Migrating VMs from VMware

Export a VM from vSphere as an OVA, put it on any http(s) server the nodes can reach, and create a Machine from it with one command. kairon-node downloads it, FluxVM converts the VMDK to raw and repairs the guest so it boots on virtio, and the Machine starts.

# vSphere: export the powered-off VM (ovftool, govc, or the UI)
govc export.ovf -vm web01 ./exports
tar -C exports/web01 -cf web01.ova web01.ovf web01-disk1.vmdk
# publish it (any static file server works)
aws s3 cp web01.ova s3://exports/ && URL=$(aws s3 presign s3://exports/web01.ova --expires-in 86400)

kaironctl import ova "$URL" --name web01 --network tap
kaironctl get machines

kaironctl import ova streams the OVA once without saving it, hashes it for spec.image.digest and reads vCPUs and memory from the OVF. It creates:

apiVersion: kairon.zyvor.dev/v1alpha1
kind: Machine
metadata:
  name: web01
  labels: {kairon.zyvor.dev/imported-from: ova}
spec:
  image:
    source: {httpURL: "https://...", format: ova, repair: true}
    digest: sha256:...
  resources: {cpu: "4", memory: 8192Mi}
  runtime: {backend: qemu}
  network: {mode: tap}
  powerState: Running

For a local file pass the URL it will be served from: kaironctl import ova ./web01.ova --url https://files.example.com/web01.ova. --dry-run prints the Machine instead of creating it. --cpu, --memory and the other kaironctl create flags override the OVF. With --sha256, --cpu and --memory all set, kaironctl doesn't read the OVA at all.

What the repair changes

Area	Change
VMware tools	vmtoolsd, open-vm-tools, vmware-tools, vgauth units disabled
Boot	virtio drivers added to dracut or initramfs-tools; initramfs rebuilt for the newest kernel
Disks	/dev/sdX and /dev/hdX in /etc/fstab, /etc/default/grub and grub.cfg become /dev/vdX
Network	70-persistent-net.rules removed; DHCP fallback for en* (netplan or NetworkManager)

The repair report (actions and warnings) is in the kairon-node log for the Machine and in <imageCacheDir>/imported/*.json on the node.

Tested on the lab: a stream-optimized VMDK OVA of Ubuntu 22.04 imports in about a minute and boots on virtio-blk and virtio-net to a login prompt.

Checklist

1. Static IPs. VMware NIC names (ens192, ens160) don't exist on virtio. The repair warns about configs that use them and adds a DHCP fallback. Re-apply a static address with cloud-init or the guest's own tooling, or use a MachineNetworkPolicy and DHCP.
2. UEFI guests. If the OVF says firmware=efi, kaironctl prints a note. Run the Machine on a node whose FluxVM config sets qemu_ovmf_code (use a node selector or label).
3. Multiple disks. Only the boot disk is attached. The other converted disks are on the node under FluxVM's images/imported/ directory; copy them to PVCs and add them as spec.volumes.
4. Windows. Install the virtio-win drivers (viostor, vioscsi, netkvm) in the VM before exporting. Offline driver injection is not done.
5. Node cache. Set the chart's node.imageCacheDir (off by default). The chart mounts that host directory into kairon-node at the same path, which is what FluxVM needs to read the downloaded OVA.

Compared with KubeVirt

KubeVirt's import path is CDI (a DataVolume that converts the disk) plus virt-v2v or the Forklift/MTV operator for the guest repair, each with its own CRDs and pods. In Kairon the download, conversion and repair are part of the Machine's first reconcile on the node, driven by two fields on spec.image.source.