It doesn't make sense that the number of CPUs would make a difference
for OVS userspace (which is all that "restart" affects), because OVS
userspace is not multithreaded.
On Thu, Sep 08, 2011 at 12:12:06PM -0400, Andres E. Moya wrote:
> I ran /etc/init.d/openvswitch restart and it gave me a warning if i wanted to
> restart it, I hit y and it restarted and everything went back to normal.
> So it seems there is a correlation between running xen with
> echo "NR_DOMAIN0_VCPUS=1" > /etc/sysconfig/unplug-vcpus
> and openvswitch not consuming cpu and memory then crashing, it does it a
> slower pace when run with "NR_DOMAIN0_VCPUS=1 instead of "NR_DOMAIN0_VCPUS=4
> does anyone know if these issues are being addressed on the release that is
> supposed to be due out tomorrow?
> ----- Original Message -----
> From: "Ben Pfaff" <blp@xxxxxxxxxx>
> To: "Andres E. Moya" <amoya@xxxxxxxxxxxxxxxxx>
> Cc: public-xen-api-GuqFBffKawuULHF6PoxzQEEOCMrvLtNR@xxxxxxxxxxxxxxx, "Andres
> E. Moya" <public-amoya-6IoTNRIFf+UAspv4Qr0y0gC/G2K4zDHf@xxxxxxxxxxxxxxx>
> Sent: Thursday, September 8, 2011 10:06:52 AM
> Subject: Re: ovs-vswitchd high cpu and memory use eventually crashes
> If OVS is running with --monitor (as it does by default on XenServer),
> then you can restart it quickly with just "kill -SEGV $(cat
> /var/run/ovs-vswitchd.conf)". Yes, it will just pause connectivity
> On Thu, Sep 08, 2011 at 06:14:47AM -0400, Andres E. Moya wrote:
> > is there anyway to restart the ovs-vswitchd process, will this jsut
> > kill connectivity while it is restarting?
> > ----- Original Message -----
> > From: "Ben Pfaff" <blp@xxxxxxxxxx>
> > To: "Andres E. Moya"
> > <public-amoya-6IoTNRIFf+UAspv4Qr0y0gC/G2K4zDHf@xxxxxxxxxxxxxxx>
> > Cc: public-xen-api-GuqFBffKawuULHF6PoxzQEEOCMrvLtNR@xxxxxxxxxxxxxxx
> > Sent: Wednesday, September 7, 2011 10:11:39 PM
> > Subject: Re: ovs-vswitchd high cpu and memory use eventually crashes
> > "Andres E. Moya"
> > <amoya-6IoTNRIFf+UAspv4Qr0y0gC/G2K4zDHf@xxxxxxxxxxxxxxxx> writes:
> > > we tried the fix
> > > echo "NR_DOMAIN0_VCPUS=1" > /etc/sysconfig/unplug-vcpus
> > >
> > > 24 hours later the same thing happened
> > The symptoms sound like a small per-flow memory leak. When I get
> > a chance, I'll test for that. (Or you can do so yourself with
> > --leak-check and ovs-parse-leaks and send a patch.)
xen-api mailing list