Details
Minor Description
This may not qualify as a bug since its so easy to provide override using hivemind, but instead an improvement for future releases. Also provided an optional page pool implementation.
Tapestry 4.0 PagePool implementation doesnt remove idle pages. When having hundred of pages with a lot of components and high user concurrency, tapestry generates many instances of each page. These instances are pooled and never unreferenced and never garbage collected. Our 1GB heap eventually fills up and reduces the memory available to other parts of the application. Eventually causes OutOfMemoryException.
We ensured caching is enabled and config change made to use unique locale, but still, heap eventually fills up.
With Hivemind its very easy to override the pagePool and provide different implementation. A page pool that uses softReferences is a good option.
Acording to java api:
All soft references to softly-reachable objects are guaranteed to have been cleared before the virtual machine throws an OutOfMemoryError.
This prevents OEM due to heap filling up.
New SoftPagePool only changes a couple of lines to use soft references.
public synchronized Object get(Object key)
{ List pooled = (List) _pool.get(key); if (pooled == null || pooled.isEmpty()) return null; _count--; SoftReference reference = (SoftReference) pooled.remove(0); //returns null if has been cleared by GC same as pool where empty return reference.get(); } public synchronized void store(Object key, Object value)
{
List pooled = (List) _pool.get(key);
if (pooled == null)
{ pooled = new LinkedList(); _pool.put(key, pooled); } SoftReference reference = new SoftReference(value);
pooled.add(reference);
_count++;
}
Additionally the page pool implementation can use a clean idle pages mechanism using same design as Tapestry 3.0 ThreadJanitor or setting a timestamp when storing pages and a TimerTask and Timer wich receives events on registry shutdown. All this by using hivemind overriding features.