def execute_step_out_of_process(step_context, step, errors, term_events):
command = InProcessExecutorChildProcessCommand(
step_context.run_config,
step_context.pipeline_run,
step_context.executor_config,
step.key,
step_context.instance.get_ref(),
term_events[step.key],
)
yield DagsterEvent.engine_event(
step_context,
'Launching subprocess for {}'.format(step.key),
EngineEventData(marker_start=DELEGATE_MARKER),
step_key=step.key,
)
for ret in execute_child_process_command(command):
if ret is None or isinstance(ret, DagsterEvent):
yield ret
elif isinstance(ret, ChildProcessEvent):
if isinstance(ret, ChildProcessSystemErrorEvent):
errors[ret.pid] = ret.error_info
elif isinstance(ret, KeyboardInterrupt):
yield DagsterEvent.engine_event(
step_context,
'Multiprocess engine: received KeyboardInterrupt - forwarding to active child processes',
EngineEventData.interrupted(list(term_events.keys())),
)
for term_event in term_events.values():
term_event.set()
else:
check.failed(
'Unexpected return value from child process {}'.format(
type(ret)))
def bounded_parallel_executor(pipeline_context, step_contexts, limit):
pending_execution = list(step_contexts)
active_iters = {}
errors = {}
term_events = {}
stopping = False
while (not stopping and pending_execution) or active_iters:
try:
while len(active_iters
) < limit and pending_execution and not stopping:
step_context = pending_execution.pop(0)
step = step_context.step
term_events[step.key] = get_multiprocessing_context().Event()
active_iters[step.key] = execute_step_out_of_process(
step_context, step, errors, term_events)
empty_iters = []
for key, step_iter in active_iters.items():
try:
event_or_none = next(step_iter)
if event_or_none is None:
continue
else:
yield event_or_none
except StopIteration:
empty_iters.append(key)
for key in empty_iters:
del active_iters[key]
if term_events[key].is_set():
stopping = True
del term_events[key]
# In the very small chance that we get interrupted in this coordination section and not
# polling the subprocesses for events - try to clean up greacefully
except KeyboardInterrupt:
yield DagsterEvent.engine_event(
pipeline_context,
'Multiprocess engine: received KeyboardInterrupt - forwarding to active child processes',
EngineEventData.interrupted(list(term_events.keys())),
)
for event in term_events.values():
event.set()
errs = {pid: err for pid, err in errors.items() if err}
if errs:
raise DagsterSubprocessError(
'During multiprocess execution errors occured in child processes:\n{error_list}'
.format(error_list='\n'.join([
'In process {pid}: {err}'.format(pid=pid, err=err.to_string())
for pid, err in errs.items()
])),
subprocess_error_infos=list(errs.values()),
)
def execute(self, pipeline_context, execution_plan):
check.inst_param(pipeline_context, "pipeline_context",
SystemPipelineExecutionContext)
check.inst_param(execution_plan, "execution_plan", ExecutionPlan)
limit = self.max_concurrent
yield DagsterEvent.engine_event(
pipeline_context,
"Executing steps using multiprocess engine: parent process (pid: {pid})"
.format(pid=os.getpid()),
event_specific_data=EngineEventData.multiprocess(
os.getpid(),
step_keys_to_execute=execution_plan.step_keys_to_execute),
)
# It would be good to implement a reference tracking algorithm here so we could
# garbage collection results that are no longer needed by any steps
# https://github.com/dagster-io/dagster/issues/811
with time_execution_scope() as timer_result:
with execution_plan.start(
retries=self.retries) as active_execution:
active_iters = {}
errors = {}
term_events = {}
stopping = False
while (not stopping
and not active_execution.is_complete) or active_iters:
try:
# start iterators
while len(active_iters) < limit and not stopping:
steps = active_execution.get_steps_to_execute(
limit=(limit - len(active_iters)))
if not steps:
break
for step in steps:
step_context = pipeline_context.for_step(step)
term_events[step.key] = multiprocessing.Event()
active_iters[
step.
