Open Stack

I write some simple code to try using the compute.rpc_api, seems ceilometer can get the diagnostics information successfully , with one thing tricky of passing a instance object to rpcapi, since the rpc_api will have no database access now.

I think if we use this method, we don't need provide pollsters to different virt driver, and reduce the dependency to nova internal (only 2~3 functions needed), also rpc_api is well maintained/versioned.

Any idea, or I missed anything?

Thanks
--jyh

> -----Original Message-----
> From: Jiang, Yunhong [mailto:yunhong.jiang at intel.com]
> Sent: Sunday, November 11, 2012 11:35 PM
> To: OpenStack Development Mailing List
> Subject: Re: [openstack-dev] [ceilometer] importutils problem in
> import_object_ns
> 
> 	I noticed a bug reported
> (https://bugs.launchpad.net/ceilometer/+bug/1074488 ) and a fix commited,
> thanks for the great effort. Are there any continous effort on this, as it should
> break compute pollster also, since libvirt driver also requires virtapi parameter.
> 
> 	This raise a curios question to me, sorry if it's too naive, I must have
> missed something  :$
> 	Currently ceilometer's compute pollster invoke functions from nova (like
> conn._conn.lookupByName()) to get compute information, which is possibly
> some nova internal implementation. However, the nova already has the
> compute api/rpcapi which provide some information already. For example,
> libvirt has already provided the get_diagnostics() to present cpu_time/volume
> access information etc.
> 	Why can't we simply import the compute api and then query through that
> (of course, we can enhance the api implementation if needed)? Is it because of
> performance issue, or it's not workable because of rpc limitation?
> 
> Thanks
> --jyh
> 
> 
> 
> > -----Original Message-----
> > From: John Tran [mailto:njhtran at gmail.com]
> > Sent: Tuesday, November 06, 2012 3:14 AM
> > To: OpenStack Development Mailing List
> > Subject: Re: [openstack-dev] [ceilometer] importutils problem in
> > import_object_ns
> >
> > I was informed that it's not a bug on nova's side as they do not
> > recommend using the libvirt fake driver outside of nova's tests.
> >
> > On Sat, Nov 3, 2012 at 1:48 AM, Julien Danjou <julien at danjou.info> wrote:
> > > On Fri, Nov 02 2012, John Tran wrote:
> > >
> > >> I figured it out, the problem is nova released a new
> > >> nova.virt.FakeDriver its __init__ method requires an arg 'virtapi',
> > >> when I think it should be a kwarg virtapi=None by default.
> > >
> > > Is there a bug/review on Nova side for that?
> > >
> > > --
> > > Julien Danjou
> > > // Free Software hacker & freelance
> > > // http://julien.danjou.info
> > >
> > > _______________________________________________
> > > OpenStack-dev mailing list
> > > OpenStack-dev at lists.openstack.org
> > > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> > >
> >
> > _______________________________________________
> > OpenStack-dev mailing list
> > OpenStack-dev at lists.openstack.org
> > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> 
> _______________________________________________
> OpenStack-dev mailing list
> OpenStack-dev at lists.openstack.org
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev