You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Till Rohrmann (JIRA)" <ji...@apache.org> on 2019/02/28 12:36:00 UTC
[jira] [Comment Edited] (FLINK-11749) Returns the stored leadership
info on ZooKeeperLeaderRetrievalService started
[ https://issues.apache.org/jira/browse/FLINK-11749?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16780454#comment-16780454 ]
Till Rohrmann edited comment on FLINK-11749 at 2/28/19 12:35 PM:
-----------------------------------------------------------------
I think I thought that the problem was that if the {{NodeCache}} is restarted (overlooking that this is not possible), that you won't refresh the {{data}} field. Looking at the code now, I think it won't work that the same {{JobMaster}} gets assigned the leadership twice when using {{ZooKeeper}}. I think this should be fixed.
It will actually be fixed with my refactoring of the {{JobMaster}} (https://github.com/tillrohrmann/flink/tree/removeSuspendFromJobMaster) for which I haven't created an issue yet. The idea there would be that you create for each leadership session a new {{JobMaster}}.
was (Author: till.rohrmann):
I think I thought that the problem was that if the {{NodeCache}} is restarted (overlooking that this is not possible), that you won't refresh the {{data}} field. Looking at the code now, I think it won't work that the same {{JobMaster}} gets assigned the leadership twice when using {{ZooKeeper}}. I think this should be fixed. It will actually be fixed with my refactoring of the {{JobMaster}} (https://github.com/tillrohrmann/flink/tree/removeSuspendFromJobMaster) for which I haven't created an issue yet.
> Returns the stored leadership info on ZooKeeperLeaderRetrievalService started
> -----------------------------------------------------------------------------
>
> Key: FLINK-11749
> URL: https://issues.apache.org/jira/browse/FLINK-11749
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Coordination
> Affects Versions: 1.9.0
> Reporter: TisonKun
> Assignee: TisonKun
> Priority: Major
> Fix For: 1.9.0
>
>
> Currently, such as in {{JobMaster}}, if we stop a {{ZooKeeperLeaderRetrievalService}} and restart it, because the cached node data isn't change, we have to manually flush it.
> I propose on {{ZooKeeperLeaderRetrievalService}} stated, get {{ChildData}} from {{NodeCache}} and if the data isn't null, generate LeadershipInfo(Address and UUID) and notify the listener.
> cc [~till.rohrmann]
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)