llamageddon83 commented on issue #32928:
URL: https://github.com/apache/airflow/issues/32928#issuecomment-1698049513

   Hi @potiuk 
   Please find my answers in line below:
   
   Is it possible that you do check which tasks are in "running" state for some 
time when it happens and maybe try to find some correlated logs from scheduler 
and task executtion - and ideally also find the logs from another task instance 
of the same task that was not found in "running" so that we could compare them 
and see the difference?
   
   > Yes we have spent quite a bit of time trying to compare logs from 
different tasks, logs of same tasks from different executions. But there is no 
oddity we could spot. As far as `running` state goes, since this happens very 
sporadically, we don't know what happens to the task state. Whenever we take a 
look, the tasks have completed and are no longer in the `running` state.
   
   I preseume that there is some kind of race happenin that will somehow skip 
removal of the task even if it has been completed. - but seeing the logs and 
comparing them could narrow down the searrch an might allow us to come up with 
a plausible hypothesis.
   
   > Sorry I wish I could help. We have spent weeks trying to narrow it down to 
some DAG implementation, some kind of pattern we may have.. but we got nothing..
   
   Question - do you run deferrable tasks / triggerer ? Maybe somehow there is 
a problem with deferrable code ? (wild guessing now after looking into some 
possible paths).
   
   > Yes, we have quite a few of them. The deferrable tasks run in a never 
ending loop. I know you are very busy and go through many discussion, but if 
you remember, I previously mentioned how we use triggers for streaming 
[here](https://github.com/apache/airflow/discussions/31074). The triggers 
schedule is stacked and they ramp up to their max_active_runs. So they will 
start one by one with a few minutes interval in between. 
   
   


-- 
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: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Reply via email to