aeroyorch opened a new issue, #72337:
URL: https://github.com/apache/airflow/issues/72337
### Under which category would you file this issue?
Airflow Core
### Apache Airflow version
3.2.2
### What happened and how to reproduce it?
After a backfill is cancelled, clearing one of its DAG runs sends the run
back to `queued`, where it
stays. In this case it had still not started four days later, with plenty of
idle capacity, and the
scheduler had never even looked at it: `last_scheduling_decision` was still
NULL and there was not a
single log line mentioning the run.
Cancelling sets both `is_paused=True` and `completed_at` on the backfill,
and never clears `is_paused`.
The scheduler's promotion query only looks at the flag:
```python
# DagRun.get_queued_dag_runs_to_set_running
not_(coalesce(cast("ColumnElement[bool]", Backfill.is_paused), False)),
```
So the flag is stale but permanent, and every run still attached to that
backfill is skipped on every
loop. `PUT /backfills/{id}/unpause` cannot help either, it returns `409
Backfill is already completed.`
**How to reproduce**
```python
from datetime import datetime
from airflow.sdk import dag, task
@dag(dag_id="cancelled_backfill_strands_run", schedule="@hourly",
start_date=datetime(2026, 1, 1), catchup=False)
def cancelled_backfill_strands_run():
@task
def work(): pass
work()
cancelled_backfill_strands_run()
```
1. Create a backfill over a past window and let at least one run finish.
2. `PUT /api/v2/backfills/{id}/cancel`.
3. Clear that finished run.
4. It returns to `queued` and never starts.
### What you think should happen instead?
Pausing a backfill should stop its runs while it is live. Once it has
completed, `is_paused` is
meaningless and should not keep its runs out of the scheduler.
### Operating System
_No response_
### Deployment
None
### Apache Airflow Provider(s)
_No response_
### Versions of Apache Airflow Providers
_No response_
### Official Helm Chart version
Not Applicable
### Kubernetes Version
_No response_
### Helm Chart configuration
_No response_
### Docker Image customizations
_No response_
### Anything else?
**Suggested fix**
```python
or_(
Backfill.completed_at.isnot(None),
not_(coalesce(cast("ColumnElement[bool]", Backfill.is_paused), False)),
)
```
A live pause still behaves as before, only the completed-and-paused
combination changes, and stranded runs
recover on the next loop with no manual intervention. PR #72054 already
proposes this exact diff, though
its regression test is filed against #69658, which turned out to be the
starvation issue rather than this
one.
### Are you willing to submit PR?
- [ ] Yes I am willing to submit a PR!
### Code of Conduct
- [x] I agree to follow this project's [Code of
Conduct](https://github.com/apache/airflow/blob/main/CODE_OF_CONDUCT.md)
--
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]
