ashb commented on a change in pull request #16401:
URL: https://github.com/apache/airflow/pull/16401#discussion_r651742864
##########
File path: airflow/jobs/scheduler_job.py
##########
@@ -1591,89 +1567,85 @@ def _create_dag_runs(self, dag_models:
Iterable[DagModel], session: Session) ->
# are not updated.
# We opted to check DagRun existence instead
# of catching an Integrity error and rolling back the session i.e
- # we need to run self._update_dag_next_dagruns if the Dag Run
already exists or if we
+ # we need to set dag.next_dagrun_info if the Dag Run already
exists or if we
# create a new one. This is so that in the next Scheduling loop we
try to create new runs
# instead of falling in a loop of Integrity Error.
- if (dag.dag_id, dag_model.next_dagrun) not in active_dagruns:
- run = dag.create_dagrun(
+ if (dag.dag_id, dag_model.next_dagrun) not in existing_dagruns:
+ dag.create_dagrun(
run_type=DagRunType.SCHEDULED,
execution_date=dag_model.next_dagrun,
- start_date=timezone.utcnow(),
- state=State.RUNNING,
+ state=State.QUEUED,
external_trigger=False,
session=session,
dag_hash=dag_hash,
creating_job_id=self.id,
)
-
- expected_start_date =
dag.following_schedule(run.execution_date)
- if expected_start_date:
- schedule_delay = run.start_date - expected_start_date
- Stats.timing(
- f'dagrun.schedule_delay.{dag.dag_id}',
- schedule_delay,
- )
-
- self._update_dag_next_dagruns(dag_models, session)
+ dag_model.next_dagrun, dag_model.next_dagrun_create_after =
dag.next_dagrun_info(
+ dag_model.next_dagrun
+ )
# TODO[HA]: Should we do a session.flush() so we don't have to keep
lots of state/object in
# memory for larger dags? or expunge_all()
- def _update_dag_next_dagruns(self, dag_models: Iterable[DagModel],
session: Session) -> None:
+ def _start_dagrun(
+ self,
+ guard,
+ session: Session,
+ ) -> int:
"""
- Bulk update the next_dagrun and next_dagrun_create_after for all the
dags.
-
- We batch the select queries to get info about all the dags at once
+ Find DagRuns in queued state and decide moving them to running state
+ :param dag_run: The DagRun to schedule
"""
- # Check max_active_runs, to see if we are _now_ at the limit for any of
- # these dag? (we've just created a DagRun for them after all)
- active_runs_of_dags = dict(
- session.query(DagRun.dag_id, func.count('*'))
- .filter(
- DagRun.dag_id.in_([o.dag_id for o in dag_models]),
- DagRun.state == State.RUNNING, # pylint:
disable=comparison-with-callable
- DagRun.external_trigger == expression.false(),
- )
- .group_by(DagRun.dag_id)
- .all()
- )
+ dag_runs = self._get_next_queued_dagruns_to_examine(session)
+
+ def _update_state(dag_run):
+ dag_run.state = State.RUNNING
+ dag_run.start_date = timezone.utcnow()
+ expected_start_date =
dag.following_schedule(dag_run.execution_date)
+ if expected_start_date:
+ schedule_delay = dag_run.start_date - expected_start_date
+ Stats.timing(
+ f'dagrun.schedule_delay.{dag.dag_id}',
+ schedule_delay,
+ )
+ # commit the session - Release the write lock on DagModel table.
+ # Commit one dag_run at a time
- for dag_model in dag_models:
- # Get the DAG in a try_except to not stop the Scheduler when a
Serialized DAG is not found
- # This takes care of Dynamic DAGs especially
+ guard.commit()
+ # END: create dagruns
+
+ for dag_run in dag_runs:
try:
- dag = self.dagbag.get_dag(dag_model.dag_id, session=session)
+ dag = dag_run.dag = self.dagbag.get_dag(dag_run.dag_id,
session=session)
except SerializedDagNotFound:
- self.log.exception("DAG '%s' not found in serialized_dag
table", dag_model.dag_id)
+ self.log.exception("DAG '%s' not found in serialized_dag
table", dag_run.dag_id)
continue
- active_runs_of_dag = active_runs_of_dags.get(dag.dag_id, 0)
- if dag.max_active_runs and active_runs_of_dag >=
dag.max_active_runs:
- self.log.info(
- "DAG %s is at (or above) max_active_runs (%d of %d), not
creating any more runs",
+ # pylint: disable=comparison-with-callable
+ total_active_dagruns = (
+ session.query(func.count(DR.dag_id))
+ .filter(DR.state == State.RUNNING, DR.dag_id == dag.dag_id)
+ .scalar()
+ )
Review comment:
Since the dag runs are locked, no other scheduler will be able to update
them, so the info won't be stale.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
