[Gluster-devel] memory leak with 2.5--patch-628

[Sascha Ottolski]

Hi,

I'm wondering if anyone may have a hint how to track down a memory leak in 
glusterfs client. The leak is easily reproduced (at least here :-)) by 
sending a "rsync -avP" or "find -exec head -c1" over the mountpoint, that 
carries millions of files. The leak happens while the rsync scans the 
directory, before transferring any files.

I didn't collect any hard numbers up to now, but I've seen glusterfs proc 
growing to about 3.5 GB after scanning 5 mio. or so files. The size of the 
process never seems to shrink, and I believe that it's not growing while 
tranferring data (I have a bonnie running for a while now, which doesn't seem 
to add to the process size).

I've seen a similar memory consumption in 1.3.0-pre4, 1.3.7 and even with the 
latest tla checkout, the latter accompanied with fuse-1.7.0-glf7, on a debian 
4.0, on amd64 hardware.

It appears as if switching on or off read-ahead/write-behind doesn't make much 
of a difference when it comes to the leaking, but I certainly need to check 
that more thoroughly. Other translators in the picture are afr and unify.

And of course I'm very much wondering if other people can reproduce this 
phenomenon...


Thanks,

Sascha