[
https://issues.apache.org/jira/browse/FLINK-22256?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17342689#comment-17342689
]
Till Rohrmann edited comment on FLINK-22256 at 5/11/21, 3:56 PM:
-----------------------------------------------------------------
We want to be able to say what properties a savepoint/checkpoint has when
looking at the savepoint/checkpoint from the user's perspective. One idea could
be to look at the metadata file. The interesting properties at the moment are
whether it is a full checkpoint or an incremental checkpoint. Moreover, in case
it is an incremental checkpoint, it would be great to be able to delete the
checkpoint data w/o having to start Flink. At the moment an incremental
checkpoint consists of the shared parts and the checkpoint specific parts.
In the end it is the information needed to manage the lifecycle of savepoints
and retained checkpoints by a third party tool.
was (Author: till.rohrmann):
We want to be able to say what properties a savepoint/checkpoint has when
looking at the savepoint/checkpoint from the user's perspective. One idea could
be to look at the metadata file. The interesting properties at the moment are
whether it is a full checkpoint or an incremental checkpoint. Moreover, in case
it is an incremental checkpoint, it would be great to be able to delete the
checkpoint data w/o having to start Flink. At the moment an incremental
checkpoint consists of the shared parts and the checkpoint specific parts.
> Persist checkpoint type information
> -----------------------------------
>
> Key: FLINK-22256
> URL: https://issues.apache.org/jira/browse/FLINK-22256
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Checkpointing
> Reporter: Fabian Paul
> Priority: Major
>
> As a user, it is retrospectively difficult to determine what kind of
> checkpoint (i.e. incremental, unaligned, ...) was performed when looking only
> at the persisted checkpoint metadata.
> The only way would be to look into the execution configuration of the job
> which might not be available anymore and can be scattered across the
> application code and cluster configuration.
> It would be highly beneficial if such information would be part of the
> persisted metadata to not track these external pointers.
> It would also be great to persist the metadata information in a standardized
> format so that external projects don't need to use Flink's metadata
> serializers to access it.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
