[ 
https://issues.apache.org/jira/browse/FLINK-25872?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17534913#comment-17534913
 ] 

Yuan Mei commented on FLINK-25872:
----------------------------------

I think [~dwysakowicz] and [~pnowojski] make a good point.

If we take changelog as a different state backend, it makes sense to follow the 
current way of switching state backend through savepoints.

However, I think changelog is a bit different from the regular backend, I'd 
instead take it as a different checkpoint type, similar to an unaligned 
checkpoint, conceptually, but implemented in the state backend wrapper. That 
says it provides the same format as the changelog checkpoint (changelog + 
materialization) vs the unaligned checkpoint (channel data + snapshot).

The other part is that you do not expect to switch between hashmap <-> rocksDB 
that often (most of the time, stick to just one). But on the other hand, enable 
<=> disable changelog is reasonable to be more lightweight based on the 
tradeoff between cp speed vs additional space/network cost.

> Restoring from non-changelog checkpoint with changelog state-backend enabled 
> in CLAIM mode discards state in use
> ----------------------------------------------------------------------------------------------------------------
>
>                 Key: FLINK-25872
>                 URL: https://issues.apache.org/jira/browse/FLINK-25872
>             Project: Flink
>          Issue Type: Bug
>          Components: Runtime / Checkpointing, Runtime / State Backends
>            Reporter: Yun Tang
>            Assignee: Yanfei Lei
>            Priority: Major
>              Labels: pull-request-available
>             Fix For: 1.16.0
>
>
> If we restore from checkpoint with changelog state-backend enabled in 
> snapshot CLAIM mode, the restored checkpoint would be discarded on subsume. 
> This invalidates newer/active checkpoints because their materialized part is 
> discarded (for incremental wrapped checkpoints, their private state is 
> discarded). This bug is like FLINK-25478.
>  
> Design doc: 
> [https://docs.google.com/document/d/1KSFWc0gL7HkhC-JNrnsp06TLnsTmZOTHITQDcGMo0cI/edit?usp=sharing,]



--
This message was sent by Atlassian Jira
(v8.20.7#820007)

Reply via email to