[Ovirt-devel] ovirt-server patch for vm failover and actual data set to db

Perry Myers pmyers at redhat.com
Tue Jan 27 18:40:50 UTC 2009


Ian Main wrote:
> Sorry I took so long to get back to you on this one.  There are some
> issues with formatting in this patch etc. but mostly I've been thinking
> that we need to have this configurable in the WUI.
> 
> I'm curious what other people think about rebooting crashed hosts?  Is
> this a desired feature?

Crashed hosts or crashed guests?

The problem with rebooting crashed hosts is you need to worry about 
fencing the host that you 'think' crashed (since there is no way to be 
completely sure without a fence of a fancy clustering algorithm that 
depends on shared storage)

Rebooting crashed guests is another story.  You can be sure the guest is 
dead because when it is crashed you can issue a virsh destroy to make sure 
it really is dead before restarting the guest.

But then the question becomes "When is a guest dead?"  That's the tricky 
part IMHO.

But if you do have a reliable way of telling when a guest has died, and 
you fence the guest properly w/ libvirt destroy then I think auto restart 
of guests is a nice feature.  However, this should be configurable.  i.e. 
each guest should have an option "Restart this guest on crash? Y/N"

Perry




More information about the ovirt-devel mailing list