potiuk commented on issue #39717:
URL: https://github.com/apache/airflow/issues/39717#issuecomment-2217444706

   > I'm not sure I agree with the "configuration specific" targeting 
problem/issue/ whatever.
   
   Yes, And as I mentioned before with "survivorship bias" you think that it 
affects everyone but it's not the case (otherwise we would be flooded with 
people stating it and it would show up in our tests and some of the service 
companies - Astronomer for one who runs latest version would also experienced 
it.
   
   So the best what you can do is to provide some guesses, results of digging 
that will narrow down the problem - ideally possibly maybe bisecting things and 
disabling it - for example if someone (one of the teams here) experiences this 
problem on a staging system, turning on and of and modifying selected 
parameters one-by-one to see what can fix it is is a great engineering 
practice. I use it daily to narrow down and find out what the problem is by 
trying out and guessing - for Airflow CI.
   
   And you can only, and exclusively use that techique when you have somewhat 
reproducible issue. Which narrows down the list of poeple who can actually do 
such check to those who experience it.  If you don't experience the problem, 
there is very little chance that you can guess what configuration or DAG might 
cause it. Especially when you have no access to the DAG and the environment 
where Airflow is run. On the other hand, when you DO experience it and DO have 
DAG access and CAN change and play with the configuration (for example on 
staging system) and DO have somewhat reproducible case, then yes - you can do 
bisecting and testing results of individual configuration changes. 
   
   And this is all we are asking here for: if any of people experiencing it and 
having somewhat reproducible case and be able to change various configuration 
parameters (by looking at the discussion above you will find a few of 
hypotheses) and creatively dig and find correlation and maybe try even 
different hypothesis - this might ACTUALLY help with diagnosis.
   
   Yes. It will require time and dedication and focused effort from someone. 
And this the smallest thing the community can ask our users to contribute back 
as a sign of gratitude and `thank you` for the free software the community 
releases for free, I think that's generally a small price to pay for all the 
cost saved (tens of thousands of dollars a month sometimes) for similar 
commercial solutions.
   
   


-- 
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