[
https://issues.apache.org/jira/browse/FLINK-29459?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17621187#comment-17621187
]
Krzysztof Chmielewski commented on FLINK-29459:
-----------------------------------------------
FYI ticets
29509
29512
29627
are fixing issue with Task manager recovery for Sink architecture with global
committer.
The 29583 is about recovering Flink 1.14 unified sinks committer state and
migrate it to the extended unified model.
> Sink v2 has bugs in supporting legacy v1 implementations with global committer
> ------------------------------------------------------------------------------
>
> Key: FLINK-29459
> URL: https://issues.apache.org/jira/browse/FLINK-29459
> Project: Flink
> Issue Type: Bug
> Components: API / DataStream
> Affects Versions: 1.16.0, 1.17.0, 1.15.2
> Reporter: Yun Gao
> Assignee: Yun Gao
> Priority: Major
> Fix For: 1.17.0, 1.15.3, 1.16.1
>
>
> Currently when supporting Sink implementation using version 1 interface,
> there are issues after restoring from a checkpoint after failover:
> # In global committer operator, when restoring SubtaskCommittableManager,
> the subtask id is replaced with the one in the current operator. This means
> that the id originally is the id of the sender task (0 ~ N - 1), but after
> restoring it has to be 0. This would cause Duplication Key exception during
> restoring.
> # For Committer operator, the subtaskId of CheckpointCommittableManagerImpl
> is always restored to 0 after failover for all the subtasks. This makes the
> summary sent to the Global Committer is attached with wrong subtask id.
> # For Committer operator, the checkpoint id of SubtaskCommittableManager is
> always restored to 1 after failover, this make the following committable sent
> to the global committer is attached with wrong checkpoint id.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
