[ 
https://issues.apache.org/jira/browse/FLINK-15116?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17177720#comment-17177720
 ] 

Till Rohrmann edited comment on FLINK-15116 at 8/14/20, 11:48 AM:
------------------------------------------------------------------

This change seems to cause another regression. When cancelling a job (using the 
per-job mode), we no longer run through the normal 
{{Dispatcher.jobReachedGLoballyTerminalState()}} method because we directly 
stop the cluster. The consequence is that jobs no longer get uploaded to the 
history server. See FLINK-18959 for more details. [~aljoscha] was this an 
intended change?


was (Author: till.rohrmann):
This change seems to cause another regression. When cancelling a job (using the 
per-job mode), we no longer run through the normal 
{{Dispatcher.jobReachedGLoballyTerminalState()}} method because we directly 
stop the cluster. The consequence is that jobs no longer get uploaded to the 
history server. [~aljoscha] was this an intended change?

> Make JobClient stateless, remove AutoCloseable
> ----------------------------------------------
>
>                 Key: FLINK-15116
>                 URL: https://issues.apache.org/jira/browse/FLINK-15116
>             Project: Flink
>          Issue Type: Sub-task
>          Components: API / DataSet, API / DataStream
>            Reporter: Aljoscha Krettek
>            Assignee: Aljoscha Krettek
>            Priority: Major
>              Labels: pull-request-available
>             Fix For: 1.10.0
>
>          Time Spent: 20m
>  Remaining Estimate: 0h
>
> Currently, {{JobClient}} is {{AutoCloseable}} and we require users to close 
> the {{JobClient}} that they get as a result from {{executeAsync()}}. This is 
> problematic because users can simply ignore the result of {{executeAsync()}} 
> and then we will leak the resources that the client has.
> We should change the {{JobClient}} so that it acquires the required 
> {{ClusterClient}} for each method call and closes it again. 
> This means that the users no longer have the burden of managing the JobClient 
> lifecycle, i.e. they can freely ignore the result of executeAsync().



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to