Description of problem:
After a round of longevity test for about one week, we found the process of broker service had about 9.2% memory usage increased in total. And the process didn't release the memory after the test stopped.

Version-Release number of selected component (if applicable):
2.0/2013-11-22.1
openshift-origin-broker-1.15.1-2.el6op.noarch

How reproducible:
always

Steps to Reproduce:
1.Set up an ose-2.0 env, creating and deleting all kinds of apps continuously
for about one week.

Initially, the two broker process cost 5.9% memory.
USER       PID %CPU %MEM    VSZ   RSS    TIME COMMAND
apache    6151  0.3  2.9 467480 116428   4:33 Rack: /var/www/openshift/broker  
apache    6159  1.5  3.0 469856 118884  22:38 Rack: /var/www/openshift/broker
                                             
                                               
Then the two process takes more and more memory along with the test running, which makes the broker machine has little free memory.
The usage of the two process in the end of the test.
USER       PID %CPU %MEM    VSZ   RSS     TIME COMMAND
apache    6151  3.5  7.6 651264 300204  450:17 Rack: /var/www/openshift/broker
apache    6159  2.8  7.5 645884 294824  362:12 Rack: /var/www/openshift/broker
                                                
2.Stop the test for some time, checking the two process, found they didn't release any memory.
USER       PID %CPU %MEM    VSZ   RSS    TIME COMMAND
apache    6151  1.9  7.6 651300 300232 454:28 Rack: /var/www/openshift/broker
apache    6159  1.5  7.5 645884 294356 362:12 Rack: /var/www/openshift/broker


Actual results:

Expected results:
The broker process should not cost more and more memory when it's working. 
Or it should release occupied memory when it's free.

Additional info: