AlexanderBLR opened a new pull request, #71380:
URL: https://github.com/apache/airflow/pull/71380

   `sync_running_jobs()` handles all active jobs in a single pass, and `sync()` 
wraps the whole heartbeat in one catch-all. When handling of a single job 
raises, the exception aborts the entire cycle: the remaining jobs are not 
synced and, more importantly, `attempt_submit_jobs()` never runs. If the error 
is deterministic for that job, the executor stops submitting tasks entirely 
until the scheduler is restarted.
   
   This is not hypothetical. `BatchJobCollection.pop_by_id()` raises `KeyError` 
before cleaning anything when the collection's bookkeeping is partially cleaned 
(e.g. a duplicate submission for the same workload key overwrites `key_to_id`, 
orphaning the older job id), so the failing job stays in `active_workers` 
forever and every subsequent heartbeat aborts at the same job. We hit exactly 
this in production (Airflow 3.2.1, `apache-airflow-providers-amazon==9.31.0`): 
one poisoned workload halted all Batch submissions cluster-wide (~20 Dags 
piling up in queued) until a scheduler restart, the only signal being `Failed 
to sync AwsBatchExecutor` repeating every heartbeat.
   
   The fix:
   
   - each job in `sync_running_jobs()` is handled in its own `try/except`; an 
unexpected per-job error evicts that job and fails its workload instead of 
aborting the cycle;
   - new `BatchJobCollection.remove_job()` used for the eviction — a defensive 
removal that tolerates partially cleaned state, unlike `pop_by_id()`.
   
   Credential-type boto errors (`ClientError` / `NoCredentialsError`) still 
reach `sync()`'s handler unchanged: they can only be raised by 
`_describe_jobs()` / `attempt_submit_jobs()`, which sit outside the per-job 
block — nothing inside the block calls AWS.
   
   Tests: two `BatchJobCollection.remove_job()` tests, eviction-with-isolation 
test (poisoned job evicted + failed, healthy job in the same cycle still 
succeeds), eviction survival when `fail()` itself raises, and a regression test 
for the production impact (`attempt_submit_jobs()` still runs when a tracked 
job cannot be synced).
   
   ---
   
   ##### Was generative AI tooling used to co-author this PR?
   
   - [X] Yes — Claude Code (Fable 5)
   
   Generated-by: Claude Code (Fable 5) following [the 
guidelines](https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#gen-ai-assisted-contributions)
   
   ---
   
   * Read the **[Pull Request 
Guidelines](https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#pull-request-guidelines)**
 for more information. Note: commit author/co-author name and email in commits 
become permanently public when merged.
   * For fundamental code changes, an Airflow Improvement Proposal 
([AIP](https://cwiki.apache.org/confluence/display/AIRFLOW/Airflow+Improvement+Proposals))
 is needed.
   * When adding dependency, check compliance with the [ASF 3rd Party License 
Policy](https://www.apache.org/legal/resolved.html#category-x).
   * For significant user-facing changes create newsfragment: 
`{pr_number}.significant.rst`, in 
[airflow-core/newsfragments](https://github.com/apache/airflow/tree/main/airflow-core/newsfragments).
 You can add this file in a follow-up commit after the PR is created so you 
know the PR number.
   


-- 
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]

Reply via email to