richox commented on PR #1694: URL: https://github.com/apache/auron/pull/1694#issuecomment-3611814828
> > @XorSum I wonder why do we need to do this manual batch split, the join algorithm itself already have batch_size handling logic, there is bug in it? > > https://github.com/apache/auron/blob/10517904367ae91cbe88ca8635bffd010601f201/native-engine/datafusion-ext-plans/src/joins/smj/full_join.rs#L55-L57 > > Thanks for the comment! I do not yet understand how the join batch_size handling logic works, however I did find a reproducible bug(#1693). An extremely large batch (100 million rows) is generated after the join operation, and lead to the `OversizedAllocationException`. In some extreme cases which have a lot of duplicated keys in both join sides, the `should_flush()` function might not be checked in time, huge batches will be created in thess cases. however, i dont think this patch can solve the issue because the huge batch has already been generated and cause oom before it is splitted. -- 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]
