[
https://issues.apache.org/jira/browse/FLINK-21084?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17271250#comment-17271250
]
Guowei Ma commented on FLINK-21084:
-----------------------------------
Currently a Task would decline the checkpoint if some operators in it are
finished. Some operators (for example sink) depend on the
`Checkpoint/Checkpoint Listener` even if its’ predecessor operators are closed.
However, the first phase of `Support Checkpoints After Tasks Finished` plans
not to support this. After some offline discussion with [~gaoyunhaii] we would
propose not to support this at first version.
> Allows tasks to report operator subtask finish state
> ----------------------------------------------------
>
> Key: FLINK-21084
> URL: https://issues.apache.org/jira/browse/FLINK-21084
> Project: Flink
> Issue Type: Sub-task
> Components: API / DataStream, Runtime / Checkpointing
> Reporter: Yun Gao
> Priority: Major
>
> Once an operator is fully finished (e.g., all its subtasks are finished) in
> one checkpoint, the state need to be recorded to skip the execution of this
> operator after recovery. Thus we should allow operators to report finished
> state so that CheckpointCoordinator could identify fully finished operators
> for each checkpoint.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
