malinjawi commented on code in PR #12592: URL: https://github.com/apache/gluten/pull/12592#discussion_r4339627384
########## docs/backends/velox/developers/MicroBenchmarks.md: ########## @@ -152,6 +153,22 @@ or 4 types of dumped file: } ``` +### Broadcast hash join + +By default the hash table of a BHJ is built once per executor and handed to the native join through +a process-local cache, so its build side never reaches the native operator as input data. While +`spark.gluten.sql.benchmark_task.stageId` or `spark.gluten.sql.benchmark_task.taskId` is set, +Gluten therefore ignores `spark.gluten.velox.buildHashTableOncePerExecutor.enabled` for every query +and builds the hash table in each task from the streamed build side, which is dumped as a regular +data file. Note that the replayed stage includes building the hash table in its measurement. + +### Empty input iterators Review Comment: Done, moved it under the Data file entry. -- 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]
