[ 
https://issues.apache.org/jira/browse/XALANJ-2195?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18014454#comment-18014454
 ] 

Joe Kesselman commented on XALANJ-2195:
---------------------------------------

FWIW, in 2.7.3 there is a releaseXMLReader() function which appears to address 
this. This is being invoked from multiple places through the code (looks like 
DTMManagerDefault, XSLTCDTMManager, TransformerImpl, TransformerIdentityImpl, 
TraxFilter). It would be nice to know whether the issue still persists, and 
ideally to have a sample which provokes it.

> Memory leak in XMLReaderManager
> -------------------------------
>
>                 Key: XALANJ-2195
>                 URL: https://issues.apache.org/jira/browse/XALANJ-2195
>             Project: XalanJ2
>          Issue Type: Bug
>          Components: Xalan
>    Affects Versions: 2.7
>            Reporter: Marko Strukelj
>            Priority: Blocker
>         Attachments: gc-roots.jpg, retained-object-sizes.jpg, 
> xalan-j2-2.6.0-xmlmemoryleak-tm2l.patch
>
>
> In class org.apache.xml.utils.XMLReaderManager 
> getXMLReader() method creates a new XMLReader (i.e. SAXParser) and stores it 
> into ThreadLocal.
> releaseXMLReader() does not remove (set to null) ThreadLocal thus creating a 
> permanent leak.
> Unfortunately the size of the cached Reader is typically dependent upon the 
> size of the XML document you process (depends on implementation but this is 
> the case with xerces SAXParser). In heavy load server environments with 
> thread pools of tens and hundreds of threads the server sustains a 
> significant memory leak (hundreds of megabytes - depending on the XML 
> document sizes and number of threads in a thread pools).
> A fix is trivial:
> Put the following line at the end of releaseXMLReader method:
> m_readers.set(null);
> I wonder, why is reader stored in ThreadLocal in the first place?



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

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to