def schedule(options, timeout):
"""
This celery task wraps the main work queue, and schedules individual builds on worker machines.
:param options: A dictionary containing build information to configure the worker with.
:param timeout: A value passed from Chorus to maintain uniformity with timeout configuration.
:return: A confirmation message indicating if the build was started successfully.
"""
LOG.info("Received Schedule Task Order!")
global SchedulerLock
while SchedulerLock.acquire(False) is False:
continue
filtered = []
while len(filtered) == 0:
filtered = []
raw = broadcast('busy', arguments={}, reply=True)
for item in raw:
for key, value in item.iteritems():
key = key.split('@')[1]
if key[0:6] == 'feeder' or key[0:6] == 'master':
continue
if value is False:
filtered.append(key)
break
time.sleep(5)
target = 'celery@' + filtered[0]
LOG.info("Scheduling to target: %s" % target)
options['host'] = target
result = broadcast('start_build', arguments={'options': options}, destination=[target], reply=True)
SchedulerLock.release()
return result
def capture(self, limit=None, timeout=None):
"""Open up a consumer capturing events.
This has to run in the main process, and it will never
stop unless forced via :exc:`KeyboardInterrupt` or :exc:`SystemExit`.
"""
consumer = self.consumer()
consumer.consume()
if self.wakeup:
from celery.task.control import broadcast
broadcast("heartbeat")
try:
for iteration in count(0):
if limit and iteration > limit:
break
try:
consumer.connection.drain_events(timeout=timeout)
except socket.timeout:
if timeout:
raise
except socket.error:
pass
finally:
consumer.close()
def handle(self, *args, **options):
try:
from celery.task.control import broadcast
except ImportError:
raise CommandError("Celery is not currently installed.")
# Shut them all down.
broadcast("shutdown")
def handle(self, *args, **options):
try:
from celery.task.control import broadcast
except ImportError:
raise CommandError("Celery is not currently installed.")
# Shut them all down.
broadcast("shutdown")
def tearDownClass(cls):
"""
Shut down the celeryd processes we created earlier.
"""
for hostname, hostinfo in cls.celeryd_hosts.items():
print "Requesting shutdown of celeryd [%s]..." % hostname
control.broadcast("shutdown", destination=[hostname])
cls.celeryd_hosts[hostname]["Popen"].wait()
print "celeryd closed."
def test_death(self):
"""
Test death broadcast command.
@param self Object reference.
"""
result = broadcast("death", arguments={"run_number":1}, reply=True)
result.sort()
print "death: %s " % result
self.validate_status(result)
# Expect subsequent attempt to succeed.
result = broadcast("death", arguments={"run_number":1}, reply=True)
self.validate_status(result)
def reset_worker(self): # pylint:disable = R0201
"""
Reset the Celery workers, via a broadcast, to have no
configuration and to use MapPyDoNothing.
@param self Object reference.
@return status of update.
"""
config_id = datetime.now().microsecond
broadcast("birth",
arguments={"configuration":self.config_doc,
"transform":"MapPyDoNothing",
"config_id":config_id}, reply=True)
def test_death(self):
"""
Test death broadcast command.
@param self Object reference.
"""
result = broadcast("death", arguments={"run_number": 1}, reply=True)
result.sort()
print "death: %s " % result
self.validate_status(result)
# Expect subsequent attempt to succeed.
result = broadcast("death", arguments={"run_number": 1}, reply=True)
self.validate_status(result)
def watcher():
try:
f = open(RUN_FLAG_FILE, 'r')
except IOError:
with open(RUN_FLAG_FILE, 'w') as f:
i = inspect()
print >>f, '1'
workers = i.active()
print >>f, '2'
if not all(workers.itervalues()):
pirnt >>f, '3'
broadcast('shutdown')
f.write('%s%s' % ('true', os.linesep))
def reset_worker(self): # pylint:disable = R0201
"""
Reset the Celery workers, via a broadcast, to have no
configuration and to use MapPyDoNothing.
