[libvirt] [PATCH] Raise default limit on number of processes for qemu user

[Daniel P. Berrange]

On Thu, Mar 10, 2011 at 11:33:24AM +0000, Daniel P. Berrange wrote:
> On Wed, Mar 09, 2011 at 02:20:09PM +0100, Jiri Denemark wrote:
> > The daemon/libvirtd.limits file (which is supposed to be copied to
> > /etc/security/limits.d/libvirtd.conf) is generated based on --qemu-user
> > option passed at configure time.
> > 
> > The file is intentionally not installed by make install since installing
> > it on distributions with higher or no limit on number of process could
> > actually result in lowering the limit. Packagers may choose whether to
> > install the file or not. It is installed by libvirt.spec for RPM based
> > distributions.

[snip]

> Hmm, did you actually test this setup to make sure it works as we
> expect ?  I have this nasty feeling in the back of my mind that
> the files under /etc/security/limits.d/ are only processed by
> PAM modules. Since PAM isn't at all involved when libvirt changes
> UID to 'qemu' to launch QEMU, how does QEMU actually see the increased
> limit being set ?

Well I've tested this now and confirm that putting a file into
/etc/security/limits.d for the 'qemu' user, has absolutely no
effect on QEMU as launched by libvirtd.

> Something needs to be calling the setrlimit() systemcall for the
> QEMU process when it is launched and I don't see what is yet ?

Because we don't use PAM, QEMU is just inheriting the limits from
libvirtd. For added fun, the limits that libvirtd sees typically
differ depending on whether libvirtd was started from a root login
shell, or from init.  So AFAICT, we have to use  setrlimit() if we
want to control this.

Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|