key] = self.execute_step_out_of_process(
step_context, step, errors,
term_events)
# process active iterators
empty_iters = []
for key, step_iter in active_iters.items():
try:
event_or_none = next(step_iter)
if event_or_none is None:
continue
else:
yield event_or_none
active_execution.handle_event(
event_or_none)
except ChildProcessCrashException as crash:
serializable_error = serializable_error_info_from_exc_info(
sys.exc_info())
yield DagsterEvent.engine_event(
pipeline_context,
("Multiprocess executor: child process for step {step_key} "
"unexpectedly exited with code {exit_code}"
).format(step_key=key,
exit_code=crash.exit_code),
EngineEventData.engine_error(
serializable_error),
step_key=key,
)
step_failure_event = DagsterEvent.step_failure_event(
step_context=pipeline_context.for_step(
active_execution.get_step_by_key(key)),
step_failure_data=StepFailureData(
error=serializable_error,
user_failure_data=None),
)
active_execution.handle_event(
step_failure_event)
yield step_failure_event
empty_iters.append(key)
except StopIteration:
empty_iters.append(key)
# clear and mark complete finished iterators
for key in empty_iters:
del active_iters[key]
if term_events[key].is_set():
stopping = True
del term_events[key]
active_execution.verify_complete(
pipeline_context, key)
# process skips from failures or uncovered inputs
for event in active_execution.skipped_step_events_iterator(
pipeline_context):
yield event
# In the very small chance that we get interrupted in this coordination section and not
# polling the subprocesses for events - try to clean up gracefully
except KeyboardInterrupt:
yield DagsterEvent.engine_event(
pipeline_context,
"Multiprocess engine: received KeyboardInterrupt - forwarding to active child processes",
EngineEventData.interrupted(
list(term_events.keys())),
)
stopping = True
for event in term_events.values():
event.set()
errs = {pid: err for pid, err in errors.items() if err}
if errs:
raise DagsterSubprocessError(
"During multiprocess execution errors occurred in child processes:\n{error_list}"
.format(error_list="\n".join([
"In process {pid}: {err}".format(
pid=pid, err=err.to_string())
for pid, err in errs.items()
])),
subprocess_error_infos=list(errs.values()),
)
yield DagsterEvent.engine_event(
pipeline_context,
"Multiprocess engine: parent process exiting after {duration} (pid: {pid})"
.format(duration=format_duration(timer_result.millis),
pid=os.getpid()),
event_specific_data=EngineEventData.multiprocess(os.getpid()),
)
def execute(pipeline_context, execution_plan):
check.inst_param(pipeline_context, 'pipeline_context',
SystemPipelineExecutionContext)
check.inst_param(execution_plan, 'execution_plan', ExecutionPlan)
intermediates_manager = pipeline_context.intermediates_manager
limit = pipeline_context.executor_config.max_concurrent
yield DagsterEvent.engine_event(
pipeline_context,
'Executing steps using multiprocess engine: parent process (pid: {pid})'
.format(pid=os.getpid()),
event_specific_data=EngineEventData.multiprocess(
os.getpid(),
step_keys_to_execute=execution_plan.step_keys_to_execute),
)
# It would be good to implement a reference tracking algorithm here so we could
# garbage collection results that are no longer needed by any steps
# https://github.com/dagster-io/dagster/issues/811
with time_execution_scope() as timer_result:
active_execution = execution_plan.start(
retries=pipeline_context.executor_config.retries)
active_iters = {}
errors = {}
term_events = {}
stopping = False
while (not stopping
and not active_execution.is_complete) or active_iters:
try:
# start iterators
while len(active_iters) < limit and not stopping:
steps = active_execution.get_steps_to_execute(
limit=(limit - len(active_iters)))
if not steps:
break
for step in steps:
step_context = pipeline_context.for_step(step)
term_events[
step.key] = get_multiprocessing_context(
).Event()
active_iters[
step.key] = execute_step_out_of_process(
step_context, step, errors, term_events)
# process active iterators
empty_iters = []
for key, step_iter in active_iters.items():
try:
event_or_none = next(step_iter)
if event_or_none is None:
continue
else:
yield event_or_none
active_execution.handle_event(event_or_none)
except StopIteration:
empty_iters.append(key)
# clear and mark complete finished iterators
for key in empty_iters:
del active_iters[key]
if term_events[key].is_set():
stopping = True
del term_events[key]
active_execution.verify_complete(pipeline_context, key)
