a/5284108/402891>. This shouldn't
#matter, since O_APPEND seeks to the end of the file before every write, but
#maybe there's something odd going on...
#Close the descriptor we used to open the file
os.close(logFh)
debugging = logging.getLogger().isEnabledFor(logging.DEBUG)
##########################################
#Worker log file trapped from here on in
##########################################
workerFailed = False
statsDict = MagicExpando()
statsDict.jobs = []
statsDict.workers.logsToMaster = []
blockFn = lambda : True
cleanCacheFn = lambda x : True
try:
#Put a message at the top of the log, just to make sure it's working.
print("---TOIL WORKER OUTPUT LOG---")
sys.stdout.flush()
#Log the number of open file descriptors so we can tell if we're leaking
#them.
logger.debug("Next available file descriptor: {}".format(
nextOpenDescriptor()))
logProcessContext(config)
##########################################
#Load the jobGraph
##########################################
jobGraph = jobStore.load(jobStoreID)
listOfJobs[0] = str(jobGraph)
logger.debug("Parsed jobGraph")
##########################################
#Cleanup from any earlier invocation of the jobGraph
##########################################
if jobGraph.command == None:
# Cleanup jobs already finished
f = lambda jobs : filter(lambda x : len(x) > 0, map(lambda x :
filter(lambda y : jobStore.exists(y.jobStoreID), x), jobs))
jobGraph.stack = f(jobGraph.stack)
jobGraph.services = f(jobGraph.services)
logger.debug("Cleaned up any references to completed successor jobs")
#This cleans the old log file which may
#have been left if the job is being retried after a job failure.
oldLogFile = jobGraph.logJobStoreFileID
if oldLogFile != None:
jobGraph.logJobStoreFileID = None
jobStore.update(jobGraph) #Update first, before deleting any files
jobStore.deleteFile(oldLogFile)
##########################################
# If a checkpoint exists, restart from the checkpoint
##########################################
# The job is a checkpoint, and is being restarted after previously completing
if jobGraph.checkpoint != None:
logger.debug("Job is a checkpoint")
if len(jobGraph.stack) > 0 or len(jobGraph.services) > 0 or jobGraph.command != None:
if jobGraph.command != None:
assert jobGraph.command == jobGraph.checkpoint
logger.debug("Checkpoint job already has command set to run")
else:
jobGraph.command = jobGraph.checkpoint
jobStore.update(jobGraph) # Update immediately to ensure that checkpoint
# is made before deleting any remaining successors
if len(jobGraph.stack) > 0 or len(jobGraph.services) > 0:
# If the subtree of successors is not complete restart everything
logger.debug("Checkpoint job has unfinished successor jobs, deleting the jobs on the stack: %s, services: %s " %
(jobGraph.stack, jobGraph.services))
# Delete everything on the stack, as these represent successors to clean
# up as we restart the queue
def recursiveDelete(jobGraph2):
# Recursive walk the stack to delete all remaining jobs
for jobs in jobGraph2.stack + jobGraph2.services:
for jobNode in jobs:
if jobStore.exists(jobNode.jobStoreID):
recursiveDelete(jobStore.load(jobNode.jobStoreID))
else:
logger.debug("Job %s has already been deleted", jobNode)
if jobGraph2 != jobGraph:
logger.debug("Checkpoint is deleting old successor job: %s", jobGraph2.jobStoreID)
jobStore.delete(jobGraph2.jobStoreID)
recursiveDelete(jobGraph)
jobGraph.stack = [ [], [] ] # Initialise the job to mimic the state of a job
# that has been previously serialised but which as yet has no successors
jobGraph.services = [] # Empty the services
# Update the jobStore to avoid doing this twice on failure and make this clean.
jobStore.update(jobGraph)
# Otherwise, the job and successors are done, and we can cleanup stuff we couldn't clean
# because of the job being a checkpoint
else:
logger.debug("The checkpoint jobs seems to have completed okay, removing any checkpoint files to delete.")
#Delete any remnant files
map(jobStore.deleteFile, filter(jobStore.fileExists, jobGraph.checkpointFilesToDelete))
##########################################
#Setup the stats, if requested
##########################################
if config.stats:
startTime = time.time()
startClock = getTotalCpuTime()
#Make a temporary file directory for the jobGraph
#localTempDir = makePublicDir(os.path.join(localWorkerTempDir, "localTempDir"))
startTime = time.time()
while True:
##########################################
#Run the jobGraph, if there is one
##########################################
if jobGraph.command is not None:
assert jobGraph.command.startswith( "_toil " )
logger.debug("Got a command to run: %s" % jobGraph.command)
#Load the job
job = Job._loadJob(jobGraph.command, jobStore)
# If it is a checkpoint job, save the command
if job.checkpoint:
jobGraph.checkpoint = jobGraph.command
# Create a fileStore object for the job
fileStore = FileStore.createFileStore(jobStore, jobGraph, localWorkerTempDir, blockFn,
caching=not config.disableCaching)
with job._executor(jobGraph=jobGraph,
stats=statsDict if config.stats else None,
fileStore=fileStore):
with fileStore.open(job):
# Get the next block function and list that will contain any messages
blockFn = fileStore._blockFn
job._runner(jobGraph=jobGraph, jobStore=jobStore, fileStore=fileStore)
# Accumulate messages from this job & any subsequent chained jobs
statsDict.workers.logsToMaster += fileStore.loggingMessages
else:
#The command may be none, in which case
#the jobGraph is either a shell ready to be deleted or has
#been scheduled after a failure to cleanup
break
if FileStore._terminateEvent.isSet():
raise RuntimeError("The termination flag is set")
##########################################
#Establish if we can run another jobGraph within the worker
##########################################
#If no more jobs to run or services not finished, quit
if len(jobGraph.stack) == 0 or len(jobGraph.services) > 0 or jobGraph.checkpoint != None:
logger.debug("Stopping running chain of jobs: length of stack: %s, services: %s, checkpoint: %s",
len(jobGraph.stack), len(jobGraph.services), jobGraph.checkpoint != None)
break
#Get the next set of jobs to run
jobs = jobGraph.stack[-1]
assert len(jobs) > 0
#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)
break
#We check the requirements of the jobGraph to see if we can run it
#within the current worker
successorJobNode = jobs[0]
if successorJobNode.memory > jobGraph.memory:
logger.debug("We need more memory for the next job, so finishing")
break
if successorJobNode.cores > jobGraph.cores:
logger.debug("We need more cores for the next job, so finishing")
break
if successorJobNode.disk > jobGraph.disk:
logger.debug("We need more disk for the next job, so finishing")
break
if successorJobNode.preemptable != jobGraph.preemptable:
logger.debug("Preemptability is different for the next job, returning to the leader")
break
if successorJobNode.predecessorNumber > 1:
logger.debug("The jobGraph has multiple predecessors, we must return to the leader.")
break
# Load the successor jobGraph
successorJobGraph = jobStore.load(successorJobNode.jobStoreID)
# add the successor to the list of jobs run
listOfJobs.append(str(successorJobGraph))
# Somewhat ugly, but check if job is a checkpoint job and quit if
# so
if successorJobGraph.command.startswith( "_toil " ):
#Load the job
successorJob = Job._loadJob(successorJobGraph.command, jobStore)
# Check it is not a checkpoint
if successorJob.checkpoint:
logger.debug("Next job is checkpoint, so finishing")
break
##########################################
#We have a single successor job that is not a checkpoint job.
#We transplant the successor jobGraph command and stack
#into the current jobGraph object so that it can be run
#as if it were a command that were part of the current jobGraph.
#We can then delete the successor jobGraph in the jobStore, as it is
#wholly incorporated into the current jobGraph.
