def start_job(self, zip_filename, uncompress=True, client=None):
if uncompress:
zip_file = os.path.join(self.zip_dir, zip_filename)
# transfer zip file to workers
for watcher in self.nodes_watchers:
if watcher.split(':')[0] == self.ip_address:
continue
file_trans_client = FileTransportClient(watcher, zip_file)
file_trans_client.send_file()
job_dir = ZipHandler.uncompress(zip_file, self.job_dir)
else:
job_dir = os.path.join(self.job_dir,
zip_filename.rsplit('.', 1)[0])
job = import_job(job_dir)
worker_port = job.context.job.port
port = job.context.job.master_port
nodes = [watcher.split(':')[0] for watcher in self.nodes_watchers]
if len(nodes) > 0:
info = MasterJobInfo(port, nodes, worker_port)
self.running_jobs[job.real_name] = info
dirname = os.path.dirname(os.path.abspath(__file__))
f = os.path.join(dirname, 'loader.py')
workers = ['%s:%s' % (node, worker_port) for node in nodes]
cmds = [
'python', f, '-j', job_dir, '-i', self.ip_address, '-n',
' '.join(workers)
]
if self.data_path is not None:
cmds.extend(['-d', self.data_path])
if self.force:
cmds.append('-f')
if client is not None:
cmds.extend(['-c', client])
popen = subprocess.Popen(cmds)
info.popen = popen
# call workers to start job
for worker_watcher in self.nodes_watchers:
client_call(worker_watcher,
'start_job',
zip_filename,
uncompress,
ignore=True)
def runLocalJob(master, job_path):
'''
push local job to cola cluster and run
'''
if not os.path.exists(job_path):
logger.error('Job path not exists!')
return
try:
import_job(job_path)
except (ImportError, AttributeError):
logger.error('Job path is illegal!')
return
start_log_server()
thread = start_rpc_server()
logger.info('Pushing job to cola cluster...')
dir_ = tempfile.mkdtemp()
try:
zip_filename = os.path.split(job_path)[1].replace(' ', '_') + '.zip'
zip_file = os.path.join(dir_, zip_filename)
ZipHandler.compress(zip_file, job_path, type_filters=("pyc", ))
FileTransportClient(master, zip_file).send_file()
logger.info('Push finished.')
finally:
shutil.rmtree(dir_)
logger.info('Start to run job.')
_client_call(master, 'start_job', zip_filename, True, client)
thread.join()
def start_job(self, zip_filename, uncompress=True, client=None):
if uncompress:
zip_file = os.path.join(self.zip_dir, zip_filename)
# transfer zip file to workers
for watcher in self.nodes_watchers:
if watcher.split(':')[0] == self.ip_address:
continue
file_trans_client = FileTransportClient(watcher, zip_file)
file_trans_client.send_file()
job_dir = ZipHandler.uncompress(zip_file, self.job_dir)
else:
job_dir = os.path.join(self.job_dir, zip_filename.rsplit('.', 1)[0])
job = import_job(job_dir)
worker_port = job.context.job.port
port = job.context.job.master_port
nodes = [watcher.split(':')[0] for watcher in self.nodes_watchers]
if len(nodes) > 0:
info = MasterJobInfo(port, nodes, worker_port)
self.running_jobs[job.real_name] = info
dirname = os.path.dirname(os.path.abspath(__file__))
f = os.path.join(dirname, 'loader.py')
workers = ['%s:%s'%(node, worker_port) for node in nodes]
cmds = ['python', f, '-j', job_dir, '-i', self.ip_address,
'-n', ' '.join(workers)]
if self.data_path is not None:
cmds.extend(['-d', self.data_path])
if self.force:
cmds.append('-f')
if client is not None:
cmds.extend(['-c', client])
popen = subprocess.Popen(cmds)
info.popen = popen
# call workers to start job
for worker_watcher in self.nodes_watchers:
client_call(worker_watcher, 'start_job', zip_filename, uncompress, ignore=True)
def start_job(self, zip_filename, uncompress=True, client=None):
if uncompress:
zip_file = os.path.join(self.zip_dir, zip_filename)
# transfer zip file to workers
for watcher in self.nodes_watchers:
if watcher.split(":")[0] == self.ip_address:
continue
file_trans_client = FileTransportClient(watcher, zip_file)
file_trans_client.send_file()
job_dir = ZipHandler.uncompress(zip_file, self.job_dir)
else:
job_dir = os.path.join(self.job_dir, zip_filename.rsplit(".", 1)[0])
job = import_job(job_dir)
worker_port = job.context.job.port
port = job.context.job.master_port
nodes = [watcher.split(":")[0] for watcher in self.nodes_watchers]
if len(nodes) > 0:
info = MasterJobInfo(port, nodes, worker_port)
self.running_jobs[job.real_name] = info
dirname = os.path.dirname(os.path.abspath(__file__))
f = os.path.join(dirname, "loader.py")
workers = ["%s:%s" % (node, worker_port) for node in nodes]
cmds = ["python", f, "-j", job_dir, "-i", self.ip_address, "-n", " ".join(workers)]
if self.data_path is not None:
cmds.extend(["-d", self.data_path])
if self.force:
cmds.append("-f")
if client is not None:
