You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Zhu Zhu (Jira)" <ji...@apache.org> on 2019/12/10 05:50:00 UTC
[jira] [Updated] (FLINK-14606) Simplify params of
Execution#processFail
[ https://issues.apache.org/jira/browse/FLINK-14606?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Zhu Zhu updated FLINK-14606:
----------------------------
Fix Version/s: (was: 1.10.0)
1.11.0
> Simplify params of Execution#processFail
> ----------------------------------------
>
> Key: FLINK-14606
> URL: https://issues.apache.org/jira/browse/FLINK-14606
> Project: Flink
> Issue Type: Sub-task
> Components: Runtime / Coordination
> Affects Versions: 1.10.0
> Reporter: Zhu Zhu
> Priority: Major
> Fix For: 1.11.0
>
>
> The 3 params fromSchedulerNg/releasePartitions/isCallback of Execution#processFail are quite a mess while they seem to be correlated.
> I'd propose to simplify the prams of processFail by using a {{isInternalError}} to replace those 3 params. {{isInternalError}} is true iff the failure is from TM(strictly speaking, notified from SchedulerBase). This also hardens the handling of cases that a task is successfully deployed but JM does not realize it(see #3 below).
> Here's why these 3 params can be simplified:
> 1. {{fromSchedulerNg}}, true iff the failure is from TM and isLegacyScheduling==false.
> It's only used like this: {{if (!fromSchedulerNg && !isLegacyScheduling()))}}. So it's the same to use {{!isInternalFailure}} to replace it.
> 2. {{releasePartitions}}, true iff the failure is from TM.
> Now the value is exactly the same as {{isInternalFailure}}, we can drop it and use {{isInternalFailure}} instead.
> 3. {{isCallback}}, true iff the failure is from TM or the task is not deployed.
> It's only used like this: {{(!isCallback && (current == RUNNING || current == DEPLOYING))}}.
> So using {{!isInternalFailure}} to replace it would be enough. It is a bit different for the case that a task deployment to a task manager fails, which set {{isCallback}} to true previously. However, it would be safer to signal a cancel call, in case the deployment is actually a success but the response is lost on network.
> cc [~GJL]
--
This message was sent by Atlassian Jira
(v8.3.4#803005)