vCenter as physical or virtual?

[Aaron270]

I currently have vCenter running as a VM in a UCS Mini chasis.  Last night the host that had vCenter on it failed and the VMs didn't fail over to a good server as expected.  I'm assuming this is because the host that failed was the one also running vCenter, but if that is expected then wouldn't it be better to have vCenter running on a physical server outside of the VMware infrastructure? 

The setup at work is relatively new so I'm a bit worried as I'm already dealing with this issue.  Thankfully there was only one other server on the host that failed and it isn't in production, but the whole lack of failover has me thinking I should repurpose an old physical box for vCenter.

[cjscol]

You can have the vCenter Server running as a virtual machine and it can be automatically restarted on another host if the host it is on fails, along with the other VMs on the failed host being restarted.

Are the ESXi hosts configured in a cluster with HA enabled?

[Aaron270]

Strange, as far as I can tell HA is enabled but neither servers on this host failed over.  I'll have to find some time to test again and see if the servers fail over correctly.

[rcporto]

All virtual machines are stored on shared storage ? Both hosts have the same network configuration ?

[Aaron270]

Yes both are using storage on an iSCSI datastore, (Nimble), and they have the same networking properties. 

[cjscol]

Can you post a screenshot of the HA settings on your cluster?

[Aaron270]

Sorry for the delay in getting back, are these the settings you were asking about?

I did notice that heartbeating is used against the datastores, but none were selected from the list.  I also did some testing by restarting the same host that failed with VMs running, including vSphere.  However in all of those tests HA worked as expected and the VMs restarted under a different host.

Initially when the host originally failed after it restarted it came to this pink screen of death as seen from the UCS console window:

I'm not sure if that helps at all, but I can't figure out why HA didn't work when the host crashed, but works just fine while testing by restarting the host.