[
https://issues.apache.org/jira/browse/FLINK-28283?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17686338#comment-17686338
]
Xintong Song commented on FLINK-28283:
--------------------------------------
I think logging all the task state changes might be necessary.
- We cannot aggregate these logs, because state changing of tasks are
independent from each other.
- We cannot only log failures, because tasks may stuck in some state and never
reaches fail / running state. In such cases, we need to know what state the
task is in. Same for changing log level.
> Improving the log of flink when job start and deploy
> ----------------------------------------------------
>
> Key: FLINK-28283
> URL: https://issues.apache.org/jira/browse/FLINK-28283
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Task
> Affects Versions: 1.14.2
> Reporter: zlzhang0122
> Priority: Major
>
> When running a large job with many operators and subtasks on flink, the
> JobManager and TaskManager will have a huge logs about the subtask executing
> msg such as "XXX switched from CREATED to SCHEDULED、XXX switched from
> SCHEDULED to DEPLOYING 、XXX switched from DEPLOYING to RUNNING 、XXX switched
> from RUNNING to CANCELING、XXX switched from CANCELING to CANCELED", etc. .
> Maybe we can do some improvement about this, such as aggregate these msg to
> reduce the log, or change the log level and only logs the failure msg and
> subtask, etc. Not so sure about the solution, but these msg is really too
> much.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