# process skips from failures or uncovered inputs
for event in active_execution.skipped_step_events_iterator(
pipeline_context):
yield event
# In the very small chance that we get interrupted in this coordination section and not
# polling the subprocesses for events - try to clean up gracefully
except KeyboardInterrupt:
yield DagsterEvent.engine_event(
pipeline_context,
'Multiprocess engine: received KeyboardInterrupt - forwarding to active child processes',
EngineEventData.interrupted(list(term_events.keys())),
)
stopping = True
for event in term_events.values():
event.set()
errs = {pid: err for pid, err in errors.items() if err}
if errs:
raise DagsterSubprocessError(
'During multiprocess execution errors occurred in child processes:\n{error_list}'
.format(error_list='\n'.join([
'In process {pid}: {err}'.format(pid=pid,
err=err.to_string())
for pid, err in errs.items()
])),
subprocess_error_infos=list(errs.values()),
)
yield DagsterEvent.engine_event(
pipeline_context,
'Multiprocess engine: parent process exiting after {duration} (pid: {pid})'
.format(duration=format_duration(timer_result.millis),
pid=os.getpid()),
event_specific_data=EngineEventData.multiprocess(os.getpid()),
)
def execute(self, pipeline_context, execution_plan):
check.inst_param(pipeline_context, "pipeline_context",
SystemPipelineExecutionContext)
check.inst_param(execution_plan, "execution_plan", ExecutionPlan)
limit = self.max_concurrent
yield DagsterEvent.engine_event(
pipeline_context,
"Executing steps using multiprocess executor: parent process (pid: {pid})"
.format(pid=os.getpid()),
event_specific_data=EngineEventData.multiprocess(
os.getpid(),
step_keys_to_execute=execution_plan.step_keys_to_execute),
)
# It would be good to implement a reference tracking algorithm here so we could
# garbage collect results that are no longer needed by any steps
# https://github.com/dagster-io/dagster/issues/811
with time_execution_scope() as timer_result:
with execution_plan.start(
retry_mode=self.retries) as active_execution:
active_iters = {}
errors = {}
term_events = {}
stopping = False
while (not stopping
and not active_execution.is_complete) or active_iters:
if active_execution.check_for_interrupts():
yield DagsterEvent.engine_event(
pipeline_context,
"Multiprocess executor: received termination signal - "
"forwarding to active child processes",
EngineEventData.interrupted(
list(term_events.keys())),
)
stopping = True
active_execution.mark_interrupted()
for key, event in term_events.items():
event.set()
# start iterators
while len(active_iters) < limit and not stopping:
steps = active_execution.get_steps_to_execute(
limit=(limit - len(active_iters)))
if not steps:
break
for step in steps:
step_context = pipeline_context.for_step(step)
term_events[step.key] = multiprocessing.Event()
active_iters[
step.key] = self.execute_step_out_of_process(
step_context,
step,
errors,
term_events,
active_execution.get_known_state(),
)
# process active iterators
empty_iters = []
for key, step_iter in active_iters.items():
try:
event_or_none = next(step_iter)
if event_or_none is None:
continue
else:
yield event_or_none
active_execution.handle_event(event_or_none)
except ChildProcessCrashException as crash:
serializable_error = serializable_error_info_from_exc_info(
sys.exc_info())
yield DagsterEvent.engine_event(
pipeline_context,
("Multiprocess executor: child process for step {step_key} "
"unexpectedly exited with code {exit_code}"
).format(step_key=key,
exit_code=crash.exit_code),
EngineEventData.engine_error(
serializable_error),
step_handle=active_execution.get_step_by_key(
key).handle,
)
step_failure_event = DagsterEvent.step_failure_event(
step_context=pipeline_context.for_step(
active_execution.get_step_by_key(key)),
step_failure_data=StepFailureData(
error=serializable_error,
user_failure_data=None),
)
active_execution.handle_event(step_failure_event)
yield step_failure_event
empty_iters.append(key)
except StopIteration:
empty_iters.append(key)
# clear and mark complete finished iterators
for key in empty_iters:
del active_iters[key]
del term_events[key]
active_execution.verify_complete(pipeline_context, key)
# process skipped and abandoned steps
yield from active_execution.plan_events_iterator(
pipeline_context)
errs = {pid: err for pid, err in errors.items() if err}
# After termination starts, raise an interrupted exception once all subprocesses
# have finished cleaning up (and the only errors were from being interrupted)
if (stopping and (not active_iters) and all([
