On Tue, Mar 12, 2013 at 05:43:29PM +0100, Guido Trotter wrote: > On Tue, Mar 12, 2013 at 5:30 PM, Iustin Pop <[email protected]> wrote: > > On Tue, Mar 12, 2013 at 05:18:54PM +0100, Guido Trotter wrote: > >> Mirroring no disk is a no-op. As such we can treat them like mirrored > >> instances, since the data they need (none) will be present on all nodes. > >> > >> This is definitely enough to failover or migrate instances with a manual > >> target node. Further work might be needed for support in hail. > > > > I suspect, if it doesn't work _already_ (e.g. node evacuation considers > > diskless as migratable), that it's a trivial fix as well. > > > > In the meantime, LGTM. > > For now I only had a way to test with manual nodes. > I wonder: should we apply this to devel-2.7 as well? (since it's just > a small change, it was requested, and it might help people to use > this, even without iallocator).
Trying to think about any way that this can fail (since 2.7 should be frozen), but I can't think of one. LGTM then! iustin
