Orphaned/zombie LocalTaskJob

[Mzhussiyev]

Several times I spot zombie/orphaned LocalTaskJobs when I was launching an Airflow. There are no active running TI and DAG runs but there are LocalTaskJobs with running state (see the screenshot below).

I thought that such orphaned tasks/jobs should be killed by the scheduler (_find_zombies function).
If you look at job runs their last heartbeat were a while ago and these jobs are still alive with running state, the current date and time you can get from latest heartbeat of SchedulerJob.

So, the questions are:

• why these jobs weren't killed by scheduler
• which process is responsible to look after such orphaned jobs (which probably lost the connection with tasks)
• are these "running" jobs affect to Airflow performance by consuming some resources
• what should we consider in such cases (mark them as 'success' directly in DB?)

[arollet-decathlon]

I'm experiencing exactly the same issue. Our Airflow is deployed on MWAA