andygrove commented on PR #5993: URL: https://github.com/apache/datafusion-comet/pull/5993#issuecomment-5709022631
Closing this. The approach was validated end to end on k8s and the plumbing works: cgroup v2 discovery inside a pod, correct limit, trip propagating as `CometNativeException`. But `memory.current` turns out to be the wrong signal. It counts reclaimable page cache, so it saturates at the container limit on any workload that reads files. In an A/B on TPC-H SF100 Q9 at a 9Gi pod limit, the guard failed the job (24 trips) while the identical run with the guard off completed in 58.62s, and nothing was OOMKilled in either. Measured `anon` peaked at 55% of the limit while `memory.current` sat at 99.99%. Details and the cgroup trace are in #4576. A future attempt should threshold on `anon`/PSI rather than iterate on this branch. -- 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]
