K8s-Copilot

Overview

K8s-Copilot is a command-line tool based on Golang which allows you to create/list/update/delete Kubernetes built-in resources interactively powered by ChatGPT.

Note: Model is currently hard-coded in gpt-4o-mini.

Use Cases

• Create a Kubernetes resource in a specