[
https://issues.apache.org/jira/browse/FLINK-36201?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17879136#comment-17879136
]
Rui Fan commented on FLINK-36201:
---------------------------------
{quote} This makes me wonder whether we should default
`execution.state-recovery.from-local` to true in 2.0. WDYT?
{quote}
Thanks [~dmvk] for pointing it out!
Actually, most of our production jobs do not enable local recovery for now. So
I'm not sure whether it's better to enable it by default.
But I think 2.0 is a good time to update the default value if some contributors
think it's useful(Pros outweigh cons). It's better to collect some feedbacks
from dev mail list if we wanna update it. :)
> StateLocalitySlotAssigner should be only used when local recovery is enabled
> for Adaptive Scheduler
> ---------------------------------------------------------------------------------------------------
>
> Key: FLINK-36201
> URL: https://issues.apache.org/jira/browse/FLINK-36201
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
> Reporter: Rui Fan
> Assignee: Rui Fan
> Priority: Major
> Labels: pull-request-available
> Fix For: 2.0.0
>
>
> SlotSharingSlotAllocator created the StateLocalitySlotAssigner[1] instead of
> DefaultSlotAssigner whenever failover happens.
> I'm curious why we use StateLocalitySlotAssigner when local recovery is
> disable.
> As I understand, the local recovery doesn't take effect if flink doesn't
> backup state on the TM local disk. So StateLocalitySlotAssigner should be
> only used when local recovery is enabled.
>
> [1]
> [https://github.com/apache/flink/blob/c869326d089705475481c2c2ea42a6efabb8c828/flink-runtime/src/main/java/org/apache/flink/runtime/scheduler/adaptive/allocator/SlotSharingSlotAllocator.java#L136]
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
