kindo-flux

Just a shell script.
The objective here is to build a play-ground for experimenting with the Flux 2 and ArgoCD GitOps toolkits on a kind based k8s cluster. The resulting setup should

• be fully self-contained, i.e. include its own git repository service
• be reproducible and disposable infrastructure-as-code
• use podman as base layer container runtime (Docker is considered deprecated)
• work at least on
  • up to date Fedora Linux (in reality, this means SilverBlue due to its native support for podman)
  • and (within the limitations imposed by the outdated kernel of this platform) standard WSL installations

kindo-flux consists of:

• the main shell script
• an auxiliary interactive shell setup for managing kubectl [contexts][kubectl-contexts] in a read-only fashion
• an auxiliary shell script for adding packages and package repositories enabling the installation of podman on Ubuntu Focal based Linux systems (as this is the standard distribution for WSL instances)

NOTE: kindo-flux is intended to be only a development / proof-of-concept helper tool – it is shellchecked, but that's about it with respect to "production quality" SW. So don't expect stuff like robust error handling, modularization, easy extensibility etc..
The result can be considered to large for implementation as a shell script, and not very eye-friendly due to excessive parametrization and embedded YAML, but OTOH it is linear in execution and straight-forward to understand.

NOTE: as Flux 2 is capable of bootstrapping and lifecycle-managing its own installation into a cluster, and has only k8s and a git server as requirements (while at least the argocd CLI needs an ingress), it is used for managing ArgoCD as a service.

Installation and usage

Download the main script, make it executable

chmod 0755 kindo-flux.sh

and execute it

./kindo-flux.sh

to install a cluster, which is named kind-k8s-<k8s-version> per default. If you pass a command line argumend to the script, this will be used as the cluster name.

NOTE: don't forget to tune host kernel parameters for serious container workloads – esp. inotify, but also (for ArgoCD) number of standard open file descriptors per systemd user session / ulimit shell settings are configured too low for running complex k8s cloud-native applicaions on many mainstream Linux distributions, including Fedora and Ubuntu.

NOTE: when using a WSL instance as host system, kind has to be called with superuser privileges, as at least the WSL default kernel as distributed by Microsoft unfortunately does not support podman with rootless containers.
However, this is not considered a showstopper as it's best practice to spin up a dedicated WSL instance for the containerized k8s anyway.