@param self Object reference.
@return status of update.
"""
config_id = datetime.now().microsecond
broadcast("birth",
arguments={
"configuration": self.config_doc,
"transform": "MapPyDoNothing",
"config_id": config_id
},
reply=True)
def shrink_worker_pool(request, name=None, num=1):
""" Kills a running worker (celeryd). However, no action will be taken
unless the request method is 'POST'.
name:
The name of the worker to stop. The special value 'all' will stop
*all* workers.
"""
if request.method != 'POST':
return HttpResponseNotAllowed(['POST'])
else:
name = _resolve_name_param(name)
dest = name and [name] # dest will be None or a list of a single name
broadcast('pool_shrink', destination=dest, arguments={'n':num})
return HttpResponse('')
def birth_celery(transform, config, config_id, run_number, timeout = 1000):
"""
Set new configuration and transforms in Celery nodes, and
birth the transforms. An initial ping is done to
identify the number of live nodes.
Each node is given up to 1000s to reply
but if they all reply then the method returns sooner.
@param transform Either a single name can be given - representing
a single transform - or a list of transforms - representing a
MapPyGroup. Sub-lists are treated as nested MapPyGroups. If None
then the current transform is deathed and rebirthed.
@param config Valid JSON configuration document
@param config_id Configuration ID from client.
@param timeout Time to wait for replies.
@return results from Celery.
@throws RabbitMQException if RabbitMQ cannot be contacted.
@throws CeleryNodeException if one or more Celery
nodes fails to configure or birth.
"""
num_nodes = CeleryUtilities.ping_celery_nodes()
try:
response = broadcast("birth", arguments={
"transform": transform,
"configuration": config,
"config_id": config_id,
"run_number": run_number},
reply=True, timeout=timeout, limit=num_nodes)
except socket.error as exc:
raise RabbitMQException(exc)
CeleryUtilities.validate_celery_response(response)
run_headers = []
for machines in response:
for name in machines.keys():
run_headers += machines[name]["run_headers"]
return run_headers
def birth(self,
config_id,
configuration=None,
transform="MapPyDoNothing",
merge_configuration=False): # pylint:disable = R0201, C0301
"""
Configure the Celery workers, via a broadcast.
@param self Object reference.
@param config_id Configuration ID.
@param configuration Configuration.
@param transform Transform specification.
@return status of update.
"""
if configuration == None:
configuration = self.config_doc
elif merge_configuration:
my_config = json.loads(self.config_doc)
config_user = json.loads(configuration)
for key in config_user:
my_config[key] = config_user[key]
configuration = json.dumps(my_config)
return broadcast("birth",
arguments={"configuration":configuration, \
"transform":transform,
"config_id":config_id,
"run_number":1}, reply=True)
def kill_worker(request, name=None):
""" Kills a running worker (celeryd). However, no action will be taken
unless the request method is 'POST'.
name:
The name of the worker to stop. The special value 'all' will stop
*all* workers.
"""
if request.method != 'POST':
return HttpResponseNotAllowed(['POST'])
else:
name = _resolve_name_param(name)
dest = name and [name] # dest will be None or a list of a single name
print 'name: {0}'.format(name)
print 'dest: {0}'.format(dest)
broadcast('shutdown', destination=dest)
return HttpResponse('success')
def broadcast(self, name, *args, **kwargs):
if 'reply' not in kwargs:
kwargs['reply'] = True
result = broadcast(name, *args, **kwargs)
result = util._merge_broadcast_result(result) # turn it into a single dict
result = util._condense_broadcast_result(result) # remove worker key
for k,v in result.iteritems():
if isinstance(v, dict) and 'error' in v:
raise RuntimeError('Found error in broadcast()')
return result
def test_death_exception(self):
"""
Test death broadcast command where the transform throws
an exception when death is called.
@param self Object reference.
