[
https://issues.apache.org/jira/browse/FLINK-22234?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17337149#comment-17337149
]
Nicholas Jiang edited comment on FLINK-22234 at 4/30/21, 7:41 AM:
------------------------------------------------------------------
[~trohrmann], [~chesnay], the behavior that read savepoint before creating
ExecutionGraph could be that create the CheckpointStorage, create the
CheckpointStorageCoordinatorView with the create storage, reading potential
savepoints, checkpoints, extract their meta information and then create an
ExecutionGraph. After the creation of ExecutionGraph, create the
CheckpointCoordinator with the created CheckpointStorageCoordinatorVIew.
Because given with the CheckpointStorage and JobID, the coordinator view could
be generated, which could pass to CheckpointCoordinator, reduce the time
reading the savepoint.
[~trohrmann], [~chesnay], what do you think about?
was (Author: nicholasjiang):
[~trohrmann], [~chesnay], the behavior that read savepoint before creating
ExecutionGraph could be that create the CheckpointStorage, create the
CheckpointStorageCoordinatorView with the create storage, reading potential
savepoints, checkpoints, extract their meta information and then create an
ExecutionGraph. After the creation of ExecutionGraph, create the
CheckpointCoordinator with the created CheckpointStorageCoordinatorVIew.
[~trohrmann], [~chesnay], what do you think about?
> Read savepoint before creating ExecutionGraph
> ---------------------------------------------
>
> Key: FLINK-22234
> URL: https://issues.apache.org/jira/browse/FLINK-22234
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Coordination
> Reporter: Chesnay Schepler
> Priority: Major
> Labels: pull-request-available
> Fix For: 1.14.0
>
>
> Savepoints are currently read when the ExecutionGraph is created.
> This is causing problems when determining the (max) parallelism the job is
> run with, because the savepoint may impose additional restrictions.
> This caused quite a few headaches in FLINK-21844, and further requires the
> execution vertices to have a mutable max parallelism.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
