[jira] [Comment Edited] (FLINK-38314) Flink dashboard incorrectly shows units for RocksDB `live-sst-files-size` metric for timer state

["Jon Cavallie Mester (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/FLINK-38314?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18023321#comment-18023321 ] 

Jon Cavallie Mester edited comment on FLINK-38314 at 9/27/25 4:44 PM:
----------------------------------------------------------------------

Interested in working on this issue.

I propose adjusting the web dashboard’s HumanizeChartNumericPipe logic:
* Treat *-size metrics and RocksDB rocksdb_live-sst-files-size as bytes.
* Restrict duration detection so that timer_state is not misclassified, while still recognizing camelCase forms (e.g. BusyTimeMs, AlignmentTime) and suffixes such as _median, _max, etc.
* Add guards to prevent counters (e.g. numRecordsIn) from being rendered as durations.
The change is web-only and does not affect any public APIs. I’ve verified the updated regexes manually (regex101) and in the dashboard against a range of metrics (RocksDB size, checkpoint times, latencies, counters, and rates).

I noticed the dashboard project does not include a test suite or test dependencies, so no automated tests are added here. If you want, I’d be happy to open a follow-up improvement ticket to add lightweight unit tests for the pipes.

If this approach sounds OK, please assign the issue to me.


was (Author: JIRAUSER311063):
Hi, I’d like to work on this issue.

My proposed fix adjusts the web dashboard’s HumanizeChartNumericPipe logic:
* Treat *-size metrics and RocksDB rocksdb_live-sst-files-size as bytes.
* Restrict duration detection so that timer_state is not misclassified, while still recognizing camelCase forms (e.g. BusyTimeMs, AlignmentTime) and suffixes such as _median, _max, etc.
* Add guards to prevent counters (e.g. numRecordsIn) from being rendered as durations.
The change is web-only and does not affect any public APIs. I’ve verified the updated regexes manually (regex101) and in the dashboard against a range of metrics (RocksDB size, checkpoint times, latencies, counters, and rates).

I noticed the dashboard project does not include a test suite or test dependencies, so no automated tests are added here. If you want, I’d be happy to open a follow-up improvement ticket to add lightweight unit tests for the pipes.

If this approach sounds OK, please assign the issue to me and I’ll prepare a PR.

> Flink dashboard incorrectly shows units for RocksDB `live-sst-files-size` metric for timer state
> ------------------------------------------------------------------------------------------------
>
>                 Key: FLINK-38314
>                 URL: https://issues.apache.org/jira/browse/FLINK-38314
>             Project: Flink
>          Issue Type: Bug
>            Reporter: Aleksandr Iushmanov
>            Priority: Minor
>
> For example metric like this: 
> `column_family._timer_state/event_user-timers.rocksdb_live-sst-files-size` we don't match and of `bytes` regular expressions, however we do match `time`: [https://github.com/apache/flink/blame/master/flink-runtime-web/web-dashboard/src/app/components/humanize-chart-numeric.pipe.ts#L41]
>  
> In result, state size for timers in rocks DB is shown in flink UI as duration in milliseconds in "Numeric" mode



--
This message was sent by Atlassian Jira
(v8.20.10#820010)