Open Stack

Xu He Jie <xuhj at linux.vnet.ibm.com> wrote on 04/06/2013 11:17:49 AM:

> Yes, the goal is heartbeat of openstack services.

So, what is missing in the current servicegroup API abstraction? Why 
defining a new one?

> Does 'VM HA' means the physical nodes that running the VMs is down, 
> then nova should startup those VMs on another physical nodes?

Yes.

> What I understand was there are two layers of HA:
> 1.The HA of physical node was done by heartbeat/corosync+pacemaker, 
> as this describe http://docs.openstack.org/trunk/openstack-ha/
> content/ch-intro.html

The problem with this approach is that heartbeat/corosync do not scale to 
more than handful of nodes. There are several recent efforts to develop a 
cloud-scale heartbeat mechanisms -- so it could make sense to have a 
framework that would enable plugging them in, to trigger VM evacuation 
(and for other purposes).

> 2.The HA of service that provided by openstack was done by this 
> common heartbeat driver(to monitoring individual services).
> The VM(compute service) provided by nova. So VM HA should be done by
> this common heartbeat driver, it needn't monitor liveness of 
> physical nodes in heartbeat driver.
> Does make sense?

Ideally, the heartbeat at the various levels should be handled in a more 
holistic way, to enable better decision making (e.g., if the service is 
down but the node isn't -- probably just need to restart the service). Not 
sure whether this should be in Nova or in a separate service, though..

Thanks,
Alex
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20130606/cc8e067c/attachment.html>