[ 
https://issues.apache.org/jira/browse/FLINK-15169?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16992455#comment-16992455
 ] 

Zhu Zhu edited comment on FLINK-15169 at 12/10/19 11:27 AM:
------------------------------------------------------------

Execution#processFail is not invoked on DefaultScheduler internal task errors 
(e.g. slot allocation timeout), #handleTaskFailure is directly invoked and it 
directly cancels all related tasks (in legacy scheduler it fails the error task 
and cancels other tasks).


was (Author: zhuzh):
1. Execution#processFail is only invoked on task failures, so global failures 
are not set (in legacy scheduler the error will be set to 
ExecutionGraph#failureInfo)
2. Execution#processFail is not invoked on DefaultScheduler internal task 
errors (e.g. slot allocation timeout), #handleTaskFailure is directly invoked 
and it cancels all related tasks (in legacy scheduler it fails the error task 
and cancels other tasks).

> Errors happen in the scheduling of DefaultScheduler is not shown in WebUI
> -------------------------------------------------------------------------
>
>                 Key: FLINK-15169
>                 URL: https://issues.apache.org/jira/browse/FLINK-15169
>             Project: Flink
>          Issue Type: Bug
>          Components: Runtime / Coordination
>    Affects Versions: 1.10.0
>            Reporter: Zhu Zhu
>            Priority: Blocker
>             Fix For: 1.10.0
>
>
> WebUI relies on {{ExecutionGraph#failureInfo}} and {{Execution#failureCause}} 
> to generate error info (via 
> {{JobExceptionsHandler#createJobExceptionsInfo}}). 
> Errors happen in the scheduling of DefaultScheduler are not recorded into 
> those fields, thus cannot be shown to users in WebUI (nor via REST queries).
> To solve it, 
> 1. global failures should be recorded into {{ExecutionGraph#failureInfo}}, 
> via {{ExecutionGraph#initFailureCause}} which can be exposed as 
> {{SchedulerBase#initFailureCause}}.
> 2. for task failures, one solution I can think of is to avoid invoking 
> {{DefaultScheduler#handleTaskFailure}} directly on scheduler's internal 
> failures. Instead, we can introduce 
> {{ExecutionVertexOperations#fail(ExecutionVertex)}} to hand the error to 
> {{ExecutionVertex}} as a common failure.
> cc [~gjy]



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

Reply via email to