With the announcement of SpinKube, this repository is being deprecated. Please see the Installing with Helm section of the spinkube.dev installation documentation for the recommended approach to installing the containerd-shim-spin going forward.
This project provides an automated method to install and configure the containerd shim for Fermyon Spin in Kubernetes.
The version of the container image and Helm chart directly correlates to the version of the containerd shim. For simplicity, here is a table depicting the version matrix between Spin and the containerd shim.
containerd-shim-spin-v* | Spin |
---|---|
v0.9.3 | v2.0.0 |
v0.9.2 | v1.5.0 |
v0.9.1 | v1.4.1 |
v0.9.0 | v1.4.1 |
v0.8.0 | v1.4.0 |
v0.7.0 | v1.2.0 |
v0.6.0 | v1.1.0 |
v0.5.1 | v1.0.0 |
At a high level, in order to add a new runtime shim to containerd we must accomplish the following:
containerd-shim-spin-v1
binary to the node's path (default location: /usr/local/bin
)containerd-shim-spin-v1
runtime to containerd's config (default location: /etc/containerd/config.toml
)RuntimeClass
that you can specify in a pod's spec for containerd to useBecause of these constraints, installing an additional runtime for containerd requires privileged access to a node. Currently this repository only contains a way to install the runtime shim via Kubernetes resources but another option would be to customize a base image for your nodes with these constraints in mind.
This project provides a Helm chart that includes a DaemonSet which runs an init container in privileged mode in order to copy the binary to the node and update the containerd config with the new runtime. This is the most generic way to install the containerd runtime shim in Kubernetes environments.
helm install spin-containerd-shim-installer oci://ghcr.io/fermyon/charts/spin-containerd-shim-installer --version 0.9.3
As mentioned above, the Helm chart's method of installation does currently require privileged access to a node. Please be sure to review the DaemonSet, install script entrypoint.sh and accompanying Dockerfile.