[Gluster-devel] io-cache exceeding cache-size value

[Gordan Bobic]

Dan Parsons wrote:
> I will do this today. I noticed that I already have vm.drop_caches set 
> to 3 via sysctl.conf, based on a suggestion from you from long ago. 
> Should I delete this under normal usage? Is it possible that this 
> setting, enabled by default, is causing my problems?

It isn't a permanent setting, it's just a real-time instruction to drop 
all current caches.

In the current context, I think the whole idea is bogus anyway because 
glusterfs process still maintains it's current resident size at hundreds 
of MB when I flush the caches (with no performance translators), so I 
don't think this affects the leak in any way.

On my setup I have / mounted on glusterfs, /tmp on ext3 and /usr/src on 
NFS, so the fact that glusterfs root daemon bloating can only be caused 
either by access to shared libraries or invocation of executables, since 
compiling a big code tree (including when it doesn't reside on 
glusterfs) seems to trigger the leak in a pretty major way.

I'll try to re-create the problem with a chroot environment, since 
debuging the rootfs daemon is extremely difficult.

Gordan