Dandandan commented on PR #21240: URL: https://github.com/apache/datafusion/pull/21240#issuecomment-4175309833
> > In principle self.right.execute only builds the stream - it shouldn't do any "actual" work, only the setup > > Is that true in practice? e.g., > > * CoalescePartitionsExec::execute() -> builder.run_input() -> self.inner.spawn(...) > * SortPreservingMergeExec::execute() -> spawn_buffered() > > Both plan nodes are used by TPC-DS Q24, as one example of a place where we saw a slowdown w/o adding additional overlapping in ScalarSubqueryExec. You are probably right in those cases it is doing potentially many things in parallel. But I think this is not what we want ideally - we want to run few independent pipelines as possible, and get (data) parallelism from the individual pipelines rather than executing all at the same time. -- 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]
