> > Nova is also not the right place to do the generic solution as many other > parts could be involved... neutron and cinder come to mind. Nova needs to > provide the basic functions but it needs something outside to make it all > happen transparently. > I would really like a shared solution rather than each deployment doing > their own and facing identical problems. A best of breed solution which can > be incrementally improved as we find problems to get the hypervisor down > event, to force detach of boot volumes, restart elsewhere and reconfigure > floating ips with race conditions is needed. > Some standards for tagging is good but we also need some code :-)
I think this would actually be a worthwhile cross-project effort but getting it done would require some higher-level guidance to keep it on track. I also do not believe Nova *contains* all of the data to perform auto-evac, but it has *access* to the data right? Or could anyway. I think Cinder would definitely play a roll, and Neutron for sure. And as far is scope is concerned, I personally think something like this should only support VM's with shared storage. Otherwise, phase 1 gets overly complex and gets into something akin to VMware's DRS which I DO think could be another step, but the first step needs to be clean to ensure it gets done. *Adam Lawson* AQORN, Inc. 427 North Tatnall Street Ste. 58461 Wilmington, Delaware 19801-2230 Toll-free: (844) 4-AQORN-NOW ext. 101 International: +1 302-387-4660 Direct: +1 916-246-2072 On Tue, Oct 14, 2014 at 12:01 PM, Mathieu Gagné <mga...@iweb.com> wrote: > On 2014-10-14 2:49 PM, Tim Bell wrote: > >> >> Nova is also not the right place to do the generic solution as many other >> parts could be involved... neutron and cinder come to mind. Nova needs to >> provide the basic functions but it needs something outside to make it all >> happen transparently. >> >> I would really like a shared solution rather than each deployment doing >> their own and facing identical problems. A best of breed solution which can >> be incrementally improved as we find problems to dget the hypervisor down >> event, to force detach of boot volumes, restart elsewhere and reconfigure >> floating ips with race conditions is needed. >> >> Some standards for tagging is good but we also need some code :-) >> >> > I agree with Tim. Nova does not have all the required information to make > a proper decision which could imply other OpenStack (and non-OpenStack) > services. Furthermore, evacuating a node might imply fencing which Nova > might not be able to do properly or have the proper tooling. What about > non-shared storage backend in Nova? You can't evacuate those without data > loss. > > -- > Mathieu > > > _______________________________________________ > OpenStack-dev mailing list > OpenStack-dev@lists.openstack.org > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev >
_______________________________________________ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev