[jira] [Commented] (CAMEL-24529) camel-djl: the loaded ZooModel is never closed (native-memory leak on restart)

["Claus Ibsen (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/CAMEL-24529?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18109204#comment-18109204 ] 

Claus Ibsen commented on CAMEL-24529:
-------------------------------------

Fixed via https://github.com/apache/camel/pull/25853 (commit 6fd66062e6e6). Resolving to sync JIRA status with the merged fix.

_Claude Code on behalf of Claus Ibsen_

> camel-djl: the loaded ZooModel is never closed (native-memory leak on restart)
> ------------------------------------------------------------------------------
>
>                 Key: CAMEL-24529
>                 URL: https://issues.apache.org/jira/browse/CAMEL-24529
>             Project: Camel
>          Issue Type: Bug
>          Components: camel-ai
>            Reporter: Andrea Cosentino
>            Assignee: Andrea Cosentino
>            Priority: Major
>             Fix For: 4.23.0
>
>
> DJL zoo predictors load a ZooModel (ModelZoo.loadModel) in their constructor, but DJLProducer has no doStop() and AbstractPredictor has no close(), so the native ZooModel (NDManager/engine memory) is never released on route stop/redeploy or CamelContext restart — a native-memory leak. The huggingface sibling cleans up (AbstractTaskPredictor.close via HuggingFaceProducer.doStop). The per-exchange Predictor objects are correctly try-with-resources; only the model is leaked.
> Fix: add close() to AbstractPredictor (closing the model), keep the predictor reference in DJLProducer, and override DJLProducer.doStop() to close it.



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