seanghaeli opened a new issue, #76992: URL: https://github.com/apache/airflow/issues/76992
With `[core] multi_team` on, an asset watcher's trigger gets the team of the Dag that declares it, and only that team's triggerer runs it (#67605). The lookup that reuses an existing trigger ignores the team, so when two teams' Dags declare the same watcher on the same asset, both use the first team's trigger. The second team's asset gets events only while the first team's triggerer runs it: once that team stops scheduling on the asset and stops its triggerer, the trigger shows as orphaned and the other team's asset gets no more events, even though its own triggerer is up. The same gap shows up in a few other ways: - Identical watchers on different assets of two teams share a trigger when no Fernet key is set. With one set, the lookup never matches across Dag files, because it compares encrypted kwargs. - When one team declares the watcher and another only schedules on the asset, each of the second team's parses removes the watcher and the first team's adds it back, so the asset gets a new trigger every parse and never fires. - A watcher created before multi-team was turned on keeps no team, so no team's triggerer runs it. Suggestion: with multi-team on, have a Dag file's parse only reuse, add and remove its own team's watchers, and let another team's identical watcher stand in only while that team still schedules on the asset. -- 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]
