[
https://issues.apache.org/jira/browse/FLINK-6295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15972561#comment-15972561
]
ASF GitHub Bot commented on FLINK-6295:
---------------------------------------
Github user WangTaoTheTonic commented on the issue:
https://github.com/apache/flink/pull/3709
I'm not a akka expert. As we observed, the status of cancelled tasks will
be updated to running only when gc happens in JM.
Way to reproduce:
1. launch a flink job with ha mode
2. restart zookeeper(to make tasks failed)
3. after tasks recovered, check if status of tasks are running or
cancelled(if there's gc happens, tasks' status showed in web frontend will be
same with the actual states, or the tasks' status are delayed, may cause
inconsistend with those in backend)
We oberved such phenomemon in yarn mode, and it is fixed after this patch.
> use LoadingCache instead of WeakHashMap to lower latency
> --------------------------------------------------------
>
> Key: FLINK-6295
> URL: https://issues.apache.org/jira/browse/FLINK-6295
> Project: Flink
> Issue Type: Bug
> Components: Webfrontend
> Reporter: Tao Wang
> Assignee: Tao Wang
>
> Now in ExecutionGraphHolder, which is used in many handlers, we use a
> WeakHashMap to cache ExecutionGraph(s), which is only sensitive to garbage
> collection.
> The latency is too high when JVM do GC rarely, which will make status of jobs
> or its tasks unmatched with the real ones.
> LoadingCache is a common used cache implementation from guava lib, we can use
> its time based eviction to lower latency of status update.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
