LuciferYang opened a new issue, #13543: URL: https://github.com/apache/gluten/issues/13543
`RssPartitionWriter::doEvict`, which the hash shuffle writer uses with Celeborn or Uniffle, compresses the block in `toBlockPayload` before its `spillTime_` timer starts, and it never adds the payload's compression time to `compressTime_`. With the default `spark.shuffle.compress=true`, `totalCompressTime` is therefore always 0 for RSS hash shuffle. Uniffle's shuffle write time is meant to include the compression time, and it is missing. The "time to split" metric of both Celeborn and Uniffle is computed as wall time minus push, write and compression time, so it absorbs the compression time instead. `sortEvict` in the same file already times the whole eviction and records the compression time. This issue was written with the assistance of AI (Claude Opus). -- 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]
