[libvirt] [PATCHv2 5/6] virNodeGetCPUTime: Implement virsh support

[Minoru Usui]

On Tue, 19 Apr 2011 12:01:34 +0100
"Daniel P. Berrange" <berrange at redhat.com> wrote:

> On Fri, Apr 08, 2011 at 08:36:52PM +0900, Minoru Usui wrote:
> > virNodeGetCPUTime: Implement virsh support
> > 
> > Add nodecputime subcommand to virsh.
> > This subcommand prints below output.
> > 
> > [Linux]
> >   # build/tools/virsh nodecputime
> >   usage:            2.8%
> >     user  :         0.8%
> >     system:         1.9%
> >     idle  :        97.2%
> >     iowait:         0.0%
> > 
> > [can get cpu utilization directly(ESX?)]
> >   # build/tools/virsh nodecputime
> >   usage:            2.8%
> > 
> > Signed-off-by: Minoru Usui <usui at mxm.nes.nec.co.jp>
> > ---
> >  tools/virsh.c   |   96 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
> >  tools/virsh.pod |    4 ++
> >  2 files changed, 100 insertions(+), 0 deletions(-)
> > 
> > diff --git a/tools/virsh.c b/tools/virsh.c
> > index 19e3449..93288ba 100644
> > --- a/tools/virsh.c
> > +++ b/tools/virsh.c
> > @@ -3388,6 +3388,101 @@ cmdNodeinfo(vshControl *ctl, const vshCmd *cmd ATTRIBUTE_UNUSED)
> >  }
> >  
> >  /*
> > + * "nodecputime" command
> > + */
> > +static const vshCmdInfo info_nodecputime[] = {
> > +    {"help", N_("Prints cpu utilizatoin of the node.")},
> > +    {"desc", N_("Returns cpu utilizatoin of the node.(%)")},
> > +    {NULL, NULL}
> > +};
> > +
> > +static int
> > +cmdNodeCpuTime(vshControl *ctl, const vshCmd *cmd ATTRIBUTE_UNUSED)
> > +{
> > +    int i, j;
> > +    int flag_utilization = FALSE;
> > +    virNodeCpuTime stats[VIR_NODE_CPU_TIME_NR];
> > +    int nr_stats;
> > +    struct cpu_time {
> > +        unsigned long long user;
> > +        unsigned long long sys;
> > +        unsigned long long idle;
> > +        unsigned long long iowait;
> > +        unsigned long long util;
> > +    } cpu_time[2];
> > +    double user_time, sys_time, idle_time, iowait_time, total_time;
> > +    double usage;
> > +
> > +    if (!vshConnectionUsability(ctl, ctl->conn))
> > +        return FALSE;
> > +
> > +    memset(cpu_time, 0, sizeof(struct cpu_time) * 2);
> > +
> > +    for (i = 0; i < 2; i++) {
> > +        memset(stats, 0, sizeof(virNodeCpuTime) * VIR_NODE_CPU_TIME_NR);
> > +        nr_stats = virNodeGetCpuTime(ctl->conn, &stats[0],
> > +                                     VIR_NODE_CPU_TIME_NR, 0);
> > +        if (nr_stats < 0) {
> > +            vshError(ctl, "%s", _("failed to get cpu time of the node."));
> > +            return FALSE;
> > +        }
> > +
> > +        for (j = 0; j < nr_stats; j++) {
> > +            switch (stats[j].tag) {
> > +            case VIR_NODE_CPU_TIME_KERNEL:
> > +                cpu_time[i].sys = stats[j].val;
> > +                break;
> > +            case VIR_NODE_CPU_TIME_USER:
> > +                cpu_time[i].user = stats[j].val;
> > +                break;
> > +            case VIR_NODE_CPU_TIME_IDLE:
> > +                cpu_time[i].idle = stats[j].val;
> > +                break;
> > +            case VIR_NODE_CPU_TIME_IOWAIT:
> > +                cpu_time[i].iowait = stats[j].val;
> > +                break;
> > +            case VIR_NODE_CPU_TIME_UTILIZATION:
> > +                flag_utilization = TRUE;
> > +                cpu_time[i].util = stats[j].val;
> > +                break;
> > +            case VIR_NODE_CPU_TIME_NR:
> > +            default:
> > +                break;
> > +            }
> > +        }
> > +
> > +        sleep(1);
> > +    }
> 
> I'm not sure about this bit. In other places in virsh, we just report
> the absolute CPU time value, and don't try to calculate the deltas.
> There is a virt-top program that might like to use this data to provide
> a continuous display of utilization.

Matthias said, ESX cannot get absolute CPU time.
It just get CPU percentage.

  http://www.mail-archive.com/libvir-list@redhat.com/msg35769.html

And I think CPU percentage is very useful for user.
The user don't care about absolute CPU time value.
They want to know about CPU load of each host.

So I wrote this specification.
 
> Alternatively, we could add an arg for this to request that it calculate
> delta over 'n' seconds.

I agree. It's very userful. 
But ESX cannot change interval, perhaps.

> 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 :|


-- 
Minoru Usui <usui at mxm.nes.nec.co.jp>