rtpsw commented on PR #34392: URL: https://github.com/apache/arrow/pull/34392#issuecomment-1501205568
Coming back to this, I see [the same race condition](https://github.com/apache/arrow/pull/34392#issuecomment-1449560506) after merging main. One experiment I made is related to [this comment](https://github.com/apache/arrow/blob/311cc52d208034214c325983f416f2f847a2a015/cpp/src/arrow/acero/asof_join_node.cc#L186-L189): - When I add a lock over `mutex_` to `UnsyncFront`, I get the same results, including the race condition. This means that `UnsyncFront` is called safely per the comment, i.e., not called concurrently with `pop` or `try_pop`. - When I add a lock over `mutex_` to `UnsyncSize`, I see a deadlock between one thread calling `AsofJoinNode::ProcessThread` -> `ConcurrentQueue::Pop and another calling `InputState::Push` -> `BackpressureConcurrentQueue::Push` -> `BackpressureConcurrentQueue::DoHandle` -> `ConcurrentQueue::UnsyncSize`. This isn't surprising, just shows that the comment does not apply to `UnsyncSize`. -- 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]
