def testToilStats_SortSimple(self):
"""
Tests the toilStats utility using the scriptTree_sort example.
"""
for test in xrange(self.testNo):
tempDir = getTempDirectory(os.getcwd())
tempFile = getTempFile(rootDir=tempDir)
outputFile = getTempFile(rootDir=tempDir)
toilDir = os.path.join(tempDir, "testToil")
lines = 10000
maxLineLength = 10
N = 1000
makeFileToSort(tempFile, lines, maxLineLength)
# Sort the file
rootPath = os.path.join(toilPackageDirPath(), "test", "sort")
system("{rootPath}/sort.py "
"--toil {toilDir} "
"--logLevel=DEBUG "
"--fileToSort={tempFile} "
"--N {N} --stats "
"--jobTime 0.5 "
"--retryCount 99".format(**locals()))
# Now get the stats
toilStats = self.getScriptPath('toilStats')
system("{toilStats} "
"--toil {toilDir} "
"--outputFile {outputFile}".format(**locals()))
# Cleanup
system("rm -rf %s" % tempDir)
def setUp(self):
super(UtilsTest, self).setUp()
self.tempDir = self._createTempDir()
self.tempFile = getTempFile(rootDir=self.tempDir)
self.outputFile = getTempFile(rootDir=self.tempDir)
self.toilDir = os.path.join(self.tempDir, "jobstore")
self.assertFalse(os.path.exists(self.toilDir))
self.lines = 1000
self.lineLen = 10
self.N = 1000
makeFileToSort(self.tempFile, self.lines, self.lineLen)
# First make our own sorted version
with open(self.tempFile, "r") as fileHandle:
self.correctSort = fileHandle.readlines()
self.correctSort.sort()
def testEncapsulation(self):
"""
Tests the Job.encapsulation method, which uses the EncapsulationJob
class.
"""
# Temporary file
outFile = getTempFile(rootDir=self._createTempDir())
try:
# Encapsulate a job graph
a = T.wrapJobFn(encapsulatedJobFn, "A", outFile)
a = a.encapsulate()
# Now add children/follow to the encapsulated graph
d = T.wrapFn(f, a.rv(), outFile)
e = T.wrapFn(f, d.rv(), outFile)
a.addChild(d)
a.addFollowOn(e)
# Create the runner for the workflow.
options = T.Runner.getDefaultOptions(self._getTestJobStorePath())
options.logLevel = "INFO"
# Run the workflow, the return value being the number of failed jobs
T.Runner.startToil(a, options)
# Check output
self.assertEquals(open(outFile, 'r').readline(), "ABCDE")
finally:
os.remove(outFile)
def testServiceDeadlock(self):
"""
Creates a job with more services than maxServices, checks that deadlock is detected.
"""
outFile = getTempFile(rootDir=self._createTempDir())
try:
def makeWorkflow():
job = Job()
r1 = job.addService(TestServiceSerialization("woot1"))
r2 = job.addService(TestServiceSerialization("woot2"))
r3 = job.addService(TestServiceSerialization("woot3"))
job.addChildFn(fnTest, [ r1, r2, r3 ], outFile)
return job
# This should fail as too few services available
try:
self.runToil(makeWorkflow(), badWorker=0.0, maxServiceJobs=2, deadlockWait=5)
except DeadlockException:
print "Got expected deadlock exception"
else:
assert 0
# This should pass, as adequate services available
self.runToil(makeWorkflow(), maxServiceJobs=3)
# Check we get expected output
assert open(outFile, 'r').read() == "woot1 woot2 woot3"
finally:
os.remove(outFile)
def testEncapsulation(self):
"""
Tests the Job.encapsulation method, which uses the EncapsulationJob
class.
