[Gluster-devel] valgrind logs for glusterfs-3.4 memory leak

Kaleb S. KEITHLEY kkeithle at redhat.com
Wed Oct 15 15:27:37 UTC 2014

As mentioned in the Gluster Community Meeting on irc today, here are the 
glusterfs client side valgrind logs. By 'glusterfs client side' I 
specifically mean the glusterfs fuse bridge daemon on the client.


The basic test is, simply, mount the gluster volume, make a deep 
directory path on the volume, e.g. /mnt/a/b/c/d/e/f/g, do three or five 
`ls -R /mnt`, and unmount.

tmp[35]/glusterfs.fuse.out are the logs from three or five ls -R.

tmp[35]+/glusterfs.fuse.out are the logs as above, but with the addition 
that the directories are populated with a few files.

Notice that, e.g. both tmp[35]/glusterfs.fuse.out show approximately the 
same amount of {definitely,indirectly,possibly} lost memory. I.e. the 
number of `ls -R` invocations did not affect how much memory was leaked.

The same is true for tmp[35]+/glusterfs.fuse.out. I.e. more `ls -R` did 
not affect the amount of memory leaked, _but_ notice that when the 
directories are populated with files that more memory was leaked, across 
the board, than when the directories were empty.

Make sense? Any questions, don't hesitate to ask.




More information about the Gluster-devel mailing list