[
https://issues.apache.org/jira/browse/FLINK-12672?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Andrey Zagrebin updated FLINK-12672:
------------------------------------
Description:
At the moment we use consumer running check in RemoteChannelStateChecker as a
shortcut if consumer is done before the check has been accomplished then we
skip the check.
Ideally we should use closeFuture in SingleInputGate instead of consumer state
for this. Although this shortcut is the best effort because the state change or
gate close can happen concurrently right after this check,
exposing consumer state to shuffle implementation looks unnecessary.
See also PR comment:
[https://github.com/apache/flink/pull/8463#discussion_r288537012]
was:
At the moment we use consumer running check in as a shortcut if consumer is
done before the check has been accomplished then we skip the check.
Ideally we should use closeFuture in SingleInputGate instead of
ConsumerExecutionState for this. Although this shortcut is the best effort
because the state change or gate close can happen concurrently right after this
check
See also PR comment:
[https://github.com/apache/flink/pull/8463#discussion_r288537012]
> Use SingleInputGate.closeFuture in triggerPartitionStateCheck instead of
> consumer state in RemoteChannelStateChecker
> --------------------------------------------------------------------------------------------------------------------
>
> Key: FLINK-12672
> URL: https://issues.apache.org/jira/browse/FLINK-12672
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Network
> Reporter: Andrey Zagrebin
> Priority: Minor
>
> At the moment we use consumer running check in RemoteChannelStateChecker as a
> shortcut if consumer is done before the check has been accomplished then we
> skip the check.
> Ideally we should use closeFuture in SingleInputGate instead of consumer
> state for this. Although this shortcut is the best effort because the state
> change or gate close can happen concurrently right after this check,
> exposing consumer state to shuffle implementation looks unnecessary.
> See also PR comment:
> [https://github.com/apache/flink/pull/8463#discussion_r288537012]
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
