Taragolis commented on issue #34339:
URL: https://github.com/apache/airflow/issues/34339#issuecomment-1719126077

   > There are a couple of other reasons why we build our own, but that may 
change in the future.
   
   If not yet, you should consider to use any process daemon/supervisor behind 
your shell entrypoint. As I mention before it could be tiny or dumb-init both 
of them presented in debian repo. `dumb-init` uses in official Airflow image, 
`tiny` uses in official Prefect image, as user I can't find any differences 
between this two init managers both of them do great job - no more zombie 
process.
   
   > highly dynamic graphs of small lightweight tasks.
   
   It also could be a reason of fantom failures, if DAG graph could be change 
during execution, e.g.: task deleted, dependency between tasks changed and etc, 
then a lot of funny side effects could happen.
   
   > Processing time is really variable. Sometimes it's 4 seconds, sometimes 
it's 25, sometimes it's 50+. And before you ask, no the box CPU is pretty 
constant. However, I'm doing my best to eliminate exogenous factors before I 
start claiming the problem is 100% in airflow.
   
   - DAG generated thought the DB?
   - DAG placed on some network share, e.g. NFS?
   
   ---
   
   The more I look at this issue the more filling that is not an issue rather 
then discussion and troubleshooting 🤔 


-- 
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