[
https://issues.apache.org/jira/browse/FLINK-33498?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17786162#comment-17786162
]
Zakelly Lan commented on FLINK-33498:
-------------------------------------
I'm afraid the exception should be thrown around the KeyedStateBackend, which
selects the scope of checkpoint stream (exclusive or shared or both) to write
during snapshot. Only KeyedStateBackend knows whether it could reuse the files
from previous full checkpoints and do an incremental one.
And later we could consider supporting this in RocksDBStateBackend, by fast
duplicating / hard linking remote state files from exclusive to shared. WDYT?
[~pnowojski]
> Do not allow manually triggering incremental checkpoints with full checkpoint
> configured
> ----------------------------------------------------------------------------------------
>
> Key: FLINK-33498
> URL: https://issues.apache.org/jira/browse/FLINK-33498
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Checkpointing
> Reporter: Zakelly Lan
> Assignee: Zakelly Lan
> Priority: Major
> Fix For: 1.19.0
>
>
> Currently, when a job is configured to run with incremental checkpoint
> disabled, user manually triggers an incremental checkpoint actually
> triggering a full checkpoint. That is because the files from full checkpoint
> cannot be shared with an incremental checkpoint. So we'd better throw some
> exception somewhere around {{CheckpointCoordinator}} and fail the request in
> this case.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
