[
https://issues.apache.org/jira/browse/FLINK-37238?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sai Sharath Dandi updated FLINK-37238:
--------------------------------------
Description:
The Flink Web UI shows an incorrect relationship between job vertices by not
considering scaled parallelism. For example, in the screenshot, we can see a
forward connection between the first 2 vertices. However, it is an implicit
rescale connection as all the task managers running the second vertex are
receiving the data.
I'm guessing the bug happens because both the connection is forward connection
when parallelism overrides are not considered(i.e.. both the vertex parallelism
are 256 without considering the parallelism overrides). This can be misleading
to users when trying to debug issues with their job.
was:
The Flink Web UI shows an incorrect relationship between job vertices by not
considering scaled parallelism. For example, in the screenshot, we can see a
forward connection between the first 2 vertices. However, it is an implicit
rescale connection as all the task managers running the second vertex are
receiving the data.
I'm guessing the bug happens because both the vertex parallelism are 256
without considering the parallelism overrides which is not being reflected in
the web UI. This can be misleading to users when trying to debug issues with
their job.
> Flink Web UI does not take scaled parallelism into account when showing job
> graph
> ---------------------------------------------------------------------------------
>
> Key: FLINK-37238
> URL: https://issues.apache.org/jira/browse/FLINK-37238
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Web Frontend
> Affects Versions: 1.18.1
> Reporter: Sai Sharath Dandi
> Priority: Minor
> Attachments: Screenshot 2025-01-29 at 4.39.00 PM.png
>
>
> The Flink Web UI shows an incorrect relationship between job vertices by not
> considering scaled parallelism. For example, in the screenshot, we can see a
> forward connection between the first 2 vertices. However, it is an implicit
> rescale connection as all the task managers running the second vertex are
> receiving the data.
>
>
> I'm guessing the bug happens because both the connection is forward
> connection when parallelism overrides are not considered(i.e.. both the
> vertex parallelism are 256 without considering the parallelism overrides).
> This can be misleading to users when trying to debug issues with their job.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
