[ https://issues.apache.org/jira/browse/AIRFLOW-972?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16786307#comment-16786307 ]
Jeff Stein commented on AIRFLOW-972: ------------------------------------ We ran into this with some third party code (rollout.io's [rox 4.0.1|https://pypi.org/project/rox/]) that was using the multiprocessing defaults (with the "fork" process start). They had a very short-lived process that they were calling `terminate()` on. Using something like this helped: {code:java} def signal_handler(signum, frame): print("Ignoring SIGTERM from bad dependency") sys.exit(0) old_handler = signal.getsignal(signal.SIGTERM) signal.signal(signal.SIGTERM, signal_handler) # call third-party code that uses multiprocessing fork processes ... signal.signal(signal.SIGTERM, old_handler){code} > Airflow kills subprocesses created by task instances > ---------------------------------------------------- > > Key: AIRFLOW-972 > URL: https://issues.apache.org/jira/browse/AIRFLOW-972 > Project: Apache Airflow > Issue Type: Bug > Components: scheduler > Affects Versions: 1.7.1 > Reporter: Richard Moorhead > Priority: Minor > > We have a task which creates multiple subprocesses via > [joblib|https://pythonhosted.org/joblib/parallel.html]; we're noticing that > airflow seems to kill the subprocesses prior to their completion. Is there > any way around this behavior? -- This message was sent by Atlassian JIRA (v7.6.3#76005)