Re: [WebMacro-user] Thread leak with ReloadingCacheManager

[Keats K. <ke...@xa...>]

I haven't had a chance to validate what you are saying, but I know there 
has been a longstanding issue with getting WM Web apps to unload 
properly.  It sounds like you've finally nailed it.

I think your approach of adding a destroy method to the Broker makes 
sense.  If you're willing to work on it that would be great.

Nice work.

Keats

Alex Twisleton-Wykeham-Fiennes wrote:

>All,
>
>I've just been spending a couple of days trying to track down why memory gets 
>consumed when reloading the same web application under tomcat which according 
>to my research is related to class loader confusion [1].
>
>During the course of my testing (running OptimizeIt across the tomcat5 process 
>and analysing class loading and object creation across multiple reloads), 
>I've been noticing the number of Threads growing with each reload which 
>didn't seem correct at all.
>
>The increasing threads were getting allocated by ReloadingCacheManager and 
>where being utilised by the ClockDaemon process for triggering the expiration 
>of assets from the template and url Providers (using the standard webmacro 
>defaults file).
>
>These would normally get destroyed when the destroy() method was invoked on 
>the ReloadingCacheManager, but I couldn't find any where that enabled me to 
>do this easily.  I would have expected this to be invoked in the destroy() 
>method of WMServlet but this is not the case:-
>    public synchronized void destroy ()
>    {
>        stop();
>        _log.notice("stopped: " + this);
>        _wm = null;
>        _started = false;
>        super.destroy();
>    }
>
>In the end, I found it very hard to find a way to cleanly persuade the 
>webmacro instance to shut itself down properly, and ended up putting 
>something like this in my stop() method for my subclass of WMServlet:-
>
>  protected void stop() 
>  {
>    // tidy up my stuff... <snip>
>    Broker broker = getBroker();
>    try {
>      Provider templateProvider = broker.getProvider("template");
>      templateProvider.destroy();
>      Provider urlProvider = broker.getProvider("url");
>      urlProvider.destroy();
>    } catch (NotFoundException nfEx) {
>      nfEx.printStackTrace();
>    }
>    super.stop();
>  }
>
>Now this nicely tidies up the appropriate resources and when I reload this 
>application lots of times the thread count stays consistent, my memory is 
>much more under control and everything is a bit smoother.
>
>So, my question is:-
>
>Assuming that ReloadingCacheManager has to be destroy()ed correctly to get it 
>to shut down, and assuming that the application author doesn't know which 
>Providers are going to be implemented using a ReloadingCacheManager (as it is 
>defined by the properties classes at runtime), then what would be the 
>"correct" way of ensuring that the framework is correctly shut down?
>
>Previously there was a definition of WebMacro.destroy() that stated "Call this 
>method when you are finished with WebMacro. If you don't call this method, 
>the Broker and all of WebMacro's caches may not be properly shut down, 
>potentially resulting in loss of data, and wasted memory. This method is 
>called in the finalizer, but it is best to call it as soon as you know you 
>are done with WebMacro.", but this has since been dropped from the package.
>
>My gut feeling is that there should be a destroy() method on the Broker that 
>then invokes destroy() on all of the sub-components that it is maintaining 
>references to, but I am not that familiar with the webmacro hierarchy...
>
>All tests have been run against 2.0b1, but the changelog to 2.0RC1 doesn't 
>touch upon any of these changes as far as I can see.
>
>Alex
>
>[1] 
>http://opensource2.atlassian.com/confluence/spring/pages/viewpage.action?pageId=2669
>
>  
>