[jboss-jira] [JBoss JIRA] (JBWEB-269) Memory Leak when Custom Tags Throw Runtime Exception

[Martin Ball (JIRA)]

     [ https://issues.jboss.org/browse/JBWEB-269?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Martin Ball updated JBWEB-269:
------------------------------

    Attachment: Generator.java.patch
                tagLeakTest.zip
                tagLeakTest.jmx


Attached:

1. Generator.java.patch which contains the proposed fix.

2. tagLeakTest.zip which containing a simple test project, built with gradle, pre-built war is in the build dir.

3. tagLeakTest.jmx which is a jmeter test that calls the test page /tagLeakTest/nestedtagWithException.jsp to provoke the leak. Also has other page requests defined which call test pages which should not leak in either case, but were added to prove the patched code worked correctly in normal use.
                
> Memory Leak when Custom Tags Throw Runtime Exception
> ----------------------------------------------------
>
>                 Key: JBWEB-269
>                 URL: https://issues.jboss.org/browse/JBWEB-269
>             Project: JBoss Web
>          Issue Type: Bug
>      Security Level: Public(Everyone can see) 
>          Components: Tomcat
>            Reporter: Martin Ball
>            Assignee: Remy Maucherat
>            Priority: Minor
>         Attachments: Generator.java.patch, tagLeakTest.jmx, tagLeakTest.zip
>
>
> There appears to be a bug in org.apache.jasper.compiler.Generator which in cases where a custom tag throws a runtime exception will lead to a memory leak. The problem is that the code generated only wraps the call to the taghandler pool reuse() method in a try/finally block if the tag implements javax.servlet.jsp.tagext.TryCatchFinally. So for any tag that does not implement TryCatchFinally if a runtime exception occurs reuse() is never called, which means the tag that threw the exception and any tags it is nested within are not returned to the pool, so new instances are created every request and the instance manager is left holding references to the associated interceptor/injection instances.
> The fix is to always wrap the code in try/finally so that reuse()/release() is always called and then optionally add the catch block/calls to the TryCatchFinally methods doCatch/doFinally when required.
> I have tested the leak/a fix against both JBoss Web 7.0.17.FINAL and 7.2.0.Final. I will attach my test war/jmeter project which demonstrates the issue along with my patch. When running the test on unpatched code we observe a continual increase in the instances of com.test.tag.SimpleTag and com.test.tag.SimpleExceptionTag, with the patched code the issue no longer occurs. The test also contains calls to jsps which do not throw exceptions/throw exceptions but implement TryCatchFinally, this is to check that the patched code still generates the try/catch/finally properly and still behaves as expected in normal operation.
> We can verify the cause of the problem by debugging and adding a breakpoint to org.apache.jasper.runtime.PerThreadTagHandlerPool.reuse(Tag handler) and then hitting the test page http://hostname:port/tagLeakTest/nestedtagWithException.jsp, without the patch reuse() will not be called, with the patch it will. 

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira