uranusjr commented on code in PR #71072:
URL: https://github.com/apache/airflow/pull/71072#discussion_r3732967216
##########
airflow-core/src/airflow/jobs/scheduler_job_runner.py:
##########
@@ -2247,13 +2256,55 @@ def _create_dagruns_for_partitioned_asset_dags(self,
session: Session) -> set[st
DagModel.is_stale.is_(False),
)
.order_by(AssetPartitionDagRun.created_at,
AssetPartitionDagRun.id)
- .limit(self._max_partition_dag_runs_per_loop),
+ .limit(self._max_partition_dag_runs_per_loop + 1),
of=AssetPartitionDagRun,
skip_locked=True,
key_share=False,
session=session,
)
).all()
+ has_backlog = len(rows) > self._max_partition_dag_runs_per_loop
+ pending_apdrs = rows[: self._max_partition_dag_runs_per_loop]
+ if has_backlog:
+ self.log.warning(
+ "Reached the per-tick cap on pending partitioned Dag runs; the
remaining backlog "
+ "will be evaluated over subsequent scheduler ticks",
+ cap=self._max_partition_dag_runs_per_loop,
+ pending_count=len(pending_apdrs),
Review Comment:
The cap+1 approach may have other issues with HA since we’re locking one
more row that does not belong to this process. Since we need a separate count
anyway, maybe something like this would be better:
```python
# No +1 in the limit!
pending_apdrs = session.scalars(with_row_locks(... .limit(cap), ...)).all()
if len(pending_apdrs) >= self._max_partition_dag_runs_per_loop:
backlog_total = session.scalar(count_query)
has_backlog = backlog_total > self._max_partition_dag_runs_per_loop
```
Also, if row count if a problem, we should add an index on APDR to prevent
table scan.
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
