rjgoyln opened a new pull request, #73131: URL: https://github.com/apache/airflow/pull/73131
## Summary A deferrable `KubernetesPodOperator` with `do_xcom_push=True` is marked SUCCESS while pushing no `return_value` whenever the pod is reclaimed between the trigger firing and the worker resuming the task. The XCom sidecar goes with the pod, and `trigger_reentry`'s return value *is* the task's `return_value`, so the 404 shortcut added in #66716 returns `None` and reports success with no result. Downstream tasks then fail at templating time on a `None` they cannot explain, and only clearing the upstream task recovers it. - Keep the silent success only when `do_xcom_push` is unset, where the pod logs are the only loss - Raise `PodNotFoundException` otherwise, so the retry re-runs the pod Of the two approaches the issue proposed, this is the first — no opt-in flag. The crash #66716 fixed stays fixed: the raise is still ahead of `_clean`, which never sees `self.pod` as `None`. The new test fails on `main`. The existing GC'd-on-success test now states `do_xcom_push=False` explicitly, since it pins the surviving shortcut rather than an unconditional one. closes: #73117 --- ##### Was generative AI tooling used to co-author this PR? - [X] Yes — Claude Code (Opus 5) Generated-by: Claude Code (Opus 5) following [the guidelines](https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#gen-ai-assisted-contributions) -- 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]
