[ 
https://issues.apache.org/jira/browse/POOL-161?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12875907#action_12875907
 ] 

Phil Steitz commented on POOL-161:
----------------------------------

We seem to be talking about three different issues here:

1. The issue that Mark's commit fixed - potential for memory leak when the 
class loader for the eviction thread is not the same loader that loaded the 
library.  IIUC, Mark's patch fully fixes this issue and any other potential for 
classloader leaks associated with the evictor.  Correct?

2. Feature request to support the use case in the issue description.  Looks to 
me like Mark's patch actually does fully fix the issue as narrowly defined 
above.  I must be missing something here.

3. Feature request for pool to work in arbitrary multi-classloader environments

I agree with Mark that it is not obvious that we should aim to support 2., and 
I think we have some work do to in defining exactly what 3. means.  I am 
strongly +1 on doing everything we can to avoid leaks. 

> ContextClassLoader problems for the Evictor thread
> --------------------------------------------------
>
>                 Key: POOL-161
>                 URL: https://issues.apache.org/jira/browse/POOL-161
>             Project: Commons Pool
>          Issue Type: Bug
>    Affects Versions: 1.5.4
>            Reporter: Sylvain Laurent
>             Fix For: 2.0
>
>         Attachments: patch_Evictor_CCL.txt, 
> TestGenericObjectPoolClassLoader.patch.txt
>
>
> Since a single Timer is used for several GenericObjectPool instances, this 
> may create classloader issues and a memory leak of one classloader :
> Let's imagine the following scenario :
> - commons-pool.jar is in the classpath of a webapp container (e.g. tomcat).
> - 2 webapps A and B are deployed, each creating an instance of 
> GenericObjectPool for its own usage.
> - each webapp makes use of the idle object evictor and sets a positive number 
> for minIdle
> - first, webapp A instantiates its GenericObjectPool. Since this is the first 
> TimerTask to be created, the Timer instance is created, thus creating a 
> Thread whose ContextClassLoader is the current one, that is webapp A's 
> ContextClassLoader.
> The TimerTask properly creates instances of idle objects in the pool, making 
> use of the ObjectFactory provided by A.
> - then B instantiates its GenericObjectPool. A new TimerTask is created, and 
> it tries to invoke the ObjectFactory provided by B. But when it needs a class 
> that only exists in B webapp, it cannot find it because the 
> ContextClassLoader of the Timer Thread is A's classloader.
> Other side effect : if webapp A is undeployed, but B is still running, then 
> A's webappClassLoader cannot be GCed because the Timer Thread keeps a strong 
> reference to A's classloader (as its context classloader).

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

Reply via email to