err_info.cls_name == "DagsterExecutionInterruptedError"
for err_info in errs.values()
])):
yield DagsterEvent.engine_event(
pipeline_context,
"Multiprocess executor: interrupted all active child processes",
event_specific_data=EngineEventData(),
)
raise DagsterExecutionInterruptedError()
elif errs:
raise DagsterSubprocessError(
"During multiprocess execution errors occurred in child processes:\n{error_list}"
.format(error_list="\n".join([
"In process {pid}: {err}".format(
pid=pid, err=err.to_string())
for pid, err in errs.items()
])),
subprocess_error_infos=list(errs.values()),
)
yield DagsterEvent.engine_event(
pipeline_context,
"Multiprocess executor: parent process exiting after {duration} (pid: {pid})"
.format(duration=format_duration(timer_result.millis),
pid=os.getpid()),
event_specific_data=EngineEventData.multiprocess(os.getpid()),
)
def core_celery_execution_loop(pipeline_context, execution_plan,
step_execution_fn):
check.inst_param(pipeline_context, "pipeline_context",
SystemPipelineExecutionContext)
check.inst_param(execution_plan, "execution_plan", ExecutionPlan)
check.callable_param(step_execution_fn, "step_execution_fn")
executor = pipeline_context.executor
# https://github.com/dagster-io/dagster/issues/2440
check.invariant(
execution_plan.artifacts_persisted,
"Cannot use in-memory storage with Celery, use filesystem (on top of NFS or "
"similar system that allows files to be available to all nodes), S3, or GCS",
)
app = make_app(executor.app_args())
priority_for_step = lambda step: (-1 * int(
step.tags.get(DAGSTER_CELERY_STEP_PRIORITY_TAG, task_default_priority)
) + -1 * _get_run_priority(pipeline_context))
priority_for_key = lambda step_key: (priority_for_step(
execution_plan.get_step_by_key(step_key)))
_warn_on_priority_misuse(pipeline_context, execution_plan)
step_results = {} # Dict[ExecutionStep, celery.AsyncResult]
step_errors = {}
with execution_plan.start(
retries=pipeline_context.executor.retries,
sort_key_fn=priority_for_step,
) as active_execution:
stopping = False
while (not active_execution.is_complete
and not stopping) or step_results:
if active_execution.check_for_interrupts():
yield DagsterEvent.engine_event(
pipeline_context,
"Celery executor: received termination signal - revoking active tasks from workers",
EngineEventData.interrupted(list(step_results.keys())),
)
stopping = True
active_execution.mark_interrupted()
for result in step_results.values():
result.revoke()
results_to_pop = []
for step_key, result in sorted(
step_results.items(),
key=lambda x: priority_for_key(x[0])):
if result.ready():
try:
step_events = result.get()
except TaskRevokedError:
step_events = []
yield DagsterEvent.engine_event(
pipeline_context,
'celery task for running step "{step_key}" was revoked.'
.format(step_key=step_key, ),
EngineEventData(marker_end=DELEGATE_MARKER),
step_handle=active_execution.get_step_by_key(
step_key).handle,
)
except Exception: # pylint: disable=broad-except
# We will want to do more to handle the exception here.. maybe subclass Task
# Certainly yield an engine or pipeline event
step_events = []
step_errors[
step_key] = serializable_error_info_from_exc_info(
sys.exc_info())
for step_event in step_events:
event = deserialize_json_to_dagster_namedtuple(
step_event)
yield event
active_execution.handle_event(event)
results_to_pop.append(step_key)
for step_key in results_to_pop:
if step_key in step_results:
del step_results[step_key]
active_execution.verify_complete(pipeline_context,
step_key)
# process skips from failures or uncovered inputs
for event in active_execution.plan_events_iterator(
pipeline_context):
yield event
# don't add any new steps if we are stopping
if stopping or step_errors:
continue
# This is a slight refinement. If we have n workers idle and schedule m > n steps for
# execution, the first n steps will be picked up by the idle workers in the order in
# which they are scheduled (and the following m-n steps will be executed in priority
# order, provided that it takes longer to execute a step than to schedule it). The test
# case has m >> n to exhibit this behavior in the absence of this sort step.
for step in active_execution.get_steps_to_execute():
try:
queue = step.tags.get(DAGSTER_CELERY_QUEUE_TAG,
task_default_queue)
yield DagsterEvent.engine_event(
pipeline_context,
'Submitting celery task for step "{step_key}" to queue "{queue}".'
