[jira] [Commented] (FLINK-35817) High RSS Memory Growth in Long-Running Flink Tasks with Low Heap Usage: Is Prometheus Metrics Reporting the Culprit?

["Zhanghao Chen (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/FLINK-35817?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17869170#comment-17869170 ] 

Zhanghao Chen commented on FLINK-35817:
---------------------------------------

The screenshot indicates high usage of JVM direct memory, which is off-heap.

> High RSS Memory Growth in Long-Running Flink Tasks with Low Heap Usage: Is Prometheus Metrics Reporting the Culprit?
> --------------------------------------------------------------------------------------------------------------------
>
>                 Key: FLINK-35817
>                 URL: https://issues.apache.org/jira/browse/FLINK-35817
>             Project: Flink
>          Issue Type: Bug
>          Components: Runtime / Metrics
>    Affects Versions: 1.16.3
>            Reporter: ZHangTianLong
>            Priority: Blocker
>         Attachments: image-2024-07-11-22-01-31-775.png
>
>
> We're encountering an issue in our Flink tasks where the RSS (Resident Set Size) memory steadily increases over time, despite low heap memory usage.
> We seek confirmation if Prometheus Metrics reporting is indeed the root cause of the RSS memory surge and suggestions for optimization.
> !image-2024-07-11-22-01-31-775.png!



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