##########################################
#Clone the jobGraph and its stack
jobGraph = copy.deepcopy(jobGraph)
#Remove the successor jobGraph
jobGraph.stack.pop()
#These should all match up
assert successorJobGraph.memory == successorJobNode.memory
assert successorJobGraph.cores == successorJobNode.cores
assert successorJobGraph.predecessorsFinished == set()
assert successorJobGraph.predecessorNumber == 1
assert successorJobGraph.command is not None
assert successorJobGraph.jobStoreID == successorJobNode.jobStoreID
#Transplant the command and stack to the current jobGraph
jobGraph.command = successorJobGraph.command
jobGraph.stack += successorJobGraph.stack
# include some attributes for better identification of chained jobs in
# logging output
jobGraph.unitName = successorJobGraph.unitName
jobGraph.jobName = successorJobGraph.jobName
assert jobGraph.memory >= successorJobGraph.memory
assert jobGraph.cores >= successorJobGraph.cores
#Build a fileStore to update the job
fileStore = FileStore.createFileStore(jobStore, jobGraph, localWorkerTempDir, blockFn,
caching=not config.disableCaching)
#Update blockFn
blockFn = fileStore._blockFn
#Add successorJobGraph to those to be deleted
fileStore.jobsToDelete.add(successorJobGraph.jobStoreID)
#This will update the job once the previous job is done
fileStore._updateJobWhenDone()
#Clone the jobGraph and its stack again, so that updates to it do
#not interfere with this update
jobGraph = copy.deepcopy(jobGraph)
logger.debug("Starting the next job")
##########################################
#Finish up the stats
##########################################
if config.stats:
totalCPUTime, totalMemoryUsage = getTotalCpuTimeAndMemoryUsage()
statsDict.workers.time = str(time.time() - startTime)
statsDict.workers.clock = str(totalCPUTime - startClock)
statsDict.workers.memory = str(totalMemoryUsage)
# log the worker log path here so that if the file is truncated the path can still be found
logger.info("Worker log can be found at %s. Set --cleanWorkDir to retain this log", localWorkerTempDir)
logger.info("Finished running the chain of jobs on this node, we ran for a total of %f seconds", time.time() - startTime)
##########################################
#Trapping where worker goes wrong
##########################################
except: #Case that something goes wrong in worker
traceback.print_exc()
logger.error("Exiting the worker because of a failed job on host %s", socket.gethostname())
FileStore._terminateEvent.set()
##########################################
#Wait for the asynchronous chain of writes/updates to finish
##########################################
blockFn()
##########################################
#All the asynchronous worker/update threads must be finished now,
#so safe to test if they completed okay
##########################################
if FileStore._terminateEvent.isSet():
jobGraph = jobStore.load(jobStoreID)
jobGraph.setupJobAfterFailure(config)
workerFailed = True
##########################################
#Cleanup
##########################################
#Close the worker logging
#Flush at the Python level
sys.stdout.flush()
sys.stderr.flush()
#Flush at the OS level
os.fsync(1)
os.fsync(2)
#Close redirected stdout and replace with the original standard output.
os.dup2(origStdOut, 1)
#Close redirected stderr and replace with the original standard error.
os.dup2(origStdErr, 2)
#sys.stdout and sys.stderr don't need to be modified at all. We don't need
#to call redirectLoggerStreamHandlers since they still log to sys.stderr
#Close our extra handles to the original standard output and standard error
#streams, so we don't leak file handles.
os.close(origStdOut)
os.close(origStdErr)
#Now our file handles are in exactly the state they were in before.
#Copy back the log file to the global dir, if needed
if workerFailed:
jobGraph.logJobStoreFileID = jobStore.getEmptyFileStoreID(jobGraph.jobStoreID)
jobGraph.chainedJobs = listOfJobs
with jobStore.updateFileStream(jobGraph.logJobStoreFileID) as w:
with open(tempWorkerLogPath, "r") as f:
if os.path.getsize(tempWorkerLogPath) > logFileByteReportLimit !=0:
if logFileByteReportLimit > 0:
f.seek(-logFileByteReportLimit, 2) # seek to last tooBig bytes of file
elif logFileByteReportLimit < 0:
f.seek(logFileByteReportLimit, 0) # seek to first tooBig bytes of file
w.write(f.read())
jobStore.update(jobGraph)
elif debugging: # write log messages
with open(tempWorkerLogPath, 'r') as logFile:
if os.path.getsize(tempWorkerLogPath) > logFileByteReportLimit != 0:
if logFileByteReportLimit > 0:
logFile.seek(-logFileByteReportLimit, 2) # seek to last tooBig bytes of file
elif logFileByteReportLimit < 0:
logFile.seek(logFileByteReportLimit, 0) # seek to first tooBig bytes of file
logMessages = logFile.read().splitlines()
statsDict.logs.names = listOfJobs
statsDict.logs.messages = logMessages
if (debugging or config.stats or statsDict.workers.logsToMaster) and not workerFailed: # We have stats/logging to report back
jobStore.writeStatsAndLogging(json.dumps(statsDict))
#Remove the temp dir
cleanUp = config.cleanWorkDir
if cleanUp == 'always' or (cleanUp == 'onSuccess' and not workerFailed) or (cleanUp == 'onError' and workerFailed):
shutil.rmtree(localWorkerTempDir)
#This must happen after the log file is done with, else there is no place to put the log
if (not workerFailed) and jobGraph.command == None and len(jobGraph.stack) == 0 and len(jobGraph.services) == 0:
# We can now safely get rid of the jobGraph
jobStore.delete(jobGraph.jobStoreID)