"""
# Set up a transform that will fail when it is deathed.
config_id = datetime.now().microsecond
transform = "MapPyTestMap"
configuration = """{"death_result":%s, "maus_version":"%s"}""" \
% (MapPyTestMap.EXCEPTION, self.__version)
result = self.birth(config_id, configuration, transform)
# Check the status is OK.
self.validate_status(result)
# Now death the transform.
result = broadcast("death", arguments={"run_number":1}, reply=True)
print "death(transform.death exception): %s " % result
self.validate_status(result, "error")
# Expect subsequent attempt to succeed.
result = broadcast("death", arguments={"run_number":1}, reply=True)
self.validate_status(result)
def test_death_exception(self):
"""
Test death broadcast command where the transform throws
an exception when death is called.
@param self Object reference.
"""
# Set up a transform that will fail when it is deathed.
config_id = datetime.now().microsecond
transform = "MapPyTestMap"
configuration = """{"death_result":%s, "maus_version":"%s"}""" \
% (MapPyTestMap.EXCEPTION, self.__version)
result = self.birth(config_id, configuration, transform)
# Check the status is OK.
self.validate_status(result)
# Now death the transform.
result = broadcast("death", arguments={"run_number": 1}, reply=True)
print "death(transform.death exception): %s " % result
self.validate_status(result, "error")
# Expect subsequent attempt to succeed.
result = broadcast("death", arguments={"run_number": 1}, reply=True)
self.validate_status(result)
def kill(host, bid):
"""
A celery task the kills a specific container on a specific host.
:param host: The host to send the kill message to.
:param bid: The bid of the build that created the container.
:return: A confirmation message from the stop_container control command.
"""
LOG.info("KILL! bid: %s on host: %s" % (bid, host))
raw = broadcast('stop_container', arguments={'bid': bid}, destination=[host], reply=True)
for r in raw:
ret = r.values()[0]
return ret
def command(self):
from celery.task.control import broadcast
self.parse_config()
self.log_break()
self.log('Starting celerystop...')
#
# Look for hosts to broadcast to in the command line arguments,
# e.g.,
# pecan celerystop config.py --hosts=8.17.172.226,8.17.12.225
# If no arguments are found, defaults to the hostname of the executing machine.
#
hosts = []
args = self.options.hosts
if args:
hosts = args.split(',')
else:
hosts = [gethostname()]
broadcast('shutdown', destination=hosts)
def get_worker_subprocesses(dest=None):
""" Retrieve the number of subprocesses for each worker. The return value
is a dict where the keys are worker names and the values are the number
of subprocesses.
"""
stats = {}
for x in broadcast("stats", destination=dest, reply=True):
stats.update(x)
workercounts = {}
for workername in stats.iterkeys():
procs = stats[workername]['pool']['processes']
workercounts[workername] = len(procs)
return workercounts
def retrieve_log_page(app_id, hostname, log_id):
"""
Get a page of log content.
"""
log_contents = broadcast(
"distlog_get_log_content",
destination=[hostname],
arguments={"app_id": app_id, "hostname": hostname, "log_id": log_id},
reply=True,
timeout=taskconfig.LOGS_CELERY_BCAST_TIMEOUT,
)
for host_response in log_contents:
for host, result in host_response.iteritems():
if result:
return result
return ""
def test_birth_map_cpp(self):
"""
Test birth can birth a MapCpp*. This is a regression of #1483
@param self Object reference.
"""
config_id = datetime.now().microsecond
transform = "MapCppExampleMAUSDataInput"
result = broadcast("birth",
arguments={"configuration":Configuration().getConfigJSON(),
"transform":transform,
"config_id":config_id,
"run_number":1}, reply=True)
print "birth(OK): %s " % result
for item in result:
for value in item.values():
self.assertEquals(value['status'], 'ok')
return True
def death_celery(run_number):
"""
Call death on transforms in Celery nodes.
