[FLINK-21178] Task failure will not trigger master hook's reset()

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.11.3, 1.12.0
• Fix Version/s: 1.13.0, 1.12.3
• Component/s: Runtime / Checkpointing
• Labels: pull-request-available
• Assignee: Brian Zhou
• Reporter: Brian Zhou
• Created: 28/Jan/21 06:34
• Updated: 11/Mar/21 09:31
• Resolved: 11/Mar/21 09:31

Description

In Pravega Flink connector integration with Flink 1.12, we found an issue with our no-checkpoint recovery test case [1].

We expect the recovery will call the ReaderCheckpointHook::reset() function which was the behaviour before 1.12. However FLINK-20222 changes the logic, the reset() call will only be called along with a global recovery. This causes Pravega source data loss when failure happens before the first checkpoint.

[1]  https://github.com/crazyzhou/flink-connectors/blob/da9f76d04404071471ebd86bf6889b307c9122ff/src/test/java/io/pravega/connectors/flink/FlinkPravegaReaderRGStateITCase.java#L78

Issue Links

• links to: GitHub Pull Request #14890
• links to: GitHub Pull Request #15067