"""
#Temporary file
outFile = getTempFile(rootDir=os.getcwd())
#Make a job graph
a = T.wrapFn(f, "A", outFile)
b = a.addChildFn(f, a.rv(), outFile)
c = a.addFollowOnFn(f, b.rv(), outFile)
#Encapsulate it
a = a.encapsulate()
#Now add children/follow to the encapsulated graph
d = T.wrapFn(f, c.rv(), outFile)
e = T.wrapFn(f, d.rv(), outFile)
a.addChild(d)
a.addFollowOn(e)
#Create the runner for the workflow.
options = T.Runner.getDefaultOptions()
options.logLevel = "INFO"
#Run the workflow, the return value being the number of failed jobs
self.assertEquals(T.Runner.startToil(a, options), 0)
T.Runner.cleanup(options) #This removes the jobStore
#Check output
self.assertEquals(open(outFile, 'r').readline(), "ABCDE")
#Cleanup
os.remove(outFile)
def issueBatchJob(self, command, memory, cores, disk, preemptable):
"""
Issues parasol with job commands.
"""
self.checkResourceRequest(memory, cores, disk)
MiB = 1 << 20
truncatedMemory = (memory / MiB) * MiB
# Look for a batch for jobs with these resource requirements, with
# the memory rounded down to the nearest megabyte. Rounding down
# meams the new job can't ever decrease the memory requirements
# of jobs already in the batch.
if len(self.resultsFiles) >= self.maxBatches:
raise RuntimeError( 'Number of batches reached limit of %i' % self.maxBatches)
try:
results = self.resultsFiles[(truncatedMemory, cores)]
except KeyError:
results = getTempFile(rootDir=self.parasolResultsDir)
self.resultsFiles[(truncatedMemory, cores)] = results
# Prefix the command with environment overrides, optionally looking them up from the
# current environment if the value is None
command = ' '.join(concat('env', self.__environment(), command))
parasolCommand = ['-verbose',
'-ram=%i' % memory,
'-cpu=%i' % cores,
'-results=' + results,
'add', 'job', command]
# Deal with the cpus
self.usedCpus += cores
while True: # Process finished results with no wait
try:
jobID = self.cpuUsageQueue.get_nowait()
except Empty:
break
if jobID in self.jobIDsToCpu.keys():
self.usedCpus -= self.jobIDsToCpu.pop(jobID)
assert self.usedCpus >= 0
while self.usedCpus > self.maxCores: # If we are still waiting
jobID = self.cpuUsageQueue.get()
if jobID in self.jobIDsToCpu.keys():
self.usedCpus -= self.jobIDsToCpu.pop(jobID)
assert self.usedCpus >= 0
# Now keep going
while True:
line = self._runParasol(parasolCommand)[1][0]
match = self.parasolOutputPattern.match(line)
if match is None:
# This is because parasol add job will return success, even if the job was not
# properly issued!
logger.info('We failed to properly add the job, we will try again after a 5s.')
time.sleep(5)
else:
jobID = int(match.group(1))
self.jobIDsToCpu[jobID] = cores
self.runningJobs.add(jobID)
logger.debug("Got the parasol job id: %s from line: %s" % (jobID, line))
return jobID
def testService(self, checkpoint=False):
"""
Tests the creation of a Job.Service with random failures of the worker.
"""
for test in xrange(2):
outFile = getTempFile(rootDir=self._createTempDir()) # Temporary file
messageInt = random.randint(1, sys.maxint)
try:
# Wire up the services/jobs
t = Job.wrapJobFn(serviceTest, outFile, messageInt, checkpoint=checkpoint)
# Run the workflow repeatedly until success
self.runToil(t)
# Check output
self.assertEquals(int(open(outFile, 'r').readline()), messageInt)
finally:
os.remove(outFile)
def testService(self):
"""
Tests the creation of a Job.Service.
"""
# Temporary file
outFile = getTempFile(rootDir=self._createTempDir())
try:
# Wire up the services/jobs
t = Job.wrapFn(f, "1", outFile)
t.addChildFn(f, t.addService(TestService("2", "3", outFile)), outFile)
