Lari Hotari

#23513 is related.

> When the cancellation messages are stored in the topic, cancellation would only work when the DelayedDeliveryTracker state is such that it has "indexed" the delayed messages and the cancellation...

Can you share the thread dump that is recorded in broker logs? Does this reproduce on Pulsar 4.0.4?

did you find the thread dump recorded in the broker logs? That would be useful since it would show the dead lock. It's very long since it contains the stack...

> Not able to rerun the failed CI pipeline, but all tests passed in my repo - https://github.com/grssam/pulsar/pull/1 and the new tests that I added are part of "Broker Group...

> PS: while cherry picking to branch-3.0 , it will contain one conflict due to the command line argument parser library change at line https://github.com/apache/pulsar/pull/23120/files#diff-63a2a931cc53a14767f5eb4a53898d7b6fb20f81a87de51cc15bbc139782bd00R77 (and the relevant class imports...

This PR is ready for initial review feedback. I'm looking into ways of adding tests for the key_shared look ahead limits.

Tomorrow I'll continue and add tests and fix broken tests. I'd appreciate any reviews along the way.

There are some test failures. Trying to understand what org.apache.pulsar.client.api.KeySharedSubscriptionTest#testCheckBetweenSkippingAndRecentlyJoinedConsumers is testing. https://github.com/apache/pulsar/blob/dccc06bf50bb5ca510b39167908c02d2b4602ca5/pulsar-broker/src/test/java/org/apache/pulsar/client/api/KeySharedSubscriptionTest.java#L1235-L1333 "Test for guaranteed message ordering in corner cases" says the PR. @equanz Do you remember which corner...

Found the cause of SimpleProducerConsumerTest#testMaxUnAckMessagesLowerThanPermits failure. It's about available permits handling. The previous implementation before this PR ignored the batch index count so I skipped the logic. There's a lot...