@throws RabbitMQException if RabbitMQ cannot be contacted.
@throws CeleryNodeException if one or more Celery
nodes fails to death.
"""
try:
response = broadcast("death", arguments={"run_number":run_number},
reply=True)
except socket.error as exc:
raise RabbitMQException(exc)
CeleryUtilities.validate_celery_response(response)
run_footers = []
for machines in response:
for name in machines.keys():
run_footers += machines[name]["run_footers"]
return run_footers
def ping():
"""
Ping is one of the essential celery control commands Chorus uses to determine worker state.
:return: A list of results from control pong commands that provides Chorus with detailed state information.
"""
LOG.info("PING!")
myself = System.get_celery_hostname().split('@')[1]
State.master = True
raw = broadcast('pong', arguments={'rabbitmq': os.environ['CHORUS_HOST']}, reply=True)
filtered = []
for result in raw:
for key, value in result.iteritems():
key = key.split('@')[1]
if key == myself or key[0:6] == 'feeder':
continue
filtered.append([key] + value)
sorted_results = sorted(filtered)
return sorted_results
def validate_configuration(self, configuration, transform,
config_id = None):
"""
Validate workers have the given configuration using
set_maus_configuration and the same MAUS version as the
test class.
@param self Object reference.
@param configuration Expected configuration.
@param transform Expected transform specification.
@param config_id Expected configuration ID.
"""
result = broadcast("get_maus_configuration", reply=True)
print "get_maus_configuration: %s " % result
# Use built-in Celery worker inspection command to get
# worker names.
check_workers = self.__inspection.stats()
check_worker_names = check_workers.keys()
self.assertEquals(len(check_worker_names), len(result),
"Number of worker entries does not match that expected")
for worker in result:
worker_name = worker.keys()[0]
self.assertTrue(worker_name in check_worker_names,
"Cannot find entry for worker %s" % worker_name)
worker_config = worker[worker_name]
self.assertTrue(worker_config.has_key("config_id"),
"Configuration has no config_id entry")
if (config_id != None):
self.assertEquals(config_id, worker_config["config_id"],
"Unexpected config_id value")
self.assertTrue(worker_config.has_key("configuration"),
"Configuration has no configuration entry")
self.assertEquals(configuration,
worker_config["configuration"],
"Unexpected configuration value\n\n%s\n\n%s" % (configuration,
worker_config["configuration"]))
self.assertTrue(worker_config.has_key("transform"),
"Configuration has no transform entry")
self.assertEquals(transform, worker_config["transform"],
"Unexpected transform value")
self.assertTrue(worker_config.has_key("version"),
"Configuration has no version entry")
self.assertEquals(self.__version,
worker_config["version"],
"Unexpected version value")
def death_celery(run_number):
"""
Call death on transforms in Celery nodes.
@throws RabbitMQException if RabbitMQ cannot be contacted.
@throws CeleryNodeException if one or more Celery
nodes fails to death.
"""
try:
response = broadcast("death",
arguments={"run_number": run_number},
reply=True)
except socket.error as exc:
raise RabbitMQException(exc)
CeleryUtilities.validate_celery_response(response)
run_footers = []
for machines in response:
for name in machines.keys():
run_footers += machines[name]["run_footers"]
return run_footers
def validate_configuration(self, configuration, transform, config_id=None):
"""
Validate workers have the given configuration using
set_maus_configuration and the same MAUS version as the
test class.
@param self Object reference.
@param configuration Expected configuration.
@param transform Expected transform specification.
@param config_id Expected configuration ID.
