jerqi commented on issue #263: URL: https://github.com/apache/incubator-uniffle/issues/263#issuecomment-1286686537
> > OK, I think it's more convenient that we shutdown the shuffle servers for a period of time. All the applications use ess , we upgrade all the shuffle servers, we let the applications use the rss again. > > Emm... I practiced by your way and it also have to wait all apps finished which will cost 1 Days(due to some ML jobs). > > Besides In our co-location environment(Uniffle ShuffleServer on NodeManager), this is acceptable and I introduce the safemode to reject all newer assignment from newly app in our internal uniffle version. > > But if the spark job is running on the k8s pod of less local disk or no ESS, how to upgrade? You're right. StateStore is useful. But if you does some poc, it have great effective, or you use this feature in your environment, it's acceptable. This feature is a big feature, we should be careful. It might be more suitable to merge this feature when Uniffle are more stable. -- 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]
