
On Wed, May 23, 2018 at 6:06 AM, Eitan Raviv <eraviv@redhat.com> wrote:
If the engine manages to 'reboot' the host as you say, it seems that the engine can communicate with the host.
So it might be that you have a bit of a chicken and egg problem: the host is missing a required network and therefore goes back into non-responsive. If you have other (non mgmt) 'required' networks on the host, set the networks to non-required under Compute | Clusters | <cluster> | Logical Networks | Manager Networks temporarily.
In Compute | Hosts | <host> | networks | setup networks can you attach 'ovirtmgmt' to the host in its current condition?
Try to reboot the host.
HTH
The configuration has only 2 logical networks: the ovirtmgmt and another one that is already marked as not required I just noticed that I configured wrong my first address of dns servers for this host. This could be a source of timeout/problems So I put the correct value for it. NOTE: from cockpit web interface it seems it didn't maintain the config, it continued to reverted with the wrong first entry of DNS servers; I had to connect to host via ssh and execute nmcli con mod "Bond connection 1" ipv4.dns "172.16.1.2,172.16.1.20" nmcli con down "Bond connection 1"; nmcli con up "Bond connection 1" and the resolv.conf configuration survived also a reboot of the node. Then at this point I was able to put host into maintenance without werrors, but not to activate it (acceptable because id doesn't have ovirtmgmt configured yet), or better a reinstall, because I got now an "invalid fingerprint" event for the node. Indeed the ng-node os was re-installed from scratch with the same parameters but I had not a backup of the ssh keys generated during the first install. But anyway at least now I could remove the host (the "Remove" button is not grayed out as before) and install as a new host with the same name/ip without problems. It remains the question if a node fails and then remains inaccessible, if I'm not wrong the "Remove" button is not a possible choice. I can try to reproduce and verify. Gianluca