potiuk commented on issue #27589:
URL: https://github.com/apache/airflow/issues/27589#issuecomment-1316114343

   Another option for pin-pointing is to selectively disable certain processes 
and compare the usage before/after. For example if you see a pod running with 
multiple processes in it 0 you can delete certain processes in some containers 
- changing  an entrypoint command to run with "sleep 3600" will run - likely 
everything else that there is to run with something that for sure does not take 
memory - and you can see which process caused it. 
   
   On top of that - again switching airfflow back to originl configuration and 
"vanilla" state might tell you for example that your configuration is your 
problem and applying configuraiotn (including logging handles, setting default 
values for host name check and many others might help with pin-pointing. It's 
almost certain Airlfow in the vanilla state has no leak. With the It would be 
far too easy to see - so it must be something on your side. The growth you 
should is pretty catastrophic and it would demand most of airflow installation 
to restart scheduler every day or so - which does not happen.
   
   I also suggest (if you get there to vanilla and the memory is stil growing) 
to test different airflow versions - maybe what you see is a mistake- and 
trying various versions might simply give more answers. And finally if you see 
it in several airflow versions, I would try to run other experiments - 
replacing scheduler with other components etc. etc. Unfortnately I cannot have 
access to your system to play with it but if I were you, this is what I'd do.
   
   


-- 
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]

Reply via email to