# Create the runner for the workflow.
options = Job.Runner.getDefaultOptions(self._getTestJobStorePath())
options.logLevel = "INFO"
# Run the workflow, the return value being the number of failed jobs
Job.Runner.startToil(t, options)
# Check output
self.assertEquals(open(outFile, 'r').readline(), "123")
finally:
os.remove(outFile)
def testStatic(self):
"""
Create a DAG of jobs non-dynamically and run it. DAG is:
A -> F
\-------
B -> D \
\ \
------- C -> E
Follow on is marked by ->
"""
outFile = getTempFile(rootDir=self._createTempDir())
try:
# Create the jobs
A = Job.wrapFn(fn1Test, "A", outFile)
B = Job.wrapFn(fn1Test, A.rv(), outFile)
C = Job.wrapFn(fn1Test, B.rv(), outFile)
D = Job.wrapFn(fn1Test, C.rv(), outFile)
E = Job.wrapFn(fn1Test, D.rv(), outFile)
F = Job.wrapFn(fn1Test, E.rv(), outFile)
# Connect them into a workflow
A.addChild(B)
A.addChild(C)
B.addChild(C)
B.addFollowOn(E)
C.addFollowOn(D)
A.addFollowOn(F)
# Create the runner for the workflow.
options = Job.Runner.getDefaultOptions(self._getTestJobStorePath())
options.logLevel = "INFO"
options.retryCount = 100
options.badWorker = 0.5
options.badWorkerFailInterval = 0.01
# Run the workflow, the return value being the number of failed jobs
Job.Runner.startToil(A, options)
# Check output
self.assertEquals(open(outFile, 'r').readline(), "ABCDEFG")
finally:
os.remove(outFile)
def testStatic(self):
"""
Create a DAG of jobs non-dynamically and run it. DAG is:
A -> F
\-------
B -> D \
\ \
------- C -> E
Follow on is marked by ->
"""
#Temporary file
outFile = getTempFile(rootDir=os.getcwd())
#Create the jobs
A = Job.wrapFn(f, "A", outFile)
B = Job.wrapFn(f, A.rv(0), outFile)
C = Job.wrapFn(f, B.rv(0), outFile)
D = Job.wrapFn(f, C.rv(0), outFile)
E = Job.wrapFn(f, D.rv(0), outFile)
F = Job.wrapFn(f, E.rv(0), outFile)
#Connect them into a workflow
A.addChild(B)
A.addChild(C)
B.addChild(C)
B.addFollowOn(E)
C.addFollowOn(D)
A.addFollowOn(F)
#Create the runner for the workflow.
options = Job.Runner.getDefaultOptions()
options.logLevel = "INFO"
#Run the workflow, the return value being the number of failed jobs
self.assertEquals(Job.Runner.startToil(A, options), 0)
Job.Runner.cleanup(options) #This removes the jobStore
#Check output
self.assertEquals(open(outFile, 'r').readline(), "ABCDEF")
#Cleanup
os.remove(outFile)
def testEvaluatingRandomDAG(self):
"""
Randomly generate test input then check that the ordering of the running
respected the constraints.
"""
jobStore = self._getTestJobStorePath()
for test in xrange(30):
# Temporary file
outFile = getTempFile(rootDir=os.getcwd())
# Make a random DAG for the set of child edges
nodeNumber = random.choice(xrange(2, 20))
childEdges = self.makeRandomDAG(nodeNumber)
# Get an adjacency list representation and check is acyclic
adjacencyList = self.getAdjacencyList(nodeNumber, childEdges)
self.assertTrue(self.isAcyclic(adjacencyList))
# Add in follow on edges - these are returned as a list, and as a set
# of augmented edges in the adjacency list
followOnEdges = self.addRandomFollowOnEdges(adjacencyList)
self.assertTrue(self.isAcyclic(adjacencyList))
# Make the job graph
rootJob = self.makeJobGraph(nodeNumber, childEdges, followOnEdges, outFile)
# Run the job graph
options = Job.Runner.getDefaultOptions()
options.jobStore = "%s.%i" % (jobStore, test)
Job.Runner.startToil(rootJob, options)
# Get the ordering add the implied ordering to the graph
with open(outFile, 'r') as fH:
ordering = map(int, fH.readline().split())
# Check all the jobs were run
self.assertEquals(set(ordering), set(xrange(nodeNumber)))
# Add the ordering to the graph
for i in xrange(nodeNumber - 1):
adjacencyList[ordering[i]].add(ordering[i + 1])
# Check the ordering retains an acyclic graph
if not self.isAcyclic(adjacencyList):
print "ORDERING", ordering
print "CHILD EDGES", childEdges
print "FOLLOW ON EDGES", followOnEdges
print "ADJACENCY LIST", adjacencyList
self.assertTrue(self.isAcyclic(adjacencyList))
# Cleanup
os.remove(outFile)
def testServiceRecursive(self, checkpoint=True):
"""
Tests the creation of a Job.Service, creating a chain of services and accessing jobs.
