[
https://issues.apache.org/jira/browse/FLINK-14271?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16940723#comment-16940723
]
Zhu Zhu commented on FLINK-14271:
---------------------------------
Hi [~stevenz3wu], I hear from [~GJL] that you are using the legacy
{{RestartPipelinedRegionStrategy}}, namely configuring
"jobmanager.execution.failover-strategy" to "region-legacy" in Flink 1.9 or
"region" in previous versions.
The legacy {{RestartPipelinedRegionStrategy}} has a limitation that it's not
able to backtrack missing input result partitions.
As a result it is superseded by the new strategy
{{AdaptedRestartPipelinedRegionStrategyNG}} in Flink 1.9. The new strategy's
configuration value is "region" in Flink 1.9.
Did you test whether the new region failover strategy works for your jobs which
use the legacy {{RestartPipelinedRegionStrategy}}?
If it works, will you turn to the new strategy?
Your feedback is critical for our decision making.
If you do not depend on it anymore, we can remove the legacy
{{RestartPipelinedRegionStrategy}} in Flink 1.10.
> Deprecate legacy RestartPipelinedRegionStrategy
> -----------------------------------------------
>
> Key: FLINK-14271
> URL: https://issues.apache.org/jira/browse/FLINK-14271
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Coordination
> Affects Versions: 1.10.0
> Reporter: Zhu Zhu
> Priority: Minor
> Fix For: 1.10.0
>
>
> The legacy {{RestartPipelinedRegionStrategy}} has been superseded by
> {{AdaptedRestartPipelinedRegionStrategyNG}} in Flink 1.9.
> It heavily depends on ExecutionGraph components and becomes a blocker for a
> clean scheduler re-architecture.
> We should deprecate it for further removal.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
