[
https://issues.apache.org/jira/browse/FLINK-33355?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17779315#comment-17779315
]
Yun Tang commented on FLINK-33355:
----------------------------------
Changing the max-parallelism (instead of parallelism), would break the
checkpoint compatibility, which is built by design. You can refer to
https://nightlies.apache.org/flink/flink-docs-master/docs/dev/datastream/execution/parallel/#setting-the-maximum-parallelism
for more details.
> can't reduce the parallelism from 'n' to '1' when recovering through a
> savepoint.
> ---------------------------------------------------------------------------------
>
> Key: FLINK-33355
> URL: https://issues.apache.org/jira/browse/FLINK-33355
> Project: Flink
> Issue Type: Bug
> Components: API / Core
> Environment: flink 1.17.1
> Reporter: zhang
> Priority: Major
>
> If the program includes operators with window, it is not possible to reduce
> the parallelism of the operators from n to 1 when restarting from a
> savepoint, and it will result in an error:
> {code:java}
> //IllegalStateException: Failed to rollback to checkpoint/savepoint
> Checkpoint Metadata. Max parallelism mismatch between checkpoint/savepoint
> state and new program. Cannot map operator 0e059b9f403cf6f35592ab773c9408d4
> with max parallelism 128 to new program with max parallelism 1. This
> indicates that the program has been changed in a non-compatible way after the
> checkpoint/savepoint. {code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
