flink-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Yun Tang (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (FLINK-12096) Flink counter state getting stuck on usin
Date Wed, 03 Apr 2019 11:53:05 GMT

    [ https://issues.apache.org/jira/browse/FLINK-12096?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16808647#comment-16808647
] 

Yun Tang commented on FLINK-12096:
----------------------------------

[~yogesh_n] Would you please provide more information such as exception stack, jobamanger
and taskmanager logs. We actually cannot get any helpful messages from your plain description.
If no exception appeared, you could login the machine which contains the stuck task to see
where java process hangs and shares the screenshot to us. It would also be welcome if you
could reproduce this with specific codes and share it to us.

Moreover, in my point of view, {{RockDBStatateBackend}} actually behaves almost the same as
{{FsStateBackend}} except the performance for users. And I cannot understand what "FS State
backend was used because the counter value state was not getting updated from the first record
itself when using RocksDBS State Backend" means here.

To be honest, from my experience, the problem you describe here might not be caused by a Flink
block bug. If you don't mind, we should better discuss in the user mail list first and then
create a issue here if we confirm this is really a bug.

> Flink counter state getting stuck on usin
> -----------------------------------------
>
>                 Key: FLINK-12096
>                 URL: https://issues.apache.org/jira/browse/FLINK-12096
>             Project: Flink
>          Issue Type: Bug
>          Components: Runtime / State Backends
>            Reporter: Yogesh Nandigamq
>            Priority: Blocker
>              Labels: Counters, ValueState
>
> Working on a job to dynamically create counters for a keyed stream using counter value
state and FileSystem State Backend. It was observed that the counters are stuck after a few
records which is most probably because of the counter value state not getting updated. 
> FS State backend was used because the counter value state was not getting updated from
the first record itself when using RocksDBS State Backend.



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

Mime
View raw message