You are viewing a plain text version of this content. The canonical link for it is here.
Posted to jobs@airflow.apache.org by GitBox <gi...@apache.org> on 2026/06/23 11:12:10 UTC
[GH] (airflow/fix-ti-history-retry-metadata): Workflow run "Check newsfragment PR number" failed!
The GitHub Actions job "Check newsfragment PR number" on airflow.git/fix-ti-history-retry-metadata has failed.
Run started by GitHub user kalluripradeep (triggered by kalluripradeep).
Head commit for run:
5bab7ede42b426b70084c98a08800346bfe22996 / Pradeep Kalluri <12...@users.noreply.github.com>
Fix KubernetesExecutor retry losing hostname
When a task fails from a non-RUNNING state (QUEUED, SCHEDULED, or DEFERRED)
— e.g. a Kubernetes pod OOMKilled before the task process starts — the
scheduler's executor-event path calls handle_failure(). Previously,
prepare_db_for_next_try() was only invoked when ti.state == RUNNING, so
TaskInstanceHistory was never recorded for non-RUNNING retries, losing the
hostname and start_date and causing 'http://:8793 No host supplied' in the
log UI.
Changes:
- taskinstance.py: extend prepare_db_for_next_try() guard from
state == RUNNING to state != RESTARTING. RESTARTING already had it
called during the clear operation; all other retryable states now get it.
- taskinstancehistory.py: use `ti.end_date or utcnow()` so a QUEUED TI
without an end_date still gets a valid timestamp in history.
- task_instances.py (execution API): capture end_date, duration, and
rendered_map_index before prepare_db_for_next_try() clears them.
- Tests: verify hostname/start_date preserved for QUEUED/SCHEDULED retry
and that rendered_map_index is stored in TaskInstanceHistory.
closes: #65366
closes: #67238
Report URL: https://github.com/apache/airflow/actions/runs/28021986283
With regards,
GitHub Actions via GitBox
---------------------------------------------------------------------
To unsubscribe, e-mail: jobs-unsubscribe@airflow.apache.org
For additional commands, e-mail: jobs-help@airflow.apache.org