In around 40% cases, when we upgrade 3 HA replication pairs of brokers, the message-load-balancing / redistribution stops working and consumer is stuck, waiting to receive/pass messages from other broker - when messages are in expected queue on other (primary) broker.

Same scenario sometimes pass, sometimes fails.
Example build 1 passed, build 2 failed.
https://main-jenkins-csb-amq-broker-qe.apps.ocp-c1.prod.psi.redhat.com/job/7.12/job/claire/job/standalone-upgrade/job/upgrade-8-oracle-21/

Test data is quite big (6 brokers, each upgraded multiple times) (~400MB). See testData.zip file for all details.

This happens on both 7.11.x and 7.12.x versions.

Test code
https://gitlab.cee.redhat.com/amq-broker/claire/-/blob/main/standalone-suite/src/test/java/io/brokerqe/claire/upgrade/HAReplicationUpgradeTests.java?ref_type=heads#L29