zanmato1984 commented on PR #41335: URL: https://github.com/apache/arrow/pull/41335#issuecomment-2088118988
> Sorry for the confusion! I think per-node thread local temp stacks are what was eventually intended but that refactor was not completed. Sorry I'm still confused. From the discussion in the email thread, it seems like the intention at that time was making the temp stack thread-local and inter-node, i.e., sharing the same temp stack instance among all the nodes being executed in a specific thread (sometimes referred to as a "pipeline"). The "unfinished refactoring" back then might be making more kinds of nodes to use the shared temp stack rather than their own. I would +1 to the "per-node temp stack" solution as I think it achieves the original goal of reducing the heap allocation overhead, as well as the possibility of overflow. Comparing to "shared stack", it only introduces moderate overhead of the amount of memory that is allocated for each node's stack. I'm asking the above question just to sort out the whole story. Thanks. -- 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]
