worker_cpu_affinity

[Igor Sysoev]

On Wed, Nov 28, 2007 at 02:47:05PM -0800, Cliff Wells wrote:

> On Wed, 2007-11-28 at 07:21 +0300, Igor Sysoev wrote:
> > On Tue, Nov 27, 2007 at 01:30:53PM -0800, Cliff Wells wrote:
> > 
> > > I'm running a dual processor (Athlon MP) server.  Gentoo Linux.  Nginx
> > > is 0.6.16.
> > > 
> > > In nginx.com I have:
> > > 
> > > worker_processes        2;
> > > worker_cpu_affinity     0001 0010;
> > > 
> > > 
> > > But I see in my error log the following:
> 
> > Try the attached patch.
> > 
> 
> Same result:
> 
> 2007/11/28 14:37:14 [notice] 2175#0: OS: Linux 2.6.18-028test007
> 2007/11/28 14:37:14 [notice] 2175#0: sysctl(KERN_RTSIGMAX): 0
> 2007/11/28 14:37:14 [notice] 2175#0: getrlimit(RLIMIT_NOFILE): 1024:1024
> 2007/11/28 14:37:14 [notice] 2176#0: start worker processes
> 2007/11/28 14:37:14 [notice] 2177#0: sched_setaffinity(0x00000001)
> 2007/11/28 14:37:14 [alert] 2177#0: sched_setaffinity(0x00000001) failed (22: Invalid argument)
> 2007/11/28 14:37:14 [notice] 2176#0: start worker process 2177
> 2007/11/28 14:37:14 [notice] 2179#0: sched_setaffinity(0x00000002)
> 2007/11/28 14:37:14 [alert] 2179#0: sched_setaffinity(0x00000002) failed (22: Invalid argument)
> 2007/11/28 14:37:14 [notice] 2176#0: start worker process 2179
> 
> Perhaps worth noting is that I'm running a kernel with OpenVZ patches
> (although Nginx is not running in a VPS). 

Well, I will return to sched_setaffinity after cache will be done.
Current sched_setaffinity() processing needs to be rewritten:
1) it does not support RedHat sched_setaffinity() interface,
2) it should use CPU_SET()/etc interface for other Linuxes.


-- 
Igor Sysoev
http://sysoev.ru/en/