"""
result = broadcast("get_maus_configuration", reply=True)
print "get_maus_configuration: %s " % result
# Use built-in Celery worker inspection command to get
# worker names.
check_workers = self.__inspection.stats()
check_worker_names = check_workers.keys()
self.assertEquals(
len(check_worker_names), len(result),
"Number of worker entries does not match that expected")
for worker in result:
worker_name = worker.keys()[0]
self.assertTrue(worker_name in check_worker_names,
"Cannot find entry for worker %s" % worker_name)
worker_config = worker[worker_name]
self.assertTrue(worker_config.has_key("config_id"),
"Configuration has no config_id entry")
if (config_id != None):
self.assertEquals(config_id, worker_config["config_id"],
"Unexpected config_id value")
self.assertTrue(worker_config.has_key("configuration"),
"Configuration has no configuration entry")
self.assertEquals(
configuration, worker_config["configuration"],
"Unexpected configuration value\n\n%s\n\n%s" %
(configuration, worker_config["configuration"]))
self.assertTrue(worker_config.has_key("transform"),
"Configuration has no transform entry")
self.assertEquals(transform, worker_config["transform"],
"Unexpected transform value")
self.assertTrue(worker_config.has_key("version"),
"Configuration has no version entry")
self.assertEquals(self.__version, worker_config["version"],
"Unexpected version value")
def test_process_after_death(self):
"""
Test process command fails after death has been called.
@param self Object reference.
"""
config_id = datetime.now().microsecond
transform = "MapPyTestMap"
configuration = """{"maus_version":"%s"}""" % self.__version
result = self.birth(config_id, configuration, transform)
self.validate_status(result)
result = broadcast("death", arguments={"run_number": 1}, reply=True)
self.validate_status(result)
# Call process.
result = execute_transform.delay("{}", 1)
# Wait for it to complete.
try:
result.wait()
except Exception: # pylint:disable = W0703
pass
self.assertTrue(result.failed(), "Expected failure")
def test_process_after_death(self):
"""
Test process command fails after death has been called.
@param self Object reference.
"""
config_id = datetime.now().microsecond
transform = "MapPyTestMap"
configuration = """{"maus_version":"%s"}""" % self.__version
result = self.birth(config_id, configuration, transform)
self.validate_status(result)
result = broadcast("death", arguments={"run_number":1}, reply=True)
self.validate_status(result)
# Call process.
result = execute_transform.delay("{}", 1)
# Wait for it to complete.
try:
result.wait()
except Exception: # pylint:disable = W0703
pass
self.assertTrue(result.failed(), "Expected failure")
def birth_celery(transform, config, config_id, run_number, timeout=1000):
"""
Set new configuration and transforms in Celery nodes, and
birth the transforms. An initial ping is done to
identify the number of live nodes.
Each node is given up to 1000s to reply
but if they all reply then the method returns sooner.
@param transform Either a single name can be given - representing
a single transform - or a list of transforms - representing a
MapPyGroup. Sub-lists are treated as nested MapPyGroups. If None
then the current transform is deathed and rebirthed.
@param config Valid JSON configuration document
@param config_id Configuration ID from client.
@param timeout Time to wait for replies.
@return results from Celery.
@throws RabbitMQException if RabbitMQ cannot be contacted.
@throws CeleryNodeException if one or more Celery
nodes fails to configure or birth.
"""
num_nodes = CeleryUtilities.ping_celery_nodes()
try:
response = broadcast("birth",
arguments={
"transform": transform,
"configuration": config,
"config_id": config_id,
"run_number": run_number
},
reply=True,
timeout=timeout,
limit=num_nodes)
except socket.error as exc:
raise RabbitMQException(exc)
CeleryUtilities.validate_celery_response(response)
run_headers = []
for machines in response:
for name in machines.keys():
run_headers += machines[name]["run_headers"]
return run_headers
def test_birth_map_cpp(self):
"""
Test birth can birth a MapCpp*. This is a regression of #1483
@param self Object reference.
