flink-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF GitHub Bot (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (FLINK-2793) Redirect to leading JobManager web fronted in non-standalone mode
Date Mon, 05 Oct 2015 16:46:26 GMT

    [ https://issues.apache.org/jira/browse/FLINK-2793?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14943628#comment-14943628
] 

ASF GitHub Bot commented on FLINK-2793:
---------------------------------------

Github user tillrohrmann commented on a diff in the pull request:

    https://github.com/apache/flink/pull/1202#discussion_r41167117
  
    --- Diff: flink-runtime-web/src/main/java/org/apache/flink/runtime/webmonitor/JobManagerArchiveRetriever.java
---
    @@ -70,28 +104,68 @@ public ActorGateway getArchiveGateway() {
     		return archiveGateway;
     	}
     
    +	/**
    +	 * Returns the current redirect address or <code>null</code> if the job
manager associated with
    +	 * this web monitor is leading. In that case, work with the gateway directly.
    +	 */
    +	public String getRedirectAddress() {
    +		return redirectWebMonitorAddress;
    +	}
     
     	@Override
     	public void notifyLeaderAddress(String leaderAddress, UUID leaderSessionID) {
     		if (leaderAddress != null && !leaderAddress.equals("")) {
     			try {
    -				ActorRef jobManager = AkkaUtils.getActorRef(
    -						leaderAddress,
    -						actorSystem,
    +				ActorRef jobManager = AkkaUtils.getActorRef(leaderAddress, actorSystem,
    --- End diff --
    
    This is a blocking call which will block the `NodeCache's` thread which calls the `NodeCacheListener`.
This should be avoided by using `AkkaUtils.getActorRefFuture` and a `Promise`.


> Redirect to leading JobManager web fronted in non-standalone mode
> -----------------------------------------------------------------
>
>                 Key: FLINK-2793
>                 URL: https://issues.apache.org/jira/browse/FLINK-2793
>             Project: Flink
>          Issue Type: Improvement
>          Components: JobManager
>    Affects Versions: master
>            Reporter: Ufuk Celebi
>            Assignee: Ufuk Celebi
>
> In case of a non-standalone recovery mode, the job manager frontend of non-leading job
managers prints the job manager information of its associated job manager. Because the job
manager is not leading, nothing shows up.
> The web frontend cannot directly communicate with the leading job manager, because many
job manager structures like the execution graph are not serializable.
> A work around is to redirect to the web frontend of the leading job manager. This makes
sure that all interesting information is presented.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message