Carsten Ziegeler wrote:

Sylvain Wallez wrote:
So the question is:
- why is the pipeline automatically released? Is to avoid memory leaks?
Yes, in 2.1.x there was no good place for explicitly releasing the
pipeline; this required some incompatible changes we did in 2.2; the
pipelines are released there explicitly and not automatically.
If the pipeline is not released properly, this means that all pipeline
components are not released, creating a hugh memory leak.

I see, and cleaning up is better than letting memory leak.

But I have a problem with the InheritableThreadLocal, which can also create leaks, as one of its effects is that child threads inherits the environment stack of their parent thread at the time where they (the children) were created.

If these child threads are created and terminated within the scope of the parent's thread request, then this is no problem (e.g. in parallel cinclude), but it is a problem for long-running background threads such as those created by the RunnableManager (used by the scheduler) or, as in my case, by user code.

So IMO, we should change the way environment inheritance between threads is managed: - if a child thread runs in the scope of its parent, then a special CocoonThread class (extends Thread) should be used, which will copy the needed environment data before launching its Runnable. - normal threads created with "new Thread()" inherit nothing and can therefore work in a clean environment.

This approach requires two changes in the 2.1 code base (haven't checked 2.2 yet):
- DefaultIncludeCacheManager.load() in o.a.c.transformation.helpers
- maybe PortalManagerImpl.loadCoplet() in o.a.c.webapps.portal.components

In these two places, the CocoonThread proposed above have to be used to inherit the parent environment. In all other places where "new Thread()" is called, inheriting the parent environment is not needed and worse, can be harmful.

WDYT?

Sylvain

--
Sylvain Wallez                        Anyware Technologies
http://apache.org/~sylvain            http://anyware-tech.com
Apache Software Foundation Member     Research & Technology Director

Reply via email to