[
https://issues.apache.org/jira/browse/FLINK-12039?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16900982#comment-16900982
]
TisonKun commented on FLINK-12039:
----------------------------------
When investigating this issue, I found a potentially deadlock in
{{Execution#tryAssignResource}}. I'll appreciate it if you can help confirm or
reason it.
{{Execution#tryAssignResource}} is called when
{{Execution#calculatePreferredLocations}} finished when, if
{{LocationPreferenceConstraint.ALL}} configured, on
{{ExecutionVertex#getPreferredLocationsBasedOnInputs}} called,
all of {{Execution#taskManagerLocationFuture}} finished when
{{Execution#tryAssignResource}} called.
> Remove ASSIGNED_SLOT_UPDATER in Execution.tryAssignResource
> -----------------------------------------------------------
>
> Key: FLINK-12039
> URL: https://issues.apache.org/jira/browse/FLINK-12039
> Project: Flink
> Issue Type: Sub-task
> Components: Runtime / Coordination
> Reporter: Andrey Zagrebin
> Assignee: vinoyang
> Priority: Minor
>
> After making access to ExecutionGraph single-threaded in FLINK-11417, we can
> simplify execution slot assignment in Execution.tryAssignResource and get rid
> of ASSIGNED_SLOT_UPDATER as it happens now only in one JM main thread.
> Though, it seems that we have to keep `assignedResource` as volatile at the
> moment which could be further investigated.
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
