Memory leak - Node clean up?

Question

Hey guys,
I'm running into a memory leak and was reading some other threads where people were having their own memory leaks. Is there a way through NetBeans to track memory usage or something to figure out what is consuming memory? My memory leak sounds similar to this gentleman's: [http://forums.sun.com/thread.jspa?threadID=5409788] and [http://forums.sun.com/thread.jspa?threadID=5357401]
Setup:
I have a mySQL database call that returns a bunch of results and in the end converts each record into its own custom node which then gets put into a sequence, which is then given to a listEventsNode (a custom node VBox) and finally put on the stage. As a note each custom node has its own animations, events, images, and shapes.
listEventsNode gets its list of custom nodes from a publicly accessible sequence variable where its contents is simply cleared and new nodes are added by the next MySQL search. I cleared the eventSequence by using delete myCustomNodeSequence.
I even go as far as setting the same sequence null (myCustomNodeSequence = null;). This unfortunately doesn't make any difference in terms of memory usage. java.exe will reach 100MB then crash.
the listEventsNode is bound with eventSequence, this is to ensure that changes to the sequence of custom nodes are immediately reflected in the Vbox (listEventsNode).
ListEventsNode is on the main stage and there is only one instance of it, but what changes is the content in the eventSequence. Even if I clear the contents of the eventSequence, it doesn't appear to "clean up" the memory.
The way I'm doing it is probably breaking every rule in the book. I should probably make it so listEventsNode is its own object which isn't bound to any external variables (such as eventSequence) and in the event a new search takes place I simply delete the listEventsNode and when a new search is complete, it re-adds the node to the scene. Am I on the right track here?
Is there a good "best practices" for JavaFX? For example, a typical mistake that would cause a node to "recreate" itself over and over again in memory when the programmer may have thought it was simply being "written over"? Does this make sense? Because I have a feeling that my application is not deleting the custom nodes that were created for the eventSequence and even if the eventSequence is "deleted" and re-assigned with new custom nodes, the original or previous custom nodes are still residing in memory.
Let me know if you need to see the source or any logs/readouts from NetBeans during execution if this will help.
Thanks for taking the time to read this.
Cheers,
Nik.

Answer 1

Your heap usage looks pretty typical. In scenario 5, I think you are simply running out of memory. I doubt its a leak inside the javafx runtime (although it could be).+
I think you might be right. It's running out of memory and I may have to increase the heap size.
Say that my application legitimately needs more memory to operate, and I need to increase the heap size. How does this work if you have a fully deployed application? Is the heap size information built into the application itself so that when the jvm runs the application, it allocates the appropriate amount of memory? I've increased the heap size from 64mb to 128mb, and I added many many nodes and I still crapped out when I hit the 128mb ceiling. I changed it to 512 and I added a TON of nodes (when you click a node from the VBox, it adds a miniature version of the node to the scene graph) and I'm just under 200MB. I plan on setting a cap to how many concurrent additional nodes can be placed on the scene graph, which will help.
If you deploy this as is, how does the application utilize memory if you've adjusted the heap size? Or is this specific to the IDE only?
Do you know what objects are on the heap? Can you compare what objects are on the heap from one scenario to the next?+
Where can I find this information in NetBeans profiler?
Do you have a lot of images? Are they thumbnails or are they images scaled down to thumbnail size?+
Actually, yes I am using a scaled down thumbnail size of the original image. The original image files are PNG format, 60x60 pixels, and about 8kb in size. I simply use the "FitWidth:" property to scale the image down. I was doing some more reading before I went to bed and I was going to use an alternative way to scale the image down. By simply doing this, the initial heap usage off the 500 node search went down form 44MB to 39MB. It's still slower on consecutive searches versus the first, but it's stable.
Edit: I've used the width: property to downsize the image and it looks like I'm not running into that heap crash as fast but this poses a problem where I need to have the full size of the image available when a custom node is selected. What's the best way of doing this? I should probably store the image location in a string and recreate the image when I need it in full size since there is only one full size version of it on the screen at a given time. I've also completely disabled the addition of a picture in my custom node; it appears these images don't take up a lot of space since they are very small. I save an additional 3-5MB in heap space if I completely disable the pictures and have just the nodes themselves. Each node has animation effects (i.e. fading in/out of colors, changing of color if selected). Although the class itself is pretty dang long in comparison with any other classes I have.
Are you clearing the nodes from your scene content before the search or after? If after, are you creating the new nodes before clearing out the old?+
Yes, I have a function that reassigns the stage.scene.content sequence omitting the custom vbox that houses the list of custom nodes prior to the next search. The "cleanUp()" function is called prior to the insertion of the new custom vbox.
It might be useful to turn on verbose garbage collection (-verbose:gc on the java command line) just to see what's happening in gc.+
What is this exactly? I tried putting in System.gc() but I'm not sure if I'm seeing any difference yet.
Edit: Actually, I've placed System.gc() after I run my cleanUp() function and I'm noticing the heap usage is more conservative. Seems to clear more often than it did before. But yes, the underlying problem of my running out of memory is to be looked at.
You might also (just as an experiment) force garbage collection between your searches.+
This seems to work well with smaller result sets. However, a search that produces over 500 custom nodes in a custom VBox uses more than half of the available heap size, so the next time it tries to do the same search it just crashes like you mentioned in your first point. The memory simply runs out. What I don't get is if I "delete" the custom vbox prior to inserting it, the memory doesn't seem to be released immediately. From what I'm reading about the garbage collector, it doesn't exactly do things in a prompt fashion.