You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@flink.apache.org by "Han Yin (Jira)" <ji...@apache.org> on 2025/02/17 08:55:00 UTC

[jira] [Updated] (FLINK-37335) ForSt fails to generate the first checkpoint in IncrementalCP and NO_CLAIM mode.

     [ https://issues.apache.org/jira/browse/FLINK-37335?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Han Yin updated FLINK-37335:
----------------------------
    Fix Version/s: 2.1.0

> ForSt fails to generate the first checkpoint in IncrementalCP and NO_CLAIM mode.
> --------------------------------------------------------------------------------
>
>                 Key: FLINK-37335
>                 URL: https://issues.apache.org/jira/browse/FLINK-37335
>             Project: Flink
>          Issue Type: Bug
>          Components: Runtime / Checkpointing, Runtime / State Backends
>    Affects Versions: 2.1.0
>            Reporter: Han Yin
>            Priority: Major
>             Fix For: 2.1.0
>
>
> The {{ForStIncrementalSnapshotStrategy}} throws an exception if it encounters a {{sharingFilesStrategy}} of {{FORWARD}} when generating asynchronous snapshots. This behavior is designed to fail early in unexpected situations where full checkpoints are required for the incremental snapshot strategy.
> However, this incorrectly handles the following scenario: When incremental checkpoints are enabled and the job is restored in {{NO_CLAIM}} mode, the first checkpoint should be a full checkpoint, meaning the {{sharingFilesStrategy}} would be {{{}FORWARD{}}}.
> To address this issue, we let {{ForStIncrementalSnapshotStrategy}} skip throwing an exception if it encounters {{FORWARD}} but there are no previous snapshots to reuse.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)