[jira] [Commented] (FLINK-10914) Prefer notifying status changed to waiting and checking in ExecutionGraph tests

["Stephan Ewen (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/FLINK-10914?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16691404#comment-16691404 ] 

Stephan Ewen commented on FLINK-10914:
--------------------------------------

In general, notification-based is always better than "poll based", so I agree.
It may be that it was tricky to implement in that case due to the concurrency model.

There are thoughts to change the concurrency model of the execution graph (single threaded dispather, like actor-style).

> Prefer notifying status changed to waiting and checking in ExecutionGraph tests
> -------------------------------------------------------------------------------
>
>                 Key: FLINK-10914
>                 URL: https://issues.apache.org/jira/browse/FLINK-10914
>             Project: Flink
>          Issue Type: Improvement
>          Components: Tests
>    Affects Versions: 1.8.0
>            Reporter: TisonKun
>            Assignee: TisonKun
>            Priority: Major
>             Fix For: 1.8.0
>
>
> Currently there are several tests of {{ExecutionGraph}} based on {{ExecutionGraphTestUtils#waitUntilJobStatus}} and {{ExecutionGraphTestUtils#waitUntilExecution(Vertex)State}}. I notice that we have {{JobStatusListener}} s and {{ExecutionStatusListener}} s registered on a {{ExecutionGraph}}. It is possible to replace the waiting version with a notifying version, which is with less cpu workload and more reliable(the waiting version might miss some very quick status/state changes).
> What do you think [~till.rohrmann] [~Zentol]?



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)