metal-stack
Control plane in k3d
Description
#254 With this PR a new MINI_LAB_FLAVOR "k3s" is added. Enabling this flavor, deploys the metal-stack and gardener control plane in k3d (k3s in docker, one docker container providing a complete k3s setup) We use 1 server (control plane in k3s) and 1 agent (worker node in k3s), because with only one node the pod limit is exceeded. The traefik and servicelb load balancers provided by k3s are disabled because gardener comes with its own istio ingress gateway, which collides with traefik and servicelb.
The CI cleanup is missing (in scripts). Without it we get errors like these from the runners:
Command Output: a71339015cab623896d29ae557a001833d875b24e71919ead43382c8167b293c
docker: Error response from daemon: failed to set up container networking: driver failed programming external connectivity on endpoint metal-control-plane-control-plane (007eaa7c54fd18adb8935a094cf498ef62e55f8de7dc662ced0675bfe04781db): Bind for 0.0.0.0:4150 failed: port is already allocated
Why do we need k3d in the mini-lab if we already have kind for the same use-case?
Why do we need k3d in the mini-lab if we already have kind for the same use-case?
k3d will be used in the autonomous control plane for hosting the initial metal-stack and gardener components.
k3d will be used in the autonomous control plane for hosting the initial metal-stack and gardener components.
#254 is focused on running k3s on the host for the control plane. Are you now looking at the option of running k3s in a container instead of on the host?
k3d will be used in the autonomous control plane for hosting the initial metal-stack and gardener components.
#254 is focused on running k3s on the host for the control plane. Are you now looking at the option of running k3s in a container instead of on the host?
No, with this PR @qrnvttrl only ensures that all components still work in k3s
No, with this PR @qrnvttrl only ensures that all components still work in k3s
I previously managed Metal-Stack and Gardener on k3s for a client without any issues, which isn't surprising given that k3s passes the Kubernetes conformance tests.
On the other hand, if @qrnvttrl is switching the mini-lab from kind to k3d, I would prefer that over introducing a new flavor.