[ https://issues.apache.org/jira/browse/AIRFLOW-678?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15743009#comment-15743009 ]
ASF subversion and git services commented on AIRFLOW-678: --------------------------------------------------------- Commit 15ff540ecd5e60e7ce080177ea3ea227582a4672 in incubator-airflow's branch refs/heads/master from [~aoen] [ https://git-wip-us.apache.org/repos/asf?p=incubator-airflow.git;h=15ff540 ] [AIRFLOW-678] Prevent scheduler from double triggering TIs At the moment there is no lock/synchronization around the loop where the scheduler puts tasks in the SCHEDULED state. This means that if somehow the task starts running or gets SCHEDULED somewhere else somehow (e.g. manually running a task via the webserver) the task can have it's state changed from RUNNING/QUEUED to SCHEDULED which can cause a single task instance to be run twice at the same time. Testing Done: - Tested this branch on the Airbnb Airflow staging cluster - Airbnb has been running very similar logic in our production for many months (not 1-1 since we are still running off of the last release branch) - In the future we ideally need an integration test to catch double triggers but this is not trivial to do properly Closes #1924 from aoen/ddavydov/fix_scheduler_race_condition > Task instances can double-trigger > --------------------------------- > > Key: AIRFLOW-678 > URL: https://issues.apache.org/jira/browse/AIRFLOW-678 > Project: Apache Airflow > Issue Type: Bug > Reporter: Dan Davydov > Assignee: Dan Davydov > > At the moment there is no lock/synchronization around the loop where the > scheduler puts tasks in the SCHEDULED state. This means that if somehow the > task starts running or gets SCHEDULED somewhere else somehow (e.g. manually > running a task via the webserver) the task can have it's state changed from > RUNNING/QUEUED to SCHEDULED which can cause a single task instance to be run > twice at the same time. -- This message was sent by Atlassian JIRA (v6.3.4#6332)