def _dbnd_run_error(self, ex):
if (
# what scenario is this aiflow filtering supposed to help with?
# I had airflow put a default airflow.cfg in .dbnd causing validation error in k8sExecutor which was invisible in the console (only in task log)
("airflow" not in ex.__class__.__name__.lower()
or ex.__class__.__name__ == "AirflowConfigException")
and "Failed tasks are:" not in str(ex)
and not isinstance(ex, DatabandRunError)
and not isinstance(ex, KeyboardInterrupt)
and not isinstance(ex, DatabandSigTermError)):
logger.exception(ex)
if (isinstance(ex, KeyboardInterrupt)
or isinstance(ex, DatabandSigTermError) or self.is_killed()):
run_state = RunState.CANCELLED
unfinished_task_state = TaskRunState.UPSTREAM_FAILED
elif isinstance(ex, DatabandFailFastError):
run_state = RunState.FAILED
unfinished_task_state = TaskRunState.UPSTREAM_FAILED
else:
run_state = RunState.FAILED
unfinished_task_state = TaskRunState.FAILED
self.set_run_state(run_state)
self.tracker.tracking_store.set_unfinished_tasks_state(
run_uid=self.run_uid, state=unfinished_task_state)
err_banner_msg = self.describe.get_error_banner()
logger.error(u"\n\n{sep}\n{banner}\n{sep}".format(
sep=console_utils.ERROR_SEPARATOR, banner=err_banner_msg))
return DatabandRunError("Run has failed: %s" % ex,
run=self,
nested_exceptions=ex)
def _get_dbnd_run_relative_cmd():
argv = list(sys.argv)
while argv:
current = argv.pop(0)
if current == "run":
return argv
raise DatabandRunError(
"Can't calculate run command from '%s'",
help_msg="Check that it has a format of '..executable.. run ...'",
)
def do_run(self):
topological_tasks = topological_sort(
[tr.task for tr in self.task_runs])
fail_fast = self.settings.run.fail_fast
task_failed = False
task_runs_to_update_state = []
for task in topological_tasks:
tr = self.run.get_task_run_by_id(task.task_id)
if tr.is_reused:
continue
if fail_fast and task_failed:
logger.info("Setting %s to %s", task.task_id,
TaskRunState.UPSTREAM_FAILED)
tr.set_task_run_state(TaskRunState.UPSTREAM_FAILED,
track=False)
task_runs_to_update_state.append(tr)
continue
if self.run.is_killed():
logger.info(
"Databand Context is killed! Stopping %s to %s",
task.task_id,
TaskRunState.FAILED,
)
tr.set_task_run_state(TaskRunState.FAILED, track=False)
task_runs_to_update_state.append(tr)
continue
logger.debug("Executing task: %s", task.task_id)
try:
tr.runner.execute()
except DatabandSigTermError as e:
raise e
except Exception as e:
task_failed = True
logger.error("Failed to execute task '%s': %s" %
(task.task_id, str(e)))
if task_runs_to_update_state:
self.run.tracker.set_task_run_states(task_runs_to_update_state)
if task_failed:
err = _collect_errors(self.run.task_runs)
if err:
raise DatabandRunError(err)
def _execute(self, session=None):
"""
Initializes all components required to run a dag for a specified date range and
calls helper method to execute the tasks.
"""
# Trigger cleaning
if self.airflow_config.clean_zombies_during_backfill:
ClearZombieJob().run()
ti_status = BackfillJob._DagRunTaskStatus()
# picklin'
pickle_id = self.dag.pickle_id
# We don't need to pickle our dag again as it already pickled on job creattion
# also this will save it into databand table, that have no use for the airflow
# if not self.donot_pickle and self.executor.__class__ not in (
# executors.LocalExecutor,
# executors.SequentialExecutor,
# ):
# pickle_id = airflow_pickle(self.dag, session=session)
executor = self.executor
executor.start()
ti_status.total_runs = 1 # total dag runs in backfill
dag_run = None
try:
dag_run = self._get_dag_run(session=session)
# Create relation DagRun <> Job
dag_run.conf = {"job_id": self.id}
session.merge(dag_run)
session.commit()
run_date = dag_run.execution_date
if dag_run is None:
raise DatabandSystemError("Can't build dagrun")
tis_map = self._task_instances_for_dag_run(dag_run, session=session)
if not tis_map:
raise DatabandSystemError("There are no task instances to run!")
ti_status.active_runs.append(dag_run)
ti_status.to_run.update(tis_map or {})
processed_dag_run_dates = self._process_dag_task_instances(
ti_status=ti_status,
executor=executor,
pickle_id=pickle_id,
session=session,
)
ti_status.executed_dag_run_dates.update(processed_dag_run_dates)
err = self._collect_errors(ti_status=ti_status, session=session)
if err:
raise DatabandRunError("Airflow executor has failed to run the run")
if run_date not in ti_status.executed_dag_run_dates:
self.log.warning(
"Dag %s is not marked as completed! %s not found in %s",
self.dag_id,
run_date,
ti_status.executed_dag_run_dates,
)
finally:
# in sequential executor a keyboard interrupt would reach here and
# then executor.end() -> heartbeat() -> sync() will cause the queued commands
# to be run again before exiting
if hasattr(executor, "commands_to_run"):
executor.commands_to_run = []
try:
executor.end()
except Exception:
logger.exception("Failed to terminate executor")
if dag_run and dag_run.state == State.RUNNING:
_kill_dag_run_zombi(dag_run, session)
session.commit()
self.log.info("Run is completed. Exiting.")