Description of problem:
dictionary leak at the time of destroying graph

Version-Release number of selected component (if applicable):


How reproducible:
Always

Steps to Reproduce:
1.Upgrade the setup from 3.3.1 to 3.4 (for brick multiplex)
2.Stop all the volumes through gluster CLI
3.Stop glusterd
4.Take the dump of /var/lib/glusterd/vols to some other location from all the nodes
5.Move all .rpmsave to volfile like below on all the nodes
 for fl in `find /var/lib/glusterd/vols/ -name "*.rpmsave" ! -name "*tcp-fuse.vol.rpmsave"`
 do
 mv $fl ${fl%.*}
 done;

6.Start glusterd
7.start all the volumes  -- for y in $(gluster v list);do gluster v start $y;done;sleep 60;echo
8.Stop glusterd
9.Move the original vol files from dump to  /var/lib/glusterd/vols/ 
10. Start glusterd 
11. Create/start/stop a single volume in the loop, running brick process has got graph failure during getspec and every xlator has one dictionary leak 

Actual results:

Running brick process is showing memory leak
Expected results:
There should not be memory leak for running brick process during volume create-start-stop.

Additional info: