Hi devs, I opened PR #25210 to fix issue #25204 (epoch can change mid-batch, causing mixed accept and filter). The fix holds incomingQueueLock across the batch loop plus the incomingMessages.size() read. The regression test was updated to avoid deadlock and assert the post-fix invariant.
PR: https://github.com/apache/pulsar/pull/25210 <https://github.com/apache/pulsar/pull/25210?utm_source=chatgpt.com> Issue: https://github.com/apache/pulsar/issues/25204 <https://github.com/apache/pulsar/issues/25204?utm_source=chatgpt.com> Could someone review the lock scope and any throughput or ordering concerns? Yours sincerely, Chimdumebi Nebolisa
