[
https://issues.apache.org/jira/browse/FLINK-12413?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16840949#comment-16840949
]
Zhu Zhu commented on FLINK-12413:
---------------------------------
I think the restart attempt number might not be enough for
RestartBackoffTimeStrategy(other than FixedDelayRestartBackoffTimeStrategy) to
make decisions, e.g. FailureRateRestartBackoffTimeStrategy.
It may be hard if we want to keep RestartBackoffTimeStrategy stateless,
especially considering future possible RestartBackoffTimeStrategy
implementations(in my mind there can be
PerTaskFixedDelayRestartBackoffTimeStrategy and
PerTaskFailureRateRestartBackoffTimeStrategy).
To properly maintain the RestartBackoffTimeStrategy state, I think we need to
add a notifyFailure(Throwable) interface to RestartBackoffTimeStrategy, or
change getBackoffTime() to be notifyFailureAndGetBackoffTime(Throwable).
Would that work?
> Implement ExecutionFailureHandler
> ---------------------------------
>
> Key: FLINK-12413
> URL: https://issues.apache.org/jira/browse/FLINK-12413
> Project: Flink
> Issue Type: Sub-task
> Components: Runtime / Coordination
> Reporter: Gary Yao
> Assignee: Zhu Zhu
> Priority: Major
>
> Implement ExecutionFailureHandler and related classes.
> *Acceptance criteria*
> * Implementation or definitions exist for:
> ** {{ExecutionFailureHandler}}
> ** {{FailureHandlingResult}}
> ** {{RestartBackoffTimeStrategy}} (interface definition is enough)
> * {{ExecutionFailureHandler}} is unit tested in isolation
> * {{FailureHandlingResult}} is unit tested in isolation
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
