[
https://issues.apache.org/jira/browse/FLINK-27051?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17696919#comment-17696919
]
Matthias Pohl commented on FLINK-27051:
---------------------------------------
This issue is about hardening the process of deleting artifacts. We have three
outcomes for cleaning up artifacts:
# The data was properly deleted.
# {{discardObject}} is called on already (partially) deleted data. -> such a
call should NOT throw an exception
# An error occurred while deleting the data. -> This should throw an exception
We have to make sure that all implementations follow this contract to allow
retries on {{{}CompletedCheckpoints{}}}.
> CompletedCheckpoint.DiscardObject.discard is not idempotent
> -----------------------------------------------------------
>
> Key: FLINK-27051
> URL: https://issues.apache.org/jira/browse/FLINK-27051
> Project: Flink
> Issue Type: Sub-task
> Components: Runtime / Coordination
> Affects Versions: 1.15.0
> Reporter: Matthias Pohl
> Priority: Major
>
> {{CompletedCheckpoint.DiscardObject.discard}} is not implemented in an
> idempotent fashion because we're losing the operatorState even in the case of
> a failure (see
> [CompletedCheckpoint:328|https://github.com/apache/flink/blob/dc419b5639f68bcb0b773763f24179dd3536d713/flink-runtime/src/main/java/org/apache/flink/runtime/checkpoint/CompletedCheckpoint.java#L328]).
> This prevents us from retrying the deletion.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
