Description of problem: In some environments VMs are pinned to specific hosts. These VMs cannot be set as highly-available because this is incompatible with pinning. The problem is: when the host crashes or is fenced one way or the other, these VMs do not start when the host is ready. Version-Release number of selected component (if applicable): This is seen on RHEV 3.5. How reproducible: Always Steps to Reproduce: 0. Configure Power Management on the host 1. Pin a VM to a host ("start on specific host") 2. Reboot or fence the host externally Actual results: The host comes up again "green" in RHEV / oVirt Expected results: The host comes up and the VM comes up too.
The VM must be down when the host is not available - and never restarted anywhere else than its dedicated host.
Michal, what is your take on it. I believe this is sort of autostart feature for pinned VMs. So the request is: 1. Select a host where the VM has to be started 2. Select "autostart option" (or however we implement it) 3. If the host is available, ensure the VM is started/running. (4. If the host crashes etc. don't bother about this VM) Thanks! Martin
Since you can select today a group of hosts for affinity, not sure this is still an issue.
Martin, there's a difference between "HA" VMs and "autostart". HA restarts VMs only when host or VM crashes, autostart assumes it tries to start a VM whenever it is down even if it was e.g. shutdown from within the guest cleanly. The original request seems to be about "HA", so as comment #7 says we may not really need anything anymore
Hi Michal, as I understood this, this is not about HA, but just about autostart. So if the host affinity does work this way, I believe I am fine. As I understood the requested workflow: 1. Pin a VM to a host (Start on host xy) 2. Enable "autostart" 3. VM is automatically started if host xy is available In case the host crashes the VM is *not* restarted anywhere else In case the host gets back into "up state" the VM is started automatically on this host. @Francois: Please correct the workflow, if I misunderstood it. Does strict host affinity solve this issue in this way? If so, I believe we can close this with CURRENTRELEASE. Thanks! Martin
@Martin: the workflow as stated is exactly what is needed, thank you. NB: I'm not clearing NEEDMOREINFO because I haven't played with host affinity in current oVirt/RHV versions.
ok, sounds indeed resolved by affinity
@Michal @Martin thank you.
What version is this in? Can you set the target milestone?
affinity is a 3.6 feature I think, so it's everywhere
(In reply to Michal Skrivanek from comment #15) > affinity is a 3.6 feature I think, so it's everywhere With auto-start?