FrankYang0529 opened a new pull request, #19262: URL: https://github.com/apache/kafka/pull/19262
If broker 1 doesn't get heartbeat promptly and it's fenced after the topic creation, the broker 1 cannot be ISR. The session timeout is 300ms. Following logs are from https://develocity.apache.org/s/tjs4dzxiphmwc/tests/task/:metadata:test/details/org.apache.kafka.controller.QuorumControllerTest/testMinIsrUpdateWithElr()/1/output: ``` [2025-03-18 07:14:45,121] DEBUG [QuorumController id=0] Processed processBrokerHeartbeat(1474681863) in 140186 us (org.apache.kafka.controller.QuorumController:542) <-- heartbeat for broker 1 ... [2025-03-18 07:14:45,147] DEBUG [QuorumController id=0] Processed processBrokerHeartbeat(399642596) in 10723 us (org.apache.kafka.controller.QuorumController:542) <-- heartbeat for broker 2 ... [2025-03-18 07:14:45,172] DEBUG [QuorumController id=0] Processed processBrokerHeartbeat(168471181) in 13236 us (org.apache.kafka.controller.QuorumController:542) <-- heartbeat for broker 3 ... [2025-03-18 07:14:45,288] INFO [QuorumController id=0] CreateTopics result(s): CreatableTopic(name='foo', ...) (org.apache.kafka.metalog.LocalLogManager$SharedLogData:258) <-- topic creation ... [2025-03-18 07:14:45,455] INFO [QuorumController id=0] Fencing broker 1 at epoch 6 because its session has timed out. (org.apache.kafka.controller.ReplicationControlManager:1693) <-- broker 1 session timeout ``` At 07:14:45,121, the broker 1 gets heartbeat and it's active. However, it doesn't get another heartbeat before 07:14:45,421, so it's fenced at 07:14:45,455. The case can be reproduced by adding Thread.sleep(300) just after `active.creatTopics` [0]. To solve the root cause, use another thread to send heartbeat request, so broker 1 doesn't have chance to get fenced. [0] https://github.com/apache/kafka/blob/1ded681684e771b16aa98ae751f39b9816345a83/metadata/src/test/java/org/apache/kafka/controller/QuorumControllerTest.java#L663-L665 -- 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: jira-unsubscr...@kafka.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
