[KAFKA-16466] QuorumController is swallowing some exception messages

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.7.0
• Fix Version/s: 3.8.0, 3.7.1
• Component/s: controller
• Labels: good-first-issue
• Assignee: Ilya Zakharov
• Reporter: David Arthur
• Reviewer: Igor Soarez
• Created: 02/Apr/24 20:20
• Updated: 23/Apr/24 01:04
• Resolved: 23/Apr/24 01:04

Description

In some cases in QuorumController, we throw exceptions from the control manager methods. Unless these are explicitly caught and handled, they will eventually bubble up to the ControllerReadEvent/ControllerWriteEvent an hit the generic error handler.

In the generic error handler of QuorumController, we examine the exception to determine if it is a fault or not. In the case where it is not a fault, we log the error like:

 log.info("{}: {}", name, failureMessage);

which results in messages like

[2024-04-02 16:08:38,078] INFO [QuorumController id=3000] registerBroker: event failed with UnsupportedVersionException in 167 microseconds. (org.apache.kafka.controller.QuorumController:544)

In this case, the exception actually has more details in its own message

Unable to register because the broker does not support version 8 of metadata.version. It wants a version between 20 and 20, inclusive.

We should include the exception's message in the log output for non-fault errors as it includes very useful debugging info.

This was found while writing an integration test for KRaft migration where the brokers and controllers have a mismatched MetadataVersion.

Issue Links

• links to: GitHub Pull Request #15701