cmds.extend(["-c", client])
popen = subprocess.Popen(cmds)
info.popen = popen
# call workers to start job
for worker_watcher in self.nodes_watchers:
client_call(worker_watcher, "start_job", zip_filename, uncompress)
def pack_job_error(self, job_name):
working_dir = os.path.join(self.working_dir, job_name)
pack_dir = pack_local_job_error(job_name, working_dir=working_dir,
logger=self.logger)
zip_filename = os.path.join(self.zip_dir,
'%s_%s_errors.zip'%(self.ctx.ip.replace('.', '_'), job_name))
if os.path.exists(zip_filename):
os.remove(zip_filename)
ZipHandler.compress(zip_filename, pack_dir)
FileTransportClient(self.master, zip_filename).send_file()
def run(self, args):
master_addr = args.master
ctx = Context(is_client=True, master_addr=master_addr)
if args.list is True:
jobs = ctx.list_jobs()
self.logger.info('list jobs at master: %s' % ctx.master_addr)
for job_id, info in jobs.iteritems():
self.logger.info(
'====> job id: %s, job description: %s, status: %s' % \
(job_id, info['name'], info['status']))
if len(jobs) == 0:
self.logger.info('no jobs exist')
elif args.kill is not None:
job_id = self._get_matched_job_name(ctx, args.kill)
if job_id is not None:
ctx.kill_job(job_id)
self.logger.info('killed job: %s' % job_id)
elif args.upload is not None:
path = os.path.abspath(args.upload)
if not os.path.exists(path):
self.logger.error('upload path does not exist')
return
job_id = None
try:
job_id = import_job_desc(path).uniq_name
except Exception, e:
self.logger.exception(e)
self.logger.error('uploading job description failed')
return
new_upload_dir = os.path.join(tempfile.gettempdir(), job_id)
if os.path.exists(new_upload_dir):
shutil.rmtree(new_upload_dir)
shutil.copytree(path, new_upload_dir)
temp_filename = os.path.join(tempfile.gettempdir(),
job_id + '.zip')
ZipHandler.compress(temp_filename,
new_upload_dir,
type_filters=('pyc', ))
try:
FileTransportClient(ctx.master_addr, temp_filename).send_file()
finally:
os.remove(temp_filename)
shutil.rmtree(new_upload_dir)
self.logger.info('upload job <id: %s> finished' % job_id)
if args.run == 'U':
client_call(ctx.master_addr, 'run_job', job_id, True)
self.logger.info('submit job <id: %s> to the cluster' % job_id)
def start_job(self, zip_filename, uncompress=True):
if uncompress:
zip_file = os.path.join(self.zip_dir, zip_filename)
# transfer zip file to workers
for watcher in self.nodes_watchers:
if watcher.split(':')[0] == self.ip_address:
continue
file_trans_client = FileTransportClient(watcher, zip_file)
file_trans_client.send_file()
job_dir = ZipHandler.uncompress(zip_file, self.job_dir)
else:
job_dir = os.path.join(self.job_dir, zip_filename.rsplit('.', 1)[0])
job = import_job(job_dir)
worker_port = job.context.job.port
port = job.context.job.master_port
nodes = [watcher.split(':')[0] for watcher in self.nodes_watchers]
if len(nodes) > 0:
info = MasterJobInfo(port, nodes, worker_port)
self.running_jobs[job.real_name] = info
dirname = os.path.dirname(os.path.abspath(__file__))
f = os.path.join(dirname, 'loader.py')
workers = ['%s:%s'%(node, worker_port) for node in nodes]
subprocess.Popen('python "%(py)s" "%(job_dir)s" %(nodes)s' % {
'py': f,
'job_dir': job_dir,
'nodes': ' '.join(workers)
})
# call workers to start job
for worker_watcher in self.nodes_watchers:
client_call(worker_watcher, 'start_job', zip_filename, uncompress)
def run_job(self, job_name, unzip=False, wait_for_workers=False):
if wait_for_workers:
while not self.stopped.is_set():
if len(self.worker_tracker.workers) > 0:
break
stopped = self.stopped.wait(3)
if stopped:
return
if unzip:
self._unzip(job_name)
job_path = os.path.join(self.job_dir, job_name)
job_desc = import_job_desc(job_path)
job_master = JobMaster(self.ctx, job_name, job_desc,
self.worker_tracker.workers.keys())
job_master.init()
self.job_tracker.register_job(job_name, job_master)
self._register_runned_job(job_name, job_desc)
zip_file = os.path.join(self.zip_dir, job_name + '.zip')
for worker in job_master.workers:
FileTransportClient(worker, zip_file).send_file()
self.logger.debug('entering the master prepare stage, job id: %s' %
job_name)
self.logger.debug('job available workers: %s' % job_master.workers)
stage = Stage(job_master.workers, 'prepare')
prepared_ok = stage.barrier(True, job_name)
if not prepared_ok:
self.logger.error("prepare for running failed")
return
self.logger.debug('entering the master run_job stage, job id: %s' %
job_name)
stage = Stage(job_master.workers, 'run_job')
run_ok = stage.barrier(True, job_name)
if not run_ok:
self.logger.error("run job failed, job id: %s" % job_name)