.format(step_key=step.key, queue=queue),
EngineEventData(marker_start=DELEGATE_MARKER),
step_handle=step.handle,
)
# Get the Celery priority for this step
priority = _get_step_priority(pipeline_context, step)
# Submit the Celery tasks
step_results[step.key] = step_execution_fn(
app, pipeline_context, step, queue, priority)
except Exception:
yield DagsterEvent.engine_event(
pipeline_context,
"Encountered error during celery task submission.".
format(),
event_specific_data=EngineEventData.engine_error(
serializable_error_info_from_exc_info(
sys.exc_info()), ),
)
raise
time.sleep(TICK_SECONDS)
if step_errors:
raise DagsterSubprocessError(
"During celery execution errors occurred in workers:\n{error_list}"
.format(error_list="\n".join([
"[{step}]: {err}".format(step=key, err=err.to_string())
for key, err in step_errors.items()
])),
subprocess_error_infos=list(step_errors.values()),
)
def execute(self, plan_context: PlanOrchestrationContext,
execution_plan: ExecutionPlan):
check.inst_param(plan_context, "plan_context",
PlanOrchestrationContext)
check.inst_param(execution_plan, "execution_plan", ExecutionPlan)
self._event_cursor = -1 # pylint: disable=attribute-defined-outside-init
yield DagsterEvent.engine_event(
plan_context,
f"Starting execution with step handler {self._step_handler.name}",
EngineEventData(),
)
with execution_plan.start(retry_mode=self.retries) as active_execution:
running_steps: Dict[str, ExecutionStep] = {}
if plan_context.resume_from_failure:
yield DagsterEvent.engine_event(
plan_context,
"Resuming execution from failure",
EngineEventData(),
)
prior_events = self._pop_events(
plan_context.instance,
plan_context.run_id,
)
for dagster_event in prior_events:
yield dagster_event
possibly_in_flight_steps = active_execution.rebuild_from_events(
prior_events)
for step in possibly_in_flight_steps:
yield DagsterEvent.engine_event(
plan_context,
"Checking on status of possibly launched steps",
EngineEventData(),
step.handle,
)
# TODO: check if failure event included. For now, hacky assumption that
# we don't log anything on successful check
if self._step_handler.check_step_health(
self._get_step_handler_context(
plan_context, [step], active_execution)):
# health check failed, launch the step
self._log_new_events(
self._step_handler.launch_step(
self._get_step_handler_context(
plan_context, [step], active_execution)),
plan_context,
{
step.key: step
for step in possibly_in_flight_steps
},
)
running_steps[step.key] = step
last_check_step_health_time = pendulum.now("UTC")
# Order of events is important here. During an interation, we call handle_event, then get_steps_to_execute,
# then is_complete. get_steps_to_execute updates the state of ActiveExecution, and without it
# is_complete can return true when we're just between steps.
while not active_execution.is_complete:
if active_execution.check_for_interrupts():
if not plan_context.instance.run_will_resume(
plan_context.run_id):
yield DagsterEvent.engine_event(
plan_context,
"Executor received termination signal, forwarding to steps",
EngineEventData.interrupted(
list(running_steps.keys())),
)
active_execution.mark_interrupted()
for _, step in running_steps.items():
self._log_new_events(
self._step_handler.terminate_step(
self._get_step_handler_context(
plan_context, [step],
active_execution)),
plan_context,
running_steps,
)
else:
yield DagsterEvent.engine_event(
plan_context,
"Executor received termination signal, not forwarding to steps because "
"run will be resumed",
EngineEventData(metadata_entries=[
EventMetadataEntry.text(
str(running_steps.keys()),
"steps_in_flight")
]),
)
active_execution.mark_interrupted()
