FluxVM vs. libvirt/virsh, and the backends
The virsh-to-fluxctl command mapping, and which backend to choose for what.
vs. libvirt/virsh
FluxVM is the Zyvor host-local replacement for libvirt/virsh VM lifecycle and networking. It is not a drop-in for KubeVirt/OpenShift.
| libvirt/virsh | FluxVM | Same job? |
|---|---|---|
virsh define + virsh start | fluxctl create | Yes |
virsh list / virsh dominfo | fluxctl list / fluxctl get | Yes |
virsh suspend / virsh resume | fluxctl pause / fluxctl resume | Yes |
virsh destroy | fluxctl delete | Yes |
virsh reboot | fluxctl restart | Yes |
virsh domrename / metadata | fluxctl rename-vm / fluxctl label | Yes (labels drive -l selectors) |
virsh console | fluxctl serial (websocket over REST too) | Yes |
virsh attach-disk / blockresize / detach-disk | fluxctl disk attach|resize|detach | Yes (QEMU) |
virsh snapshot-create / -list / -delete | fluxctl snapshot / snapshot-list / snapshot-delete | Yes (QEMU internal snapshots) |
virsh backup-begin | fluxctl backup [--all-disks] | Yes (standalone qcow2) |
virt-clone | fluxctl clone-vm | Yes |
virsh event | fluxctl events -f (SSE: /v1/events/stream) | Yes |
virsh -c qemu+ssh://host | fluxctl --server / fluxctl context | Yes (REST, bearer token) |
| XML domain definitions | JSON VM spec (fluxctl create --spec vm.json) | Different format, same purpose |
| No REST API | Full REST API (fluxctl serve) | FluxVM adds this |
No libvirtd and no XML: FluxVM speaks its own REST API and manages TAP/bridge/netns networking directly via netlink.
Backends
| Backend | Best for |
|---|---|
| QEMU/KVM | Broad guest/device compatibility, qcow2 CoW overlays, QMP socket |
| Cloud Hypervisor | Modern cloud workloads on a Rust VMM; direct-kernel or firmware boot |
| Firecracker | MicroVMs from a Linux kernel + raw root filesystem, with the jailer |
FluxVM hypervisor ("backend":"flux-vm") | Agent sandboxes: memory snapshots, /v1/sandboxes, guest HTTP proxy + AutoResume, L7 egress, AutoPause, /console. With fluxvm_engine = "kvm" it is a pure in-tree KVM VMM with no QEMU or Firecracker process: multi-vCPU (verified at 2/4/8), pause/snapshot across all vCPUs, virtio queues on ioeventfd, direct-kernel boot from a raw ext4 — docs/native-kvm-no-qemu.md · docs/agent-sandbox-gaps.md |
procbox ("procbox": {} on /v1/sandboxes) | The lightest tier: a rootless Landlock + seccomp process sandbox on the host kernel (no VM, no root; a weaker boundary than a microVM, off by default) — docs/procbox.md · docs/procbox-backend.md |