[
https://issues.apache.org/jira/browse/FLINK-6042?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17265799#comment-17265799
]
Matthias edited comment on FLINK-6042 at 1/15/21, 8:08 AM:
-----------------------------------------------------------
{quote}Thanks for the proposal [~mapohl]. I have a few comments
1) When doing restarts with the new scheduler, then we will recreate the
{{ExecutionGraph}}. Hence, exposing these error infos on the
{{AccessExecutionGraph}} might not work.
{quote}
My plan was to not expose the {{ErrorInfo}} through {{ExecutionGraph}} but
through {{ArchivedExecutionGraph}} which is a kind of serializable copy of
{{ExecutionGraph}} and also implements {{AccessExecutionGraph}}. Restarting the
{{ExecutionGraph}} shouldn't harm the {{ErrorInfo}} collection as it is held in
the {{SchedulerNG}} implementation. The {{ErrorInfo}} collection would be used
as an additional parameter when instantiating {{ArchivedExecutionGraph}} in
[SchedulerBase.requestJob()|https://github.com/apache/flink/blob/ac968b83675e64712b4d35dbc166e09808c2156b/flink-runtime/src/main/java/org/apache/flink/runtime/scheduler/SchedulerBase.java#L800].
{quote}2) {{UpdateSchedulerNgOnInternalFailuresListener}} will only be called
if an exception on the JM occurs. If there is a normal task failure, then we
will call {{updateTaskExecutionState}}
{quote}
Thanks for the hint: {{SchedulerNG.updateTaskExecutionState(..)}} should work
as well. This would then collect all the {{Exception}}s thrown in the different
tasks. We could then utilize the {{JobStatusListener}} interface to identify
the {{Exception}} that causes the job to restart and group all previously
collected {{ErrorInfo}}s (that were not already assigned to a different root
cause) under this root cause. Here, I'm not 100% sure, yet, whether triggering
the consolidation of the {{ErrorInfo}}s works that easily.
{quote}3) It would be great to group the exceptions wrt to their restart cycles
in the web UI. So seeing the root causes for a restart and then being able to
expand the view to see the task failures for this specific restart would be
awesome.
{quote}
The {{ErrorInfo}} groups mentioned in 2) could be then returned through the
newly introduced access method described in 1) and forwarded by the
{{JobExceptionsHandler}} to the web UI.
was (Author: mapohl):
{quote}Thanks for the proposal [~mapohl]. I have a few comments
1) When doing restarts with the new scheduler, then we will recreate the
{{ExecutionGraph}}. Hence, exposing these error infos on the
{{AccessExecutionGraph}} might not work.
{quote}
My plan was to not expose the {{ErrorInfo}} through {{ExecutionGraph}} but
through {{ArchivedExecutionGraph}} which is a kind of serializable copy of
{{ExecutionGraph}} and also implements {{AccessExecutionGraph}}. Restarting the
{{ExecutionGraph}} shouldn't harm the {{ErrorInfo}} collection as it is held in
the {{SchedulerNG}} implementation. The {{ErrorInfo}} collection would be used
as an additional parameter when instantiating {{ArchivedExecutionGraph}} in
[SchedulerBase.requestJob()|https://github.com/apache/flink/blob/ac968b83675e64712b4d35dbc166e09808c2156b/flink-runtime/src/main/java/org/apache/flink/runtime/scheduler/SchedulerBase.java#L800].
{quote}2) {{UpdateSchedulerNgOnInternalFailuresListener}} will only be called
if an exception on the JM occurs. If there is a normal task failure, then we
will call {{updateTaskExecutionState}}
{quote}
Thanks for the hint: {{SchedulerNG.updateTaskExecutionState(..)}} should work
as well. This would then collect all the {{Exception}}s thrown in the different
tasks. We could then utilize the {{JobStatusListener}} interface to identify
the {{Exception}} that causes the job to restart and group all previously
collected {{ErrorInfo}}s (that were not already assigned to a different root
cause) under this root cause.
{quote}3) It would be great to group the exceptions wrt to their restart cycles
in the web UI. So seeing the root causes for a restart and then being able to
expand the view to see the task failures for this specific restart would be
awesome.
{quote}
The {{ErrorInfo}} groups mentioned in 2) could be then returned through the
newly introduced access method described in 1) and forwarded by the
{{JobExceptionsHandler}} to the web UI.
> Display last n exceptions/causes for job restarts in Web UI
> -----------------------------------------------------------
>
> Key: FLINK-6042
> URL: https://issues.apache.org/jira/browse/FLINK-6042
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Coordination, Runtime / Web Frontend
> Affects Versions: 1.3.0
> Reporter: Till Rohrmann
> Assignee: Matthias
> Priority: Major
>
> Users requested that it would be nice to see the last {{n}} exceptions
> causing a job restart in the Web UI. This will help to more easily debug and
> operate a job.
> We could store the root causes for failures similar to how prior executions
> are stored in the {{ExecutionVertex}} using the {{EvictingBoundedList}} and
> then serve this information via the Web UI.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
