Files
kata-containers/src/runtime-rs
Manuel Huber 7a409a1295 nvidia/runtime-rs: use EROFS without virtio-fs
Disable filesystem sharing for the non-confidential NVIDIA runtime-rs
handler and use the EROFS snapshotter as its Kubernetes image-layer
transport. This moves the runtime class toward a guest-owned storage
model instead of relying on virtio-fs for container image layers and
writable cache volumes.

Configure kata-deploy's NVIDIA GPU values to install EROFS and select
it for the qemu-nvidia-gpu-runtime-rs Kubernetes handler. Use
memory-backed writable layers and dm-verity for lower-layer integrity.

Adjust the NVIDIA GPU Kubernetes CI matrix so the non-confidential
runtime-rs job exercises EROFS, while the Go and TEE jobs keep their
existing snapshotter choices.

Keep the Docker smoke test path on virtio-fs. The EROFS setup is
targeted at the Kubernetes runtime-rs handler, where containerd can use
the EROFS snapshotter for image layers.

Adjust the runtime-rs NIM test selection and manifests closer to the
TEE case, where filesystem sharing is already disabled and cache
storage is guest-owned and ephemeral.

Signed-off-by: Manuel Huber <manuelh@nvidia.com>
Assisted-by: OpenAI Codex <codex@openai.com>
2026-07-19 09:12:45 +02:00
..
2026-02-26 09:37:46 +00:00

runtime-rs

What is runtime-rs

runtime-rs is a core component of Kata Containers 4.0. It is a high-performance, Rust-based implementation of the containerd shim v2 runtime.

Key characteristics:

  • Implementation Language: Rust, leveraging memory safety and zero-cost abstractions
  • Project Maturity: Production-ready component of Kata Containers 4.0
  • Architectural Design: Modular framework optimized for Kata Containers 4.0

For architecture details, see Architecture Overview.

Architecture Overview

Key features:

  • Built-in VMM (Dragonball): Deeply integrated into shim lifecycle, eliminating IPC overhead for peak performance
  • Asynchronous I/O: Tokio-based async runtime for high-concurrency with reduced thread footprint
  • Extensible Framework: Pluggable hypervisors, network interfaces, and storage backends
  • Resource Lifecycle Management: Comprehensive sandbox and container resource management

crates overview

Crates

Crate Description
shim Containerd shim v2 entry point (start, delete, run commands)
service Services including TaskService for containerd shim protocol
runtimes Runtime handlers: VirtContainer (default), LinuxContainer(experimental), WasmContainer(experimental)
resource Resource management: network, share_fs, rootfs, volume, cgroups, cpu_mem
hypervisor Hypervisor implementations
agent Guest agent communication (KataAgent)
persist State persistence to disk (JSON format)
shim-ctl Development tool for testing shim without containerd

shim

Entry point implementing containerd shim v2 binary protocol:

  • start: Start new shim process
  • delete: Delete existing shim process
  • run: Run ttRPC service

service

Extensible service framework. Currently implements TaskService conforming to containerd shim protocol.

runtimes

Runtime handlers manage sandbox and container operations:

Handler Feature Flag Description
VirtContainer virt (default) Virtual machine-based containers
LinuxContainer linux Linux container runtime (experimental)
WasmContainer wasm WebAssembly runtime (experimental)

resource

All resources abstracted uniformly:

  • Sandbox resources: network, share-fs
  • Container resources: rootfs, volume, cgroup

Sub-modules: cpu_mem, cdi_devices, coco_data, network, share_fs, rootfs, volume

hypervisor

Supported hypervisors:

Hypervisor Mode Description
Dragonball Built-in Integrated VMM for peak performance (default)
QEMU External Full-featured emulator
Cloud Hypervisor External Modern VMM (x86_64, aarch64)
Firecracker External Lightweight microVM
Remote External Remote hypervisor

The built-in VMM mode (Dragonball) is recommended for production, offering superior performance by eliminating IPC overhead.

agent

Communication with guest OS agent via ttRPC. Supports KataAgent for full container lifecycle management.

persist

State serialization to disk for sandbox recovery after restart. Stores state.json under /run/kata/<sandbox-id>/.

Build from Source and Install

Prerequisites

Download Rustup and install Rust. For Rust version, see languages.rust.meta.newest-version in versions.yaml.

Example for x86_64:

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
source $HOME/.cargo/env
rustup install ${RUST_VERSION}
rustup default ${RUST_VERSION}-x86_64-unknown-linux-gnu

Musl Support (Optional)

For fully static binary:

# Add musl target
rustup target add x86_64-unknown-linux-musl

# Install musl libc (example: musl 1.2.3)
curl -O https://git.musl-libc.org/cgit/musl/snapshot/musl-1.2.3.tar.gz
tar vxf musl-1.2.3.tar.gz
cd musl-1.2.3/
./configure --prefix=/usr/local/
make && sudo make install

Install Kata 4.0 Rust Runtime Shim

git clone https://github.com/kata-containers/kata-containers.git
cd kata-containers/src/runtime-rs
make && sudo make install

After installation:

  • Config file: /usr/share/defaults/kata-containers/configuration.toml
  • Binary: /usr/local/bin/containerd-shim-kata-v2

Install Without Built-in Dragonball VMM

To build without the built-in Dragonball hypervisor:

make USE_BUILTIN_DB=false

Specify hypervisor during installation:

sudo make install HYPERVISOR=qemu
# or
sudo make install HYPERVISOR=clh-runtime-rs

Configuration

Configuration files in config/:

Config File Hypervisor Notes
configuration-dragonball.toml.in Dragonball Built-in VMM
configuration-qemu-runtime-rs.toml.in QEMU Default external
configuration-clh-runtime-rs.toml.in Cloud Hypervisor Modern VMM
configuration-rs-fc.toml.in Firecracker Lightweight microVM
configuration-remote.toml.in Remote Remote hypervisor
configuration-qemu-tdx-runtime-rs.toml.in QEMU + TDX Intel TDX confidential computing
configuration-qemu-snp-runtime-rs.toml.in QEMU + SEV-SNP AMD SEV-SNP confidential computing
configuration-qemu-se-runtime-rs.toml.in QEMU + SEV AMD SEV confidential computing
configuration-qemu-coco-dev-runtime-rs.toml.in QEMU + CoCo CoCo development

See runtime configuration for configuration options.

Logging

See Developer Guide - Troubleshooting.

Debugging

For development, use shim-ctl to test shim without containerd dependencies.

Limitations

See Limitations for details.