[
https://issues.apache.org/jira/browse/ARIES-1603?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15488098#comment-15488098
]
Timothy Ward commented on ARIES-1603:
-------------------------------------
bq. What is the purpose of the classloader clone? The prevent an execution
after a service shutdown?
The class loader clone exists for a few reasons:
* For consistency with the class type mediation that requires a class loader
clone for CGLib
* To avoid future issues if/when additional types need to be referenced by the
proxy
* To avoid lifecycle issues if the async service is uninstalled - using the raw
bundle's class loader would leave proxy types defined on it after the Async
bundle was removed from the system.
Technically the way in which the code is written is not leaking class loaders.
A different class loader is used for each mediation, but these will all be
eligible for garbage collection as soon as the mediator is garbage collected.
Mediators are intended to be cached and reused but if the application code
makes a lot of them and holds onto them then that will cause a lot of
ClassLoaders to be created. In your application's case I'm not sure where the
mediators end up being held, but that's the reason that an OOM occurs. This may
indicate a leak in the application, or just a high load of concurrent tasks
that should be throttled more carefully.
Obviously the application code needs some attention, but having looked more
deeply at this the Async service *may* be able to weakly cache the class loader
to be used for mediation. This should dramatically reduce the memory overhead
for your application (although it won't fix a leak in it if there is one) and
help with the scaling.
> Aries async classloader leak
> ----------------------------
>
> Key: ARIES-1603
> URL: https://issues.apache.org/jira/browse/ARIES-1603
> Project: Aries
> Issue Type: Bug
> Affects Versions: async-1.0.2
> Reporter: Paul Thevenot
>
> Using the async service with repeated tasks during one hour leads to an out
> of memory. It's due to the very large amount of class loaded (in 10 min we
> get 20000 classes loaded). The metaspace is growing indefinitely.
> It seems to come from the privMediate method of the AsyncService where the
> classloader of the target service is cloned.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