Randomly fails the worker.
"""
for test in xrange(1):
# Temporary file
outFile = getTempFile(rootDir=self._createTempDir())
messages = [ random.randint(1, sys.maxint) for i in xrange(3) ]
try:
# Wire up the services/jobs
t = Job.wrapJobFn(serviceTestRecursive, outFile, messages, checkpoint=checkpoint)
# Run the workflow repeatedly until success
self.runToil(t)
# Check output
self.assertEquals(map(int, open(outFile, 'r').readlines()), messages)
finally:
os.remove(outFile)
def testServiceParallelRecursive(self, checkpoint=True):
"""
Tests the creation of a Job.Service, creating parallel chains of services and accessing jobs.
Randomly fails the worker.
"""
for test in xrange(1):
# Temporary file
outFiles = [ getTempFile(rootDir=self._createTempDir()) for j in xrange(2) ]
messageBundles = [ [ random.randint(1, sys.maxint) for i in xrange(3) ] for j in xrange(2) ]
try:
# Wire up the services/jobs
t = Job.wrapJobFn(serviceTestParallelRecursive, outFiles, messageBundles, checkpoint=True)
# Run the workflow repeatedly until success
self.runToil(t, retryCount=2)
# Check output
for (messages, outFile) in zip(messageBundles, outFiles):
self.assertEquals(map(int, open(outFile, 'r').readlines()), messages)
finally:
map(os.remove, outFiles)
def main():
parser = ArgumentParser()
Job.Runner.addToilOptions(parser)
addCactusWorkflowOptions(parser)
parser.add_argument("seqFile", help = "Seq file")
parser.add_argument("outputHal", type=str, help = "Output HAL file")
#Progressive Cactus Options
parser.add_argument("--database", dest="database",
help="Database type: tokyo_cabinet or kyoto_tycoon"
" [default: %(default)s]",
default="kyoto_tycoon")
parser.add_argument("--configFile", dest="configFile",
help="Specify cactus configuration file",
default=None)
parser.add_argument("--root", dest="root", help="Name of ancestral node (which"
" must appear in NEWICK tree in <seqfile>) to use as a "
"root for the alignment. Any genomes not below this node "
"in the tree may be used as outgroups but will never appear"
" in the output. If no root is specifed then the root"
" of the tree is used. ", default=None)
parser.add_argument("--latest", dest="latest", action="store_true",
help="Use the latest version of the docker container "
"rather than pulling one matching this version of cactus")
parser.add_argument("--containerImage", dest="containerImage", default=None,
help="Use the the specified pre-built containter image "
"rather than pulling one from quay.io")
parser.add_argument("--binariesMode", choices=["docker", "local", "singularity"],
help="The way to run the Cactus binaries", default=None)
options = parser.parse_args()
setupBinaries(options)
setLoggingFromOptions(options)
# Mess with some toil options to create useful defaults.
# Caching generally slows down the cactus workflow, plus some
# methods like readGlobalFileStream don't support forced
# reads directly from the job store rather than from cache.
options.disableCaching = True
# Job chaining breaks service termination timing, causing unused
# databases to accumulate and waste memory for no reason.
options.disableChaining = True
# The default deadlockWait is currently 60 seconds. This can cause
# issues if the database processes take a while to actually begin
# after they're issued. Change it to at least an hour so that we
# don't preemptively declare a deadlock.
if options.deadlockWait is None or options.deadlockWait < 3600:
options.deadlockWait = 3600
if options.retryCount is None:
# If the user didn't specify a retryCount value, make it 5
# instead of Toil's default (1).
options.retryCount = 5
with Toil(options) as toil:
importSingularityImage()
#Run the workflow
if options.restart:
halID = toil.restart()
else:
options.cactusDir = getTempDirectory()
#Create the progressive cactus project
projWrapper = ProjectWrapper(options)
projWrapper.writeXml()
pjPath = os.path.join(options.cactusDir, ProjectWrapper.alignmentDirName,
'%s_project.xml' % ProjectWrapper.alignmentDirName)
assert os.path.exists(pjPath)
project = MultiCactusProject()
if not os.path.isdir(options.cactusDir):
os.makedirs(options.cactusDir)
project.readXML(pjPath)
#import the sequences
seqIDs = []
print "Importing %s sequences" % (len(project.getInputSequencePaths()))
for seq in project.getInputSequencePaths():
if os.path.isdir(seq):
tmpSeq = getTempFile()
catFiles([os.path.join(seq, subSeq) for subSeq in os.listdir(seq)], tmpSeq)
seq = tmpSeq
seq = makeURL(seq)
seqIDs.append(toil.importFile(seq))
project.setInputSequenceIDs(seqIDs)
#import cactus config
if options.configFile:
cactusConfigID = toil.importFile(makeURL(options.configFile))
else:
cactusConfigID = toil.importFile(makeURL(project.getConfigPath()))
project.setConfigID(cactusConfigID)
project.syncToFileStore(toil)
configNode = ET.parse(project.getConfigPath()).getroot()
configWrapper = ConfigWrapper(configNode)
configWrapper.substituteAllPredefinedConstantsWithLiterals()
project.writeXML(pjPath)
halID = toil.start(RunCactusPreprocessorThenProgressiveDown(options, project, memory=configWrapper.getDefaultMemory()))
toil.exportFile(halID, makeURL(options.outputHal))
def issueBatchJob(self, jobNode):
"""
Issues parasol with job commands.
"""
self.checkResourceRequest(jobNode.memory, jobNode.cores, jobNode.disk)
MiB = 1 << 20
truncatedMemory = (jobNode.memory / MiB) * MiB
# Look for a batch for jobs with these resource requirements, with
# the memory rounded down to the nearest megabyte. Rounding down
# meams the new job can't ever decrease the memory requirements
# of jobs already in the batch.
if len(self.resultsFiles) >= self.maxBatches:
raise RuntimeError('Number of batches reached limit of %i' %
self.maxBatches)
try:
results = self.resultsFiles[(truncatedMemory, jobNode.cores)]
except KeyError:
results = getTempFile(rootDir=self.parasolResultsDir)
self.resultsFiles[(truncatedMemory, jobNode.cores)] = results
# Prefix the command with environment overrides, optionally looking them up from the
# current environment if the value is None
command = ' '.join(concat('env', self.__environment(),
jobNode.command))
parasolCommand = [
'-verbose',
'-ram=%i' % jobNode.memory,
'-cpu=%i' % jobNode.cores, '-results=' + results, 'add', 'job',
command
]
# Deal with the cpus
self.usedCpus += jobNode.cores
while True: # Process finished results with no wait
try:
jobID = self.cpuUsageQueue.get_nowait()
except Empty:
break
if jobID in self.jobIDsToCpu.keys():
self.usedCpus -= self.jobIDsToCpu.pop(jobID)
assert self.usedCpus >= 0
while self.usedCpus > self.maxCores: # If we are still waiting
jobID = self.cpuUsageQueue.get()
if jobID in self.jobIDsToCpu.keys():
self.usedCpus -= self.jobIDsToCpu.pop(jobID)
assert self.usedCpus >= 0
# Now keep going
while True:
line = self._runParasol(parasolCommand)[1][0]
match = self.parasolOutputPattern.match(line)
if match is None:
# This is because parasol add job will return success, even if the job was not
# properly issued!
logger.debug(
'We failed to properly add the job, we will try again after a 5s.'
)
time.sleep(5)
else:
jobID = int(match.group(1))
self.jobIDsToCpu[jobID] = jobNode.cores
self.runningJobs.add(jobID)
logger.debug("Got the parasol job id: %s from line: %s" %
(jobID, line))
return jobID
