mridulm commented on PR #2373: URL: https://github.com/apache/celeborn/pull/2373#issuecomment-2097428792
>> OR we can treat skewed shuffleIDs as INDETERMINATE stages and let Spark handle the stages retries and abortion accordingly. > I think this is the key idea, if this is doable, I would prefer this approach : ) This gets tricky, and depends on when we 'apply' the diff. It has to be detected and applied before `DAGScheculer` uses it, when the stage fails with fetch failure. Not that it cant be done - but it is invasive, and unfortunately unstable (this can arbitrarily evolve within scheduler) -- 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]
