squito commented on issue #23438: [SPARK-26525]Fast release ShuffleBlockFetcherIterator on completion of the iteration URL: https://github.com/apache/spark/pull/23438#issuecomment-457728641 want to make sure I understand what is going on here -- the point is *not* to be calling `ShuffleBlockFetcherIterator.cleanup()` earlier, as soon as all the data is read. The point is just to make sure there aren't any more lingering references to `ShuffleBlockFetcherIterator` from the taskCompletionListener? It looks to me that calling `cleanup()` earlier won't help much with memory when the iterator has been successfully read from to completion, as then all the buffers would have already been released. But I see that you do need to make sure it happens, to delete `shuffleFileSet`, so you can remove the taskCompletionListener. is that correct?
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
