Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

Root cause: AsyncWindowJoinRecordCursor.close() was freeing slaveTimeFrameAddressCache before waiting for worker threads to finish.

That's not the only scenario when this bug would show up and the added test demonstrates that. While the AsyncWindowJoinRecordCursor bugfix is good there is an inefficiency in frameSequence.await() calls that are made by all async record cursors. The problem is that when closing a cursor we should skip all tasks that belong to the frame sequence under close instead of reducing them. Instead, we seem to reduce the published tasks which is a waste of CPU time and only increases query latency.

That's why the segfault happened reliably when LIMIT n clause was added - query owner thread (if not one of the worker threads) was guaranteed to reduce the remaining tasks.

I'll create a GH issue for this performance enhancement.

I'll create a GH issue for this performance enhancement.

GH issue: #6665

[PR Coverage check]

😍 pass : 84 / 84 (100.00%)

file detail