Building off of Fabiano's work in https://github.com/kata-containers/kata-containers/pull/13219, we add docs explain basic Kata concepts and how to get oriented with the project. Co-authored-by: Fabiano Fidêncio <ffidencio@nvidia.com> Signed-off-by: LandonTClipp <lclipp@coreweave.com>
6.4 KiB
Kata Containers Quick Start Guide
New to Kata Containers? This guide gives you just enough context and terminology to understand the project, then points you at the fastest way to try it.
For full installation steps see the installation guide; for deeper background see the overview and the architecture documentation.
What is Kata Containers?
Kata Containers is an open source runtime that runs each container (or
Kubernetes pod) inside its own lightweight virtual machine. Unlike runc,
where containers share the host kernel and are isolated only by namespaces,
cgroups, and seccomp, each Kata pod gets its own guest kernel — a second layer
of defense between the workload and the host.
When you schedule a Kata pod, the container manager hands it off to the Kata
shim, which launches a hypervisor to boot a VM. The container runs inside that
VM on its own guest kernel. Its files (including the image's root filesystem)
are shared into the guest over virtio-fs, served by a host daemon (virtiofsd,
or nydusd with the nydus snapshotter):
flowchart TB
subgraph host["Host"]
containerd["containerd / CRI-O"]
shim["Kata shim (containerd-shim-kata-v2)"]
vmm["Hypervisor / VMM (QEMU, Cloud Hypervisor, ...)"]
virtiofs["virtio-fs daemon (virtiofsd / nydusd)"]
containerd -->|"1. create pod"| shim
shim -->|"2. launch VM"| vmm
shim -->|"2. start fs daemon"| virtiofs
end
subgraph vm["Lightweight VM (own guest kernel)"]
agent["kata-agent"]
workload["Container workload"]
agent -->|"4. start & manage"| workload
end
vmm ==>|"3. boot guest"| vm
shim <-.->|"control channel over VSOCK"| agent
virtiofs ==>|"share host content (virtio-fs)"| workload
!!! note The diagram shows the shim, VMM, and virtio-fs daemon as separate host processes — the case for QEMU and Cloud Hypervisor. With the built-in Dragonball VMM, all three run inside a single process.
Why use Kata Containers?
- Stronger isolation by default. Each pod runs in its own VM with a dedicated guest kernel, so a container breakout or guest-kernel exploit stays in the VM rather than reaching the kernel shared by every other workload on the node.
- A building block for untrusted or multi-tenant workloads. The hardware virtualization boundary is much harder to cross than namespaces and cgroups alone, making it safer to run third-party code or mutually distrusting tenants on shared infrastructure.
- Drop-in compatibility. Kata implements the OCI and CRI shim interface, so
it works with Kubernetes, containerd, CRI-O, and Docker. Opt in per workload
via a
RuntimeClass(or Docker's--runtime) — no application changes. - Reduced host attack surface and flexibility. Workloads never talk directly to the host kernel, and each guest can run its own kernel version and configuration.
!!! warning "Isolation is not multi-tenancy on its own" Kata strengthens workload isolation, but multi-tenancy also depends on network, storage, and control-plane isolation.
Why use Kata Containers with a TEE?
Kata can boot its VMs inside a hardware Trusted Execution Environment (TEE) — Intel TDX, AMD SEV-SNP, or IBM Secure Execution — so the guest's memory is encrypted and integrity-protected by the CPU. This protects data in use: even a compromised host, hypervisor, or cloud operator cannot read or tamper with the workload, and remote attestation lets you cryptographically verify the environment before secrets are released to it.
This is the foundation of the Confidential Containers project. For how to deploy and attest confidential workloads, see its documentation.
Key terminology
- Runtime / shim
- The
containerd-shim-kata-v2process the container manager calls to create and manage the VM behind a pod. Since the 4.0 release the default and recommended runtime isruntime-rs, the Rust implementation. - Agent
- The
kata-agentprocess running inside the guest VM, managing the container's lifecycle on behalf of the runtime. - Hypervisor
- The VMM that boots the guest — QEMU, Cloud Hypervisor, Firecracker, or the built-in Dragonball. See the hypervisors document.
- virtio-fs
- How Kata shares files (including the container's root filesystem) from the
host into the guest. Served by
virtiofsd, ornydusdwith the nydus snapshotter for lazy image pulling. RuntimeClass- The Kubernetes object that tells the cluster to schedule a pod with Kata.
Select it per pod with
runtimeClassName(for example,kata-qemu-runtime-rs). kata-deploy- The recommended installer — a DaemonSet that lays down the Kata binaries
on each node and wires up the container manager and
RuntimeClassobjects.
Try it out
The fastest way to try Kata is the kata-deploy Helm chart on a Kubernetes
cluster. The installation guide covers prerequisites,
other installation methods, and verification in full.
!!! tip "Before you start"
Confirm your host supports hardware virtualization and that /dev/kvm is
available. On x86_64, grep -E -o '(vmx|svm)' /proc/cpuinfo | sort -u
should print vmx (Intel) or svm (AMD).
-
Install the chart:
export VERSION=$(curl -sSL https://api.github.com/repos/kata-containers/kata-containers/releases/latest | jq -r .tag_name) export CHART="oci://ghcr.io/kata-containers/kata-deploy-charts/kata-deploy" helm install kata-deploy "${CHART}" --version "${VERSION}" --namespace kata-system --create-namespace -
Run a pod with a Kata
RuntimeClass:apiVersion: v1 kind: Pod metadata: name: kata-quickstart spec: runtimeClassName: kata-qemu-runtime-rs containers: - name: test image: quay.io/libpod/ubuntu:latest command: ["uname", "-r"]kubectl apply -f kata-quickstart.yaml kubectl logs kata-quickstart
The printed kernel version is the Kata guest kernel, normally different from
the host's (uname -r) — confirming the workload ran inside a VM.