[Openstack-operators] minimum scale openstack

Alex Leonhardt aleonhardt.py at gmail.com
Wed Jun 18 22:36:22 UTC 2014


Thanks Clint!

Yeah i'd do the DBs as Active / Passive first and do separate testing when
trying A/A our DBAs are actually testing something that may work (A/A) with
almost anything :)

Rabbit - cool - I dont know Rabbit too well but dont expect much issues
there ...

We are planning to run nova and neutron agent/s on every Hyper Visor to
avoid bottlenecks with "network nodes" and instead every HV would use its
own neutron/network service etc. - do you see any issues with that or have
you had any experience running that config ?

Thanks again!
Alex



On 18 June 2014 22:30, Clint Byrum <clint at fewbar.com> wrote:

> Excerpts from Alex Leonhardt's message of 2014-06-18 13:51:41 -0700:
> > hi guys,
> >
> > am looking for some resource which parts of the system are required to be
> > HA - I naturally would do  :
> >
> > - DB
> > - Rabbit
> > - Controller node/s
> >
> > we dont use glance and swift too much and wont really as I'd use it as a
> > private / internal cloud - what other services would you consider needing
> > HA and if so, is it possible to run Active/Active with the DB and Rabbit
> at
> > least ?
> >
>
> Active/Active DB would need to be done using a sync replication system
> like MySQL cluster or Galera. These both recommend more than 2 nodes.
> Also there are likely issues with the way OpenStack handles conflicting
> commits to the database, so Active/Active may not even be possible
> without a lot of testing.
>
> RabbitMQ can be run in Active/Active w/ mirrored queues.
>
> Both of the above would require some kind of STONITH mechanism to be
> sure that split-brain doesn't corrupt your data.
>
> All of the OpenStack controller services should be O-K except two that
> I know of:
>
> * Ceilometer's collector must be Active/Passive
> * Neutron L3 agent needs to be Active/Passive unless patches have landed
>   to change that
>
> _______________________________________________
> OpenStack-operators mailing list
> OpenStack-operators at lists.openstack.org
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-operators/attachments/20140618/ccfe2587/attachment.html>


More information about the OpenStack-operators mailing list