"""
config_id = datetime.now().microsecond
transform = "MapCppExampleMAUSDataInput"
result = broadcast("birth",
arguments={
"configuration":
Configuration().getConfigJSON(),
"transform": transform,
"config_id": config_id,
"run_number": 1
},
reply=True)
print "birth(OK): %s " % result
for item in result:
for value in item.values():
self.assertEquals(value['status'], 'ok')
return True
def birth(self, config_id, configuration = None,
transform = "MapPyDoNothing", merge_configuration = False): # pylint:disable = R0201, C0301
"""
Configure the Celery workers, via a broadcast.
@param self Object reference.
@param config_id Configuration ID.
@param configuration Configuration.
@param transform Transform specification.
@return status of update.
"""
if configuration == None:
configuration = self.config_doc
elif merge_configuration:
my_config = json.loads(self.config_doc)
config_user = json.loads(configuration)
for key in config_user:
my_config[key] = config_user[key]
configuration = json.dumps(my_config)
return broadcast("birth",
arguments={"configuration":configuration, \
"transform":transform,
"config_id":config_id,
"run_number":1}, reply=True)
def find_available_logs(app_id):
"""
Query all log-keeping servers for logs pertaining to the given app.
"""
found_logs = broadcast(
"distlog_get_available_logs",
arguments={"app_id": app_id},
reply=True,
timeout=taskconfig.LOGS_CELERY_BCAST_TIMEOUT,
)
if not found_logs:
return LOG_SERVICE_NOT_AVAILABLE
result = []
for nodedict in found_logs:
for node, logs_from_node in nodedict.items():
if logs_from_node:
for loginfo in logs_from_node:
if not isinstance(loginfo, dict):
raise ValueError("Unexpected loginfo entry: " "%r in found_logs %r" % (loginfo, found_logs))
if "mod_time" in loginfo:
loginfo["mod_dt"] = datetime.datetime.fromtimestamp(loginfo["mod_time"])
result.append(loginfo)
return result
def enable_events(self, request, queryset):
broadcast("enable_events", destination=[n.hostname for n in queryset])
def enable_events(self, request, queryset):
broadcast('enable_events',
destination=[n.hostname for n in queryset])
def process(self, jobs, runner='unit2 $TEST', callback=None):
"""
``jobs`` is a list of path.to.TestCase strings to process.
``runner`` should be defined as command exectuable in bash, where $TEST is
the current job.
``callback`` will execute a callback after each result is returned, in
addition to return the aggregate of all results after completion.
"""
self.logger.info("Processing build %s", self.build_id)
self.logger.info("Provisioning (up to) %d worker(s)", self.max_workers)
actual = None
while not actual:
# We need to determine which queues are available to use
i = inspect()
active_queues = i.active_queues() or {}
if not active_queues:
self.logger.error('No queue workers available, retrying in 1s')
time.sleep(1)
continue
available = [host for host, queues in active_queues.iteritems() if conf.DEFAULT_QUEUE in [q['name'] for q in queues]]
if not available:
# TODO: we should probably sleep/retry (assuming there were *any* workers)
self.logger.info('All workers are busy, retrying in 1s')
time.sleep(1)
continue
# Attempt to provision workers which reported as available
actual = []
for su_response in broadcast('mule_setup',
arguments={'build_id': self.build_id,
'workspace': self.workspace,
'script': load_script(self.workspace, 'setup')},
destination=available[:self.max_workers],
reply=True,
timeout=0):
for host, message in su_response.iteritems():
if message.get('error'):
self.logger.error('%s failed to setup: %s', host, message['error'])
elif message.get('status') == 'ok':
actual.append(host)
if message.get('stdout'):
self.logger.info('stdout from %s: %s', host, message['stdout'])
if message.get('stderr'):
self.logger.info('stderr from %s: %s', host, message['stderr'])
if not actual:
# TODO: we should probably sleep/retry (assuming there were *any* workers)
self.logger.info('Failed to provision workers (busy), retrying in 1s')
time.sleep(1)
continue
if len(actual) != len(available):
# We should begin running tests and possibly add more, but its not a big deal
pass
self.logger.info('%d worker(s) were provisioned', len(actual))
self.logger.info("Building queue of %d test job(s)", len(jobs))
try:
taskset = TaskSet(run_test.subtask(
build_id=self.build_id,
runner=runner,
workspace=self.workspace,
job='%s.%s' % (job.__module__, job.__name__),
options={
# 'routing_key': 'mule-%s' % self.build_id,
'queue': 'mule-%s' % self.build_id,
# 'exchange': 'mule-%s' % self.build_id,
}) for job in jobs)
result = taskset.apply_async()
self.logger.info("Waiting for response...")
# response = result.join()
# propagate=False ensures we get *all* responses
response = []
try:
for task_response in result.iterate():
response.append(task_response)
if callback:
callback(task_response)
except KeyboardInterrupt, e:
print '\nReceived keyboard interrupt, closing workers.\n'
finally:
self.logger.info("Tearing down %d worker(s)", len(actual))
# Send off teardown task to all workers in pool
for td_response in broadcast('mule_teardown',
arguments={'build_id': self.build_id,
'workspace': self.workspace,
'script': load_script(self.workspace, 'teardown')},
destination=actual,
reply=True
):
for host, message in td_response.iteritems():
if message.get('error'):
self.logger.error('%s failed to teardown: %s', host, message['error'])
if message.get('stdout'):
self.logger.info('stdout from %s: %s', host, message['stdout'])
if message.get('stderr'):
self.logger.info('stderr from %s: %s', host, message['stderr'])
self.logger.info('Finished')
return response
def process(self, jobs, runner='unit2 $TEST', callback=None):
"""
``jobs`` is a list of path.to.TestCase strings to process.
``runner`` should be defined as command exectuable in bash, where $TEST is
the current job.
``callback`` will execute a callback after each result is returned, in
addition to return the aggregate of all results after completion.
"""
self.logger.info("Processing build %s", self.build_id)
self.logger.info("Provisioning (up to) %d worker(s)", self.max_workers)
actual = None
while not actual:
# We need to determine which queues are available to use
i = inspect()
active_queues = i.active_queues() or {}
if not active_queues:
self.logger.error('No queue workers available, retrying in 1s')
time.sleep(1)
continue
available = [
host for host, queues in active_queues.iteritems()
if conf.DEFAULT_QUEUE in [q['name'] for q in queues]
]
if not available:
# TODO: we should probably sleep/retry (assuming there were *any* workers)
self.logger.info('All workers are busy, retrying in 1s')
time.sleep(1)
continue
# Attempt to provision workers which reported as available
actual = []
for su_response in broadcast(
'mule_setup',
arguments={
'build_id': self.build_id,
'workspace': self.workspace,
'script': load_script(self.workspace, 'setup')
},
destination=available[:self.max_workers],
reply=True,
timeout=0):
for host, message in su_response.iteritems():
if message.get('error'):
self.logger.error('%s failed to setup: %s', host,
message['error'])
elif message.get('status') == 'ok':
actual.append(host)
if message.get('stdout'):
self.logger.info('stdout from %s: %s', host,
message['stdout'])
if message.get('stderr'):
self.logger.info('stderr from %s: %s', host,
message['stderr'])
if not actual:
# TODO: we should probably sleep/retry (assuming there were *any* workers)
self.logger.info(
'Failed to provision workers (busy), retrying in 1s')
time.sleep(1)
continue
if len(actual) != len(available):
# We should begin running tests and possibly add more, but its not a big deal
pass
self.logger.info('%d worker(s) were provisioned', len(actual))
self.logger.info("Building queue of %d test job(s)", len(jobs))
try:
taskset = TaskSet(
run_test.subtask(
build_id=self.build_id,
runner=runner,
workspace=self.workspace,
job='%s.%s' % (job.__module__, job.__name__),
options={
# 'routing_key': 'mule-%s' % self.build_id,
'queue': 'mule-%s' % self.build_id,
# 'exchange': 'mule-%s' % self.build_id,
}) for job in jobs)
result = taskset.apply_async()
self.logger.info("Waiting for response...")
