[FLINK-6227] Introduce the abstract PartitionException for downstream task failure

• Type: Sub-task
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 1.9.0
• Component/s: Runtime / Coordination, Runtime / Network
• Labels: pull-request-available
• Assignee: Zhijiang
• Reporter: Zhijiang
• Created: 31/Mar/17 08:56
• Updated: 03/Jul/20 09:23
• Resolved: 23/May/19 16:19

Description

It is part of FLIP-1.

 

We define a new abstract special exception called PartitionException to indicate the downstream task failure in consuming upstream partition. This special exception is reported to JobMaster as a hint for unavailable partition data, and then JobMaster could track the upstream executions to restart for reproducing the partition data.

We make current PartitionNotFoundException extend PartitionException, because it is reasonable to restart producer task when the requested partition is already released.

Issue Links

• is duplicated by: FLINK-10573 Support task revocation
• is related to: FLINK-12458 Introduce PartitionConnectionException for unreachable producer
• is related to: FLINK-10289 Classify Exceptions to different category for apply different failover strategy
• relates to: FLINK-12068 Backtrack failover regions if intermediate results are unavailable
• links to: GitHub Pull Request #8242