stale containers exclusion and handling improvement

[sangreal]

Description... in previous logic we need two passes to get the normal containers and put stale containers into mailbox queue. But now we only need one pass, therefore when the traffic is huge, there will be performance gain.

Checklist

• [ ] Documentation (if applicable)
• [x] Changelog

[rkolesnev]

Hey @sangreal - is the added complexity worth the improvement? My understanding that performance improvement will only apply when there is stale work due to rebalance / epoch change - but such an event should really be rare relatively speaking - so few millisecond performance improvement on one-off scans wouldn't warrant additional changes in my opinion. What do you think?

[sangreal]

@rkolesnev Thanks for the reply.

1. I think the probability is not that low since when I encountered the staled containers causing poller paused last year, the assigned to the same partition causing stale containers happened a time a few days using cooperativestickyassignor. Therefore the fix could be valuable.
2. the new logic is quite straightforward and with UTs, I don't think this is quite complex.

Happy to hear your opinion on these.

[rkolesnev]

/sem-approve

[rkolesnev]

Overall logic looks solid - i've suggested to move the filtering earlier in the flow to eliminate re-scanning and make it a bit more clear upfront.

[rkolesnev]

And if you can please run license:format or update license headers in the changed test files to 2024

[sangreal]

And if you can please run license:format or update license headers in the changed test files to 2024

Sure, headers have been updated

[rkolesnev]

/sem-approve