# response = result.join()
# propagate=False ensures we get *all* responses
response = []
try:
for task_response in result.iterate():
response.append(task_response)
if callback:
callback(task_response)
except KeyboardInterrupt, e:
print '\nReceived keyboard interrupt, closing workers.\n'
finally:
self.logger.info("Tearing down %d worker(s)", len(actual))
# Send off teardown task to all workers in pool
for td_response in broadcast('mule_teardown',
arguments={
'build_id':
self.build_id,
'workspace':
self.workspace,
'script':
load_script(
self.workspace, 'teardown')
},
destination=actual,
reply=True):
for host, message in td_response.iteritems():
if message.get('error'):
self.logger.error('%s failed to teardown: %s', host,
message['error'])
if message.get('stdout'):
self.logger.info('stdout from %s: %s', host,
message['stdout'])
if message.get('stderr'):
self.logger.info('stderr from %s: %s', host,
message['stderr'])
self.logger.info('Finished')
return response
def disable_events(self, request, queryset):
broadcast("disable_events",
destination=[n.hostname for n in queryset])
def disable_events(self, request, queryset):
broadcast('disable_events', destination=[n.hostname for n in queryset])
nm.SetProp("_Name", str(i) + "_test")
mol_list_2.append(nm)
#removing duplicates and writing new generation
new_mols = ga_setup_selection.check_previous_gens(mol_list_2)
ga_setup_selection.new_gen_dumper(args.population_size, ng, new_mols)
else:
mol_list = cPickle.load(open(p_name, "rb"))
print len(mol_list)
run_mols = [
mol for mol in mol_list if mol.GetProp("_Energy") == "None"
]
print len(run_mols)
if len(run_mols) <= 1:
print "GA has hit generation of size 1, will quit now"
print "Shutting down workers"
broadcast("shutdown")
exit()
all_mol_coords = ga_setup_selection.coords_generator(run_mols)
if args.gauss_calc == "reorg_en":
gen_result = ga_gauss.reorg_en_calc(all_mol_coords)
if args.gauss_calc == "hl_diff":
gen_result = ga_gauss.hl_diff_calc(all_mol_coords)
if args.gauss_calc == "max_dipole":
gen_result = ga_gauss.max_dipole_calc(all_mol_coords)
if args.gauss_calc == "lumo":
gen_result = ga_gauss.lumo_calc(all_mol_coords)
print gen_result
if args.gauss_calc == "lumo" or "max_dipole" or "hl_diff":
ml_sorted_by_en = ga_setup_selection.result_matcher(
gen_result, mol_list, maximise="True")
if args.gauss_calc == "reorg_en":
def test_broadcast_limit(self):
control.broadcast("foobarbaz1", arguments=[], limit=None,
destination=[1, 2, 3])
self.assertIn("foobarbaz1", MockBroadcastPublisher.sent)
def test_broadcast(self):
control.broadcast("foobarbaz", arguments=[])
self.assertIn("foobarbaz", MockBroadcastPublisher.sent)
def shutdown_nodes(self, request, queryset):
broadcast("shutdown", destination=[n.hostname for n in queryset])
def ping_workers():
from celery.task.control import broadcast
broadcast('report_alive')
def shutdown_nodes(self, request, queryset):
broadcast('shutdown', destination=[n.hostname for n in queryset])
def shutdown_workers(workers):
return control.broadcast('shutdown', destination=workers)
def handle(self, *args, **options):
self.stdout.write("Sending shutdown event")
broadcast("shutdown")
self.stdout.write("Shutdown event sent")
def set_rate_limit(task_name,rate_limit,workers=None):
return control.broadcast("rate_limit",
{"task_name": "myapp.mytask",
"rate_limit": "200/m"}, reply=True,
destination=workers)