return
for dagster_event in self._pop_events(
plan_context.instance,
plan_context.run_id,
): # type: ignore
# STEP_SKIPPED events are only emitted by ActiveExecution, which already handles
# and yields them.
if dagster_event.is_step_skipped:
assert isinstance(dagster_event.step_key, str)
active_execution.verify_complete(
plan_context, dagster_event.step_key)
else:
yield dagster_event
active_execution.handle_event(dagster_event)
if dagster_event.is_step_success or dagster_event.is_step_failure:
assert isinstance(dagster_event.step_key, str)
del running_steps[dagster_event.step_key]
active_execution.verify_complete(
plan_context, dagster_event.step_key)
# process skips from failures or uncovered inputs
for event in active_execution.plan_events_iterator(
plan_context):
yield event
curr_time = pendulum.now("UTC")
if (curr_time - last_check_step_health_time).total_seconds(
) >= self._check_step_health_interval_seconds:
last_check_step_health_time = curr_time
for _, step in running_steps.items():
self._log_new_events(
self._step_handler.check_step_health(
self._get_step_handler_context(
plan_context, [step], active_execution)),
plan_context,
running_steps,
)
for step in active_execution.get_steps_to_execute():
running_steps[step.key] = step
self._log_new_events(
self._step_handler.launch_step(
self._get_step_handler_context(
plan_context, [step], active_execution)),
plan_context,
running_steps,
)
time.sleep(self._sleep_seconds)
def execute(self, pipeline_context: PlanOrchestrationContext,
execution_plan: ExecutionPlan):
check.inst_param(pipeline_context, "pipeline_context",
PlanOrchestrationContext)
check.inst_param(execution_plan, "execution_plan", ExecutionPlan)
self._event_cursor = -1 # pylint: disable=attribute-defined-outside-init
yield DagsterEvent.engine_event(
pipeline_context,
f"Starting execution with step handler {self._step_handler.name}",
EngineEventData(),
)
with execution_plan.start(retry_mode=self.retries) as active_execution:
stopping = False
running_steps: Dict[str, ExecutionStep] = {}
last_check_step_health_time = pendulum.now("UTC")
while (not active_execution.is_complete
and not stopping) or running_steps:
events = []
if active_execution.check_for_interrupts():
yield DagsterEvent.engine_event(
pipeline_context,
"Executor received termination signal, forwarding to steps",
EngineEventData.interrupted(list(
running_steps.keys())),
)
stopping = True
active_execution.mark_interrupted()
for _, step in running_steps.items():
events.extend(
self._log_new_events(
self._step_handler.terminate_step(
self._get_step_handler_context(
pipeline_context, [step],
active_execution)),
pipeline_context,
running_steps,
))
running_steps.clear()
events.extend(
self._pop_events(
pipeline_context.plan_data.instance,
pipeline_context.plan_data.pipeline_run.run_id,
))
if not stopping:
curr_time = pendulum.now("UTC")
if (curr_time - last_check_step_health_time).total_seconds(
) >= self._check_step_health_interval_seconds:
last_check_step_health_time = curr_time
for _, step in running_steps.items():
events.extend(
self._log_new_events(
self._step_handler.check_step_health(
self._get_step_handler_context(
pipeline_context, [step],
active_execution)),
pipeline_context,
running_steps,
))
for step in active_execution.get_steps_to_execute():
running_steps[step.key] = step
events.extend(
self._log_new_events(
self._step_handler.launch_step(
self._get_step_handler_context(
pipeline_context, [step],
active_execution)),
pipeline_context,
running_steps,
))
for dagster_event in events:
yield dagster_event
active_execution.handle_event(dagster_event)
if (dagster_event.is_step_success
or dagster_event.is_step_failure
or dagster_event.is_step_skipped):
assert isinstance(dagster_event.step_key, str)
del running_steps[dagster_event.step_key]
active_execution.verify_complete(
pipeline_context, dagster_event.step_key)
# process skips from failures or uncovered inputs
for event in active_execution.plan_events_iterator(
pipeline_context):
yield event
time.sleep(self._sleep_seconds)
