[jira] [Commented] (LIVY-1063) Fix resource leak in Kubernetes service account token refresh

["Nilesh Rathi (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/LIVY-1063?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18101663#comment-18101663 ] 

Nilesh Rathi commented on LIVY-1063:
------------------------------------

Observed in a long-running Livy deployment on Kubernetes: native thread count grew ~every 5 minutes while RefreshServiceAccountTokenThread was active. After reusing the existing client (only updating the OAuth token), thread count stayed flat over a multi-hour soak. 
Token refresh logs continued every ~5 minutes.

> Fix resource leak in Kubernetes service account token refresh
> -------------------------------------------------------------
>
>                 Key: LIVY-1063
>                 URL: https://issues.apache.org/jira/browse/LIVY-1063
>             Project: Livy
>          Issue Type: Bug
>            Reporter: Nilesh Rathi
>            Assignee: Nilesh Rathi
>            Priority: Major
>              Labels: kubernetes
>          Time Spent: 10m
>  Remaining Estimate: 0h
>
> RefreshServiceAccountTokenThread in SparkKubernetesApp recreates DefaultKubernetesClient every 5 minutes during service account token refresh. The previous client is never closed, causing a slow resource leak (native threads / HTTP client pools) in long-running Livy deployments on Kubernetes.
> The client is documented as thread-safe and intended to be created once at init(). The refresh loop should only update the OAuth token on the existing client via config.setOauthToken(), not instantiate a new DefaultKubernetesClient.
> Affected code: server/src/main/scala/org/apache/livy/utils/SparkKubernetesApp.scala (introduced in LIVY-702).



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