fhan688 commented on PR #18897: URL: https://github.com/apache/hudi/pull/18897#issuecomment-4622997140
> > This makes the routing side and bucket-loading side use the same remote assignment source, instead of maintaining separate local assignment calculations. > > The motive for remote partitioner is to resolve data skew since it has the global view of assignment, does the change in this PR also addreses the similiar issue? Yes. This PR brings the same remote bucket-to-task assignment mechanism to the Flink bucket index path. The remote partitioner keeps the partition-to-start-offset mapping centrally in the timeline service, so assignment is based on the global sequence of partition bucket ranges instead of independent local hash-based calculations. To be precise, this helps with assignment-level skew across partition buckets. It does not rebalance record-volume skew inside a hot bucket. In this PR, both the Flink routing path and the bucket-loading path use the same remote assignment source, so they get the same skew-mitigation behavior and remain consistent. -- 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]
