KubeVirt — Run Virtual Machines on Kubernetes

Introduction

KubeVirt is a CNCF incubating project that turns every Kubernetes node into a hypervisor host via the Linux kernel KVM stack. Instead of running two platforms — one for containers, one for VMs — operators can manage stateful legacy workloads and modern cloud-native services using the same scheduling, networking and storage primitives.

What KubeVirt Does

• Introduces VirtualMachine, VirtualMachineInstance and VirtualMachineInstancePreset CRDs
• Runs each VM inside a Pod via virt-launcher + libvirt + qemu-kvm
• Integrates with CNI for L2/L3 networking and with CSI for persistent storage
• Enables live migration, snapshots, cloning and HotPlug devices
• Ships companion tools (CDI, HCO, Forklift, MTV) for imports and data management

Architecture Overview

A KubeVirt install includes an operator that manages several DaemonSets/Deployments: virt-api (admission and API), virt-controller (VM lifecycle), virt-handler (per-node agent) and virt-launcher (Pod wrapper for a VM). The handler programs the kernel through libvirt; the launcher runs a single qemu-kvm process per VM. CDI (Containerized Data Importer) manages disk images in PVCs, and the virtctl CLI covers console, VNC and live migration.

Self-Hosting & Configuration

• Install via operator manifests or the Hyperconverged Cluster Operator (HCO) for batteries-included setups
• Node requirements: KVM-enabled CPU, nested virtualisation enabled if running in a cloud VM
• Use MultusCNI for VLAN-aware VM networking
• Use CDI to import qcow2/ISO images from HTTP, S3 or container registries
• Configure Storage Profiles to pick the right CSI snapshot/clone strategy per StorageClass

Key Features

• First-class Kubernetes CRDs for VMs, with familiar kubectl workflows
• Live migration between nodes for maintenance and rebalancing
• HotPlug CPU, memory, disks and NICs
• Integrated with Multus, Calico, Cilium and SR-IOV for advanced networking
• Backed by a rich ecosystem (CDI, KubeSAN, Forklift) for VM migrations from oVirt/VMware

Comparison with Similar Tools

• Proxmox / vSphere — dedicated hypervisors; KubeVirt unifies them with Kubernetes control
• OpenStack Nova — heavyweight VM platform; KubeVirt sits on top of Kubernetes primitives
• Firecracker — microVM focus for serverless; KubeVirt targets general-purpose VMs
• kata-containers — runs pods in VMs; KubeVirt exposes VMs themselves
• virt-manager / libvirt alone — great for single hosts, KubeVirt clusters them

FAQ

Q: Can I run Windows workloads? A: Yes. KubeVirt includes drivers and documentation for Windows guests, including VirtIO drivers and device passthrough.

Q: Does live migration work without shared storage? A: Requires a ReadWriteMany volume or block-level replication for most storage classes; local storage can use storage-live-migration with CDI.

Q: How much overhead does virt-launcher add? A: A few hundred MB per VM plus the qemu-kvm footprint. KVM itself runs near bare-metal speeds.

Q: Is it production ready? A: Yes, it is used in production by Red Hat OpenShift Virtualization and many Kubernetes platforms.

Sources

• https://github.com/kubevirt/kubevirt
• https://kubevirt.io