You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Maximilian Michels (JIRA)" <ji...@apache.org> on 2016/08/01 16:06:20 UTC

[jira] [Commented] (FLINK-4299) Show loss of job manager in Client

    [ https://issues.apache.org/jira/browse/FLINK-4299?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15402305#comment-15402305 ] 

Maximilian Michels commented on FLINK-4299:
-------------------------------------------

Good idea! We should add some messages in the {{JobClientActor}} class upon receiving a new leader or disconnecting from an existing.

> Show loss of job manager in Client
> ----------------------------------
>
>                 Key: FLINK-4299
>                 URL: https://issues.apache.org/jira/browse/FLINK-4299
>             Project: Flink
>          Issue Type: Improvement
>          Components: Client
>            Reporter: Ufuk Celebi
>             Fix For: 1.1.0
>
>
> If the client looses the connection to a job manager and the job recovers from this, the client will only print the job status as {{RUNNING}} again. It is hard to actually notice that something went wrong and a job manager was lost.
> {code}
> ...
> 08/01/2016 14:35:43     Flat Map -> Sink: Unnamed(8/8) switched to RUNNING
> 08/01/2016 14:35:43     Source: Custom Source(6/8) switched to RUNNING
> <------ EVERYTHING'S RUNNING ------>
> 08/01/2016 14:40:40     Job execution switched to status RUNNING <--- JOB MANAGER FAIL OVER
> 08/01/2016 14:40:40     Source: Custom Source(1/8) switched to SCHEDULED
> 08/01/2016 14:40:40     Source: Custom Source(1/8) switched to DEPLOYING
> 08/01/2016 14:40:40     Source: Custom Source(2/8) switched to SCHEDULED
> ...
> {code}
> After {{14:35:43}} everything is running and the client does not print any execution state updates. When the job manager fails, the job will be recovered and enter the running state again eventually (at 14:40:40), but the user might never notice this.
> I would like to improve on this by printing some messages about the state of the job manager connection. For example, between {{14:35:43}} and {{14:40:40}} it might say that the job manager connection was lost, a new one established, etc.



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