Question about "live migration" or "placement"

Product Support
1

Hello,

My OpenNebula scenario is:

  • Frontend server that acts, also, as KVM hypervisor
  • Two nodes that act as KVM hypervisor
  • Datastore “default” (IMAGE) with TM_MAD=qcow2 to allow creation of VM from “base” image in thin mode, that is to say, if an image has a size of 20 GB in disk, the VM has a size of only the ocuppied filesystem. When I ran some tests with TM_MAD=ssh, I noticed that VM had a size of the 20 GB. Also, this datastore has attribute TM_MAD_SYSTEM=ssh.
  • Datastore “system” (SYSTEM) with TM_MAD=qcow2 and SHARED=yes.
  • Datastore “default” (IMAGE) is shared using NFS in both KVM nodes so, in that nodes, they see “locally” base images
  • However, datastore “system” (SYSTEM) is really local to each KVM hypervisor, so each VM is created locally in one of the three KVM hypervisors (assigned in “template”), so the placement is fixed in the “template” and the assigned policy is “HYPERVISOR=kvm & ID=“XX”” (the number of the corresponding hypervisor)

When I instantiate a VM in hypervisor #1, it has assigned by the template the hypervisor where it will be create and deployed. However, if I stop the machine and try to rerun in one of the others hypervisors (for example hypervisor #2), I get this error:

Tue Oct 13 12:54:01 2026 [Z0][VM][I]: New LCM state is PROLOG_UNDEPLOY
Tue Oct 13 12:54:01 2026 [Z0][VM][I]: New LCM state is BOOT_UNDEPLOY
Tue Oct 13 12:54:01 2026 [Z0][VMM][I]: Generating deployment file: /var/lib/one/vms/8826/deployment.5
Tue Oct 13 12:54:01 2026 [Z0][VMM][I]: Successfully execute transfer manager driver operation: tm_context.
Tue Oct 13 12:54:02 2026 [Z0][VMM][I]: ExitCode: 0
Tue Oct 13 12:54:02 2026 [Z0][VMM][I]: Successfully execute network driver operation: pre.
Tue Oct 13 12:54:02 2026 [Z0][VMM][I]: ExitCode: 0
Tue Oct 13 12:54:02 2026 [Z0][VMM][I]: Successfully execute virtualization driver operation: /bin/mkdir -p.
Tue Oct 13 12:54:02 2026 [Z0][VMM][I]: ExitCode: 0
Tue Oct 13 12:54:02 2026 [Z0][VMM][I]: Successfully execute virtualization driver operation: /bin/cat - >/var/lib/one//datastores/0/8826/vm.xml.
Tue Oct 13 12:54:02 2026 [Z0][VMM][I]: ExitCode: 0
Tue Oct 13 12:54:02 2026 [Z0][VMM][I]: Successfully execute virtualization driver operation: /bin/cat - >/var/lib/one//datastores/0/8826/ds.xml.
Tue Oct 13 12:54:02 2026 [Z0][VMM][I]: Command execution fail (exit code: 255): cat << 'EOT' | /var/tmp/one/vmm/kvm/deploy '/var/lib/one//datastores/0/8826/deployment.5' 'host-2' 8826 host-2
Tue Oct 13 12:54:02 2026 [Z0][VMM][I]: error: Failed to create domain from /var/lib/one//datastores/0/8826/deployment.5
Tue Oct 13 12:54:02 2026 [Z0][VMM][I]: error: Cannot access storage file '/var/lib/one//datastores/0/8826/disk.0' (as uid:9869, gid:9869): No such file or directory
Tue Oct 13 12:54:02 2026 [Z0][VMM][I]: Could not create domain from /var/lib/one//datastores/0/8826/deployment.5
Tue Oct 13 12:54:02 2026 [Z0][VMM][I]: ExitCode: 255
Tue Oct 13 12:54:03 2026 [Z0][VMM][I]: ExitCode: 0
Tue Oct 13 12:54:03 2026 [Z0][VMM][I]: Successfully execute network driver operation: clean.
Tue Oct 13 12:54:03 2026 [Z0][VMM][I]: Failed to execute virtualization driver operation: deploy.
Tue Oct 13 12:54:03 2026 [Z0][VMM][E]: DEPLOY: error: Failed to create domain from /var/lib/one//datastores/0/8826/deployment.5error: Cannot access storage file '/var/lib/one//datastores/0/8826/disk.0' (as uid:9869, gid:9869): No such file or directoryCould not create domain from /var/lib/one//datastores/0/8826/deployment.5ExitCode: 255
Tue Oct 13 12:54:03 2026 [Z0][VM][I]: New LCM state is BOOT_UNDEPLOY_FAILURE

I suppose I have something bad configured… but, really, excluding this feature of “live migrate” or “replacement in other hypervisor”, my scenario is working FINE for more than 6 or 7 years…

Could someone help me and explain how could have “live migrate” (or “deploy” in other KVM hypervisor) with my configuration?

Thanks in advance!!!

2

Hello,

The error is related to the System Datastore configuration rather than to the VM template or scheduler placement.

The important point is that your System Datastore is configured as:

TM_MAD = qcow2
SHARED = yes

while the actual /var/lib/one/datastores/<SYSTEM_DS_ID> storage is local and independent on each KVM hypervisor.

If both the Image and System Datastores are on shared NFS storage, they can use the qcow2 transfer mode. Both datastore paths must then point to the same shared storage from every KVM node. This is the simplest setup for live migration because the VM disks are accessible from both the source and destination hypervisors.

Alternatively, if you want to keep the Image Datastore on shared NFS but run the VMs from local disks on each HV, both Datastores should use the local-storage transfer mode (local on current OpenNebula versions, or ssh on older installations), rather than using qcow2 .