I do use  automatic migration policy.

The main question you have to solve is:
1. Why the nodes became 'Non-operational' .Usually this happens when the management interface (in your case HoatedEngine VM) could not reach the nodes over the management network.

By default, management is going over the ovirtmgmt network. I guess you have created the new network. Marked that new network as management network and then the switch was off , causing 'Non-Operational  state'.

2. Migrating VMs is usually a safe approach, but this behavior is quite strange. If a node ia Non-operational ->  there could be no successful migration.

3. Some of the VMs got paused  due to storage issue.  Are you using GlusterFS, NFS or iSCSI ? If yes, you need to clarify why you lost your storage.

I guess  for now you can mark each VM to be migrated only manually (VM -> Edit) and if they are critical VMs , set a high Availability from each VM's Edit options.

In such case, if a node fails, the VMs will be restarted on another node.

4.  Have you setup node fencing ? For example APC, iLo, iDRAC and other fencing mechanisms can allow the HostedEngine use another Host as fencing proxy and to reset the problematic Hypervisor.

P.S.: You can define the following alias  in '~/.bashrc' :
alias virsh='virsh -c qemu:///system?authfile=/etc/ovirt-hosted-engine/virsh_auth.conf'
Then you can verify your VMs even when a HostedEngine is down:
'virsh list --all'

Best Regards,
Strahil Nikolov

On Dec 27, 2019 08:40, zhouhao@vip.friendtimes.net wrote:

I had a crash yesterday in my ovirt cluster, which is made up of 3 nodes.

I just tried to add a new network, but the whole cluster crashed

I added a new network to my cluster, but while I was debugging the newswitch, when the switch was poweroff, the node detected the network card status down and then moved to Non-Operational state.



At this time  all of 3 nodes moved to Non-Operational state.

All virtual machines have started automatic migration,When I received the alert email, all virtual machines were suspended





In 15 minutes my newswitch were power up again.The 3 ovirt-nodes become active again, but many virtual machines become unresponsive or suspended due to forced migration, and only a few virtual machines are pulled up again due to cancelled migration

After I tried to terminate the migration tasks and restart ovirt-engine  service, I was still unable to restore most of the virtual machines, so I had to restart 3 ovirt-nodes to restore my virtual machine

I didn't recover all the virtual machines until an hour later


Then  I modify my migration policy  to " Do Not Migrate Virtual Machines"

Which migration Policy do you recommend?

I'm afraid to use cluster...


zhouhao@vip.friendtimes.net