Open Stack

Please see my responses inline, prefixed by [SL].


On May 30, 2014, at 1:55 PM, hossein zabolzadeh <zabolzadeh at gmail.com> wrote:

> Any Idea?
> 
> ---------- Forwarded message ----------
> From: hossein zabolzadeh <zabolzadeh at gmail.com>
> Date: Fri, May 30, 2014 at 12:01 PM
> Subject: If an instance going down!
> To: "openstack at lists.openstack.org" <openstack at lists.openstack.org>
> 
> 
> Hi,
> If an instance running on openStack going down, which infrastructure component is responsible to recover the vm from the failure by doing some mechanism such as live migration?


[SL] If you are in a failure scenario, live migration isn't going to do you any good. Live migration assumes that the source and destination hypervisor are up and running, and transfers the instance between them. If the hypervisor on which the instance is running has failed, then live migration won't help; if the instance itself has failed, then live migration will just move the failed instance to a new host.


> If a host with multiple VMs going down, is there any openStack component to automatically recover all of VMs from the failure?(For example automatically start other host, and migrate all of the instances to the new one).


[SL] If you are proactively wanting to move instances to a new hypervisor in order to take a hypervisor down, then nova evacuate (as suggested by Erik) is what you need. If you are wanting to recover an instance reactively after a hypervisor has already failed, then neither live migration nor nova evacuate will help (AFAIK).

If you are using VMware vSphere with the VCDriver under Nova, then you can leverage vSphere HA, which *will* automatically recover instances after a hypervisor failure. I think that Hyper-V might have similar functionality based on Windows Failover Clustering, but I don't know if that works/is supported under OpenStack. I don't believe that KVM or Xen offer any similar functionality.

(Disclaimer: I work for VMware.)


> Some experts said, this type of disaster recovery must be address at application layer(DR module on application bundled in VM), instead of infrastructure layer?
> What is your idea?


[SL] In general, OpenStack tends to target applications that are capable of providing recovery/resiliency in the application itself. This is why there aren't strong efforts within OpenStack to make the infrastructure reliable/resilient.

I hope this helps.

-- 
Scott