tgravescs commented on issue #25078: [SPARK-28305][YARN] Request GetExecutorLossReason to use a smaller timeout parameter URL: https://github.com/apache/spark/pull/25078#issuecomment-510505529 right, so to clarify I think you are say yes if you set askTimeout separate it also fixes the problem, correct? So my concern is that you are fixing this in a single location, I think there are other places that could have the same issue, although the question is whether those are recoverable or not. In this case it is recoverable because its just trying to get the loss reason and if it doesn't it can go on just fine and the AM will restart and connect back to the driver. My other concern here is that we are breaking what the configs actually do. This is no longer just the asktimeout or network timeout, its 0.95 times one of those. Which is inconsistent with what is documented and everywhere else. Let say I set ask timeout different from network timeout, now you are decreasing the ask timeout even more then what I set it to. so I'm wondering if you can just set the ask timeout globally different or if that causes other issues?
---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
