def __init__(self, root_task_descriptor, block_store, master_proxy,
execution_features, worker):
self._lock = Lock()
self.task_records = []
self.current_task = None
self.current_td = None
self.block_store = block_store
self.master_proxy = master_proxy
self.execution_features = execution_features
self.worker = worker
self.reference_cache = dict([
(ref.id, ref) for ref in root_task_descriptor["inputs"]
])
self.initial_td = root_task_descriptor
self.task_graph = LocalTaskGraph(execution_features,
[self.initial_td["task_id"]])
self.job_output = LocalJobOutput(self.initial_td["expected_outputs"])
for ref in self.initial_td["expected_outputs"]:
self.task_graph.subscribe(ref, self.job_output)
self.task_graph.spawn_and_publish([self.initial_td],
self.initial_td["inputs"])
def __init__(self, root_task_descriptor, block_store, master_proxy, execution_features, worker):
self._lock = Lock()
self.task_records = []
self.current_task = None
self.current_td = None
self.block_store = block_store
self.master_proxy = master_proxy
self.execution_features = execution_features
self.worker = worker
self.reference_cache = dict([(ref.id, ref) for ref in root_task_descriptor["inputs"]])
self.initial_td = root_task_descriptor
self.task_graph = LocalTaskGraph(execution_features, [self.initial_td["task_id"]])
self.job_output = LocalJobOutput(self.initial_td["expected_outputs"])
for ref in self.initial_td["expected_outputs"]:
self.task_graph.subscribe(ref, self.job_output)
self.task_graph.spawn_and_publish([self.initial_td], self.initial_td["inputs"])
def __init__(self, id, worker, tickets=1000000):
self.id = id
self.incoming_queues = {}
self.runnable_queues = {}
for scheduling_class in worker.scheduling_classes.keys():
self.incoming_queues[scheduling_class] = Queue.Queue()
self.runnable_queues[scheduling_class] = Queue.Queue()
self.reference_cache = {}
self.task_graph = LocalTaskGraph(worker.execution_features,
runnable_queues=self.runnable_queues)
self.active_or_queued_tasksets = 0
self.running_tasks = 0
self.active_tasksets = {}
self.tickets = tickets
self.job_aborted = False
self._tasksets_lock = threading.Lock()
self.task_cost = INITIAL_TASK_COST
class TaskSetExecutionRecord:
def __init__(self, root_task_descriptor, block_store, master_proxy, execution_features, worker):
self._lock = Lock()
self.task_records = []
self.current_task = None
self.current_td = None
self.block_store = block_store
self.master_proxy = master_proxy
self.execution_features = execution_features
self.worker = worker
self.reference_cache = dict([(ref.id, ref) for ref in root_task_descriptor["inputs"]])
self.initial_td = root_task_descriptor
self.task_graph = LocalTaskGraph(execution_features, [self.initial_td["task_id"]])
self.job_output = LocalJobOutput(self.initial_td["expected_outputs"])
for ref in self.initial_td["expected_outputs"]:
self.task_graph.subscribe(ref, self.job_output)
self.task_graph.spawn_and_publish([self.initial_td], self.initial_td["inputs"])
def run(self):
ciel.log.error("Running taskset starting at %s" % self.initial_td["task_id"], "TASKEXEC", logging.INFO)
while not self.job_output.is_complete():
next_td = self.task_graph.get_runnable_task()
if next_td is None:
ciel.log.error("No more runnable tasks", "TASKEXEC", logging.INFO)
break
next_td["inputs"] = [self.retrieve_ref(ref) for ref in next_td["dependencies"]]
task_record = TaskExecutionRecord(next_td, self, self.execution_features, self.block_store, self.master_proxy, self.worker)
with self._lock:
self.current_task = task_record
self.current_td = next_td
try:
task_record.run()
except:
ciel.log.error('Error during executor task execution', 'TASKEXEC', logging.ERROR, True)
with self._lock:
self.current_task.cleanup()
self.current_task = None
self.current_td = None
self.task_records.append(task_record)
if task_record.success:
self.task_graph.spawn_and_publish(task_record.spawned_tasks, task_record.published_refs, next_td)
else:
break
ciel.log.error("Taskset complete", "TASKEXEC", logging.INFO)
def retrieve_ref(self, ref):
if ref.is_consumable():
return ref
else:
try:
return self.reference_cache[ref.id]
except KeyError:
raise ReferenceUnavailableException(ref.id)
def publish_ref(self, ref):
self.reference_cache[ref.id] = ref
def abort_task(self, task_id):
with self._lock:
if self.current_td["task_id"] == task_id:
self.current_task.executor.abort()
class TaskSetExecutionRecord:
def __init__(self, root_task_descriptor, block_store, master_proxy,
execution_features, worker):
self._lock = Lock()
self.task_records = []
self.current_task = None
self.current_td = None
self.block_store = block_store
self.master_proxy = master_proxy
self.execution_features = execution_features
self.worker = worker
self.reference_cache = dict([
(ref.id, ref) for ref in root_task_descriptor["inputs"]
])
self.initial_td = root_task_descriptor
self.task_graph = LocalTaskGraph(execution_features,
[self.initial_td["task_id"]])
self.job_output = LocalJobOutput(self.initial_td["expected_outputs"])
for ref in self.initial_td["expected_outputs"]:
self.task_graph.subscribe(ref, self.job_output)
self.task_graph.spawn_and_publish([self.initial_td],
self.initial_td["inputs"])
def run(self):
ciel.log.error(
"Running taskset starting at %s" % self.initial_td["task_id"],
"TASKEXEC", logging.INFO)
while not self.job_output.is_complete():
next_td = self.task_graph.get_runnable_task()
if next_td is None:
ciel.log.error("No more runnable tasks", "TASKEXEC",
logging.INFO)
break
next_td["inputs"] = [
self.retrieve_ref(ref) for ref in next_td["dependencies"]
]
task_record = TaskExecutionRecord(next_td, self,
self.execution_features,
self.block_store,
self.master_proxy, self.worker)
with self._lock:
self.current_task = task_record
self.current_td = next_td
try:
task_record.run()
except:
ciel.log.error('Error during executor task execution',
'TASKEXEC', logging.ERROR, True)
with self._lock:
self.current_task.cleanup()
self.current_task = None
self.current_td = None
self.task_records.append(task_record)
if task_record.success:
self.task_graph.spawn_and_publish(task_record.spawned_tasks,
task_record.published_refs,
next_td)
else:
break
ciel.log.error("Taskset complete", "TASKEXEC", logging.INFO)
def retrieve_ref(self, ref):
if ref.is_consumable():
return ref
else:
try:
return self.reference_cache[ref.id]
except KeyError:
raise ReferenceUnavailableException(ref.id)
def publish_ref(self, ref):
self.reference_cache[ref.id] = ref
def abort_task(self, task_id):
with self._lock:
if self.current_td["task_id"] == task_id:
self.current_task.executor.abort()