Skip to content

Latest commit

 

History

History
87 lines (60 loc) · 2.52 KB

run-with-kata-containers.md

File metadata and controls

87 lines (60 loc) · 2.52 KB
Raw

Run with Kata Containers

Get the image from dockerhub.

Kata Containers is a lightweight virtual machine (VM) for containers. It is designed to provide the speed of containers and the isolation of VMs. Kata Containers can run on multiple hypervisors and is compatible CRI for Kubernetes.

The documentation is based on Kata Containers 3.1.3, but in practice kata 3.x should be able to follow this documentation.

The pre-built kernel from Kata Containers does not support eBPF, so we need to build the kernel ourselves. this documentation is about how to build the kernel and configure kata containers to use it.

If you haven't installed Kata Containers yet, please refer to the official documentation.

Get code from github

git clone --depth 1 --branch 3.1.3 https://github.com/kata-containers/kata-containers.git

Setup the kernel

  1. Download the kernel source code, version 6.1.38 as an example
cd kata-containers/tools/packaging/kernel
./build-kernel.sh -v 6.1.38 setup
  1. Change tools/packaging/kernel/configs/fragments/x86_64/.config to meet kernel_config requirements, given an full example here
  2. Build the kernel
./build-kernel.sh -v 6.1.38 build
  1. Install the kernel to /usr/share/kata-containers/
sudo ./build-kernel.sh install

Configure kata containers

Kata-qemu config file as example: Edit /opt/kata/share/defaults/kata-containers/configuration-qemu.toml 

# Replace the kernel path
#kernel = "/opt/kata/share/kata-containers/vmlinux.container"
kernel = "/usr/share/kata-containers/vmlinuz.container"

Verify

Start container via nerdctl

sudo nerdctl run -it --runtime=io.containerd.kata.v2 --rm alpine

Run uname -a

Linux 856b63ebabcc 6.1.38 #2 SMP Fri Aug 11 16:20:36 CST 2023 x86_64 Linux

Run duetector images

sudo nerdctl run \
-it \
-p 8888:8888 \
-p 8120:8120 \
--runtime=io.containerd.kata.v2 \
--cap-add=sys_admin \
--entrypoint bash \
--rm \
dataucon/duetector

If use bash as entrypoint, you need to mount debugfs manually

mount -t debugfs debugfs /sys/kernel/debug

More information about tracking mlflow with duetector, please refer to usercases