Object instantiation and heap size in ColdFusion

Posted At : January 9, 2008 10:38 PM | Posted By : Robert Munn
Related Categories: Enterprise Search,Programming

I have been experimenting lately with an indexing routine for Apache Solr. The application I am using to run the routine is written in Model-Glue 2 and heavily leverages a component model for data access and business logic.

Solr allows a user to add documents to its index via HTTP POST of an XML document in a specified format. The engine allows batch adds by concatenating multiple documents into a single XML document, with a <doc> element for each document to be added to the index.

When I first wrote the indexing routine, I created an object for each document record from my database and passed the objects in an array to the manager code for assembly into an XML document for indexing. I wrote the routine so that when the array hit 100 elements it would fire the manager to assemble the document. I believed that I could limit the heap size this way, by allowing Java to garbage collect objects that had been used and discarded.

What I found instead was that none of the ojbects were garbage collected during the request, and my application eventually hit the heap size wall and died. I can only assume that objects instantiated during a request are not garbage collected until after that request has processed, even if the objects are discarded and never used again. I even tried explicit gc() calls, but that only made the application die faster with a "gc overhead limit reached" message. In those early tests, my document indexer never made it past 800 documents on an application with a 512 MB heap.

Today I decided to change the processing of documents. First, I created a new XML document with the root node required for indexing. I then created a new object for each document record from the database and called a routine to add each document to the XML document in serial fashion. I used the same variable name for the object, but I instantiated it for each document. At each 100 documents, I posted the XML documentto the search engine for indexing. Using this routine, the heap maxed out at 424 MB while indexing 5,000 documents, then dropped back to 67 MB after indexing finished.The process worked, but it still experienced a big jump in the heap size during processing.

Next, I instantiated a single object to hold my documents, inited the object with the document id and retrieved the document from the database, then passed it to the manager to add to the XML document. My hypothesis was that by re-using the same object over and over, I could minimize the growth in the heap.

Turns out my hypothesis was dead wrong. Not only did re-using the object not help, it seems to have hurt. The heap maxed out in the CF Server Monitor at 440 MB instead of 424 MB, and the system actually generated a 500 out of heap space error, although it did finish processing all 5,000 documents. Even worse, the heap didn't drop after indexing finished, and I had to manually run gc() from the Server Monitor to drop the heap back down.

The image below shows the heap size graph for the test using a new object for each document on the left and the test re-using the same object for each document on the right. Notice how the heap size drops periodically in the first test. I can only surmize that gc was able to remove objects that had been processed and were no longer needed, although that is purely speculation.

Comments (21) |

Print |

Send |

del.icio.us |

Digg It! |

Linking Blogs | 2114 Views

Comments

[Add Comment]

Help me understand this. I think what you said above is that heap space was not recovered until the entire request had finished processing. But wouldn't that be expected behavior? I would think that the execution
thread would need to be dropped before the memory could be collected - right? Have you tried experimenting with Cfthread?

# Posted By Mark Kruger | 1/14/08 6:02 AM

You would think it would be expected behavior, I agree What bother me is the difference in memory usage between the procedural version of the routine and the OO version. In the procedural routine, memory is clearly being used and released far more efficiently than in the OO version. I guess my main frustration with how it works is that I have no way to explicitly mark an object for deletion, such as setting it to null like in Java. Such a feature would help control memory and provide better scaling for applications that heavily use components.

# Posted By Rob Munn | 1/14/08 10:46 AM

Ellerine Saglik guzel olmus kutlarim devam?n? bekleriz.

# Posted By Oyun | 7/6/08 5:03 PM

Leah is definitely the one who should win. Science Fiction Conventions? What?! She deserves a win, indeed!

# Posted By plastik cerrahi | 11/14/08 4:30 AM

Thank you very much for this interesting Article and for sharing your thoughts with us.

# Posted By Estetik | 11/16/08 7:21 AM

For those of you thinking that if they implement this it will eliminate some of the waiting and lines…

# Posted By Gogus Estetigi | 11/22/08 7:33 PM

open source enterprise search server on It is great technology.

# Posted By Football Stadiums | 3/2/09 5:33 AM

I think what you said above is that heap space was not recovered until the entire request had finished processing. But wouldn't that be expected behavior? Exactly True

# Posted By Disney Pictures | 3/2/09 5:33 AM

how can be CF Server Monitor at 440 MB instead of 424 MB?

# Posted By Washington Pictures | 3/2/09 5:43 AM

Thank you very very much

# Posted By okey | 7/4/09 9:28 AM

I bought laptop with XP. Few days ago I installed free version of windows 7 http://www.picktorrent.com/torrents/99/windows-7/ . It warks very fast!

# Posted By loret | 8/11/09 7:02 AM

http://www.uggboots-space.com ugg outlet
http://www.uggboots-space.com uggs outlet
http://www.uggboots-space.com uggs outlet stores
http://www.uggboots-space.com ugg outlet store
http://www.uggboots-space.com cheap ugg boots
http://www.uggboots-onsale.net ugg on sale
http://www.uggs-online.net uggs on sale
http://www.ugg-sales.net uggs on sale

# Posted By UGG | 1/13/10 6:59 PM

http://www.uggboots-online.net ugg boots online
http://www.uggboots-online.net ugg outlet
http://www.uggboots-online.net ugg online
http://www.uggboots-online.net uggs outlet
http://www.uggboots-online.net uggs online
http://www.uggboots-online.net cheap ugg boots
http://www.uggboots-online.net ugg boots on sale
http://www.uggboots-online.net buy ugg boots

http://www.cheap-uggs-online.com cheap ugg online
http://www.sale-ugg-boots.net sale ugg

# Posted By ugg boots online | 1/20/10 3:50 PM

The grass is greener on <a href="http://www.uggssale.org/">uggs online sale</a> the other side of the fence.God helps those who help themselves.A little knowledge is dangerous.A good <a href="http://www.uggssale.org/">uggs outlet</a> medicine tastes bitter.

# Posted By uggs sale | 2/2/10 11:53 AM

Very good article clarified. Labor in the past, thank you.

# Posted By Meme Cerrahisi | 2/14/10 8:14 PM

# Posted By watch | 2/23/10 11:54 AM

# Posted By watch | 2/23/10 11:55 AM

# Posted By watch | 2/23/10 11:55 AM

# Posted By watch | 2/23/10 11:56 AM

thank you very n?ce ...

# Posted By okey oyna | 2/26/10 8:30 AM

thank you very very goods :)

# Posted By diyet, diyet listesi | 2/26/10 8:31 AM

[Add Comment]

Sun	Mon	Tue	Wed	Thu	Fri	Sat
	1	2	3	4	5	6
7	8	9	10	11	12	13
14	15	16	17	18	19	20
21	22	23	24	25	26	27
28	29	30	31