yl09099 commented on PR #1129: URL: https://github.com/apache/incubator-uniffle/pull/1129#issuecomment-1674363963
> @yl09099 thanks for your work. This PR is large enough to split into multiple parts: > > 1. the proto and the corresponding rpc client and service part. > 2. the coordinator part > 3. Spark part, you may divided it into reader and writers parts. > 4. The unified client part that shares shuffle assignment info across tasks, current approach would retrieve shuffle assignment per task, that may put a lot pressure to the driver. > > Before dividing it into multiple parts, would you mind to write a design doc for this pr. If not, a control & data flow diagram about how dynamic shuffle server assignment are generated, transferred, when triggered would be needed. So we can discus a bit more and determines current approach is the good design choice or not. > > I did a quick overview of this PR, I didn't notice extra data check for this new procedure. Maybe I'm a bit of overlooking, but like I noted in the #825, we may need to revise the data check to protect data integrity. ok,I'll break it up as you want. -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
