[FLINK-16443] Fix wrong fix for user-code CheckpointExceptions

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.11.4, 1.12.2, 1.13.0
• Component/s: Runtime / Checkpointing
• Labels: pull-request-available
• Assignee: Arvid Heise
• Reporter: Stephan Ewen
• Created: 05/Mar/20 13:50
• Updated: 22/Jun/21 14:05
• Resolved: 03/Dec/20 08:02

Description

The problem of having exceptions that are only in the user code classloader was fixed by proactively serializing them inside the CheckpointException. That means all consumers of CheckpointException now need to be aware of that and unwrap the serializable exception.

I believe the right way to fix this would have been to use a SerializedException in the DeclineCheckpoint message instead, which would have localized the change to the actual problem: RPC transport.

I would suggest to revert https://github.com/apache/flink/pull/9742 and instead apply the above described change.

Issue Links

• links to: GitHub Pull Request #14177