k8s-controller

An implementation of a simple k8s controller:

1. Create custom resource
2. Launch a RPC worker pod with specific memory, cpu and concurrency limit

Generate K8S code

docker build -t code-gen .
docker run -it --rm -v `pwd`:/go/src/github.com/willeslau/k8s-controller code-gen bash

# inside docker
/go/src/k8s.io/code-generator/generate-groups.sh all \
github.com/willeslau/k8s-controller/pkg/client \
github.com/willeslau/k8s-controller/pkg/apis \
worker:v1

K8S generated code should be populated.

Write the controller

The code is in internal/controller, the state machine (work in progress) is

Launch the controller

Get the master location by kubectl cluster-info

./controller --kubeconfig ~/.kube/config

Lessions from Deployment-Controller

1. New status is derived only based on current deployment and all the replicas sets but not the "past" states. In this case, try to set the status purely based on the current deployment and worker. Think about what parameters are needed to deduce the status!

Good References:

1. Replica Controller: https://blog.csdn.net/yan234280533/article/details/78312620
2. Deployment Controller: https://blog.hdls.me/15763918313590.html
3. Deployment Controller Src: https://github.com/kubernetes/kubernetes/blob/27aca3f7c2f06925dd1890cd8b0522b775b0e323/pkg/controller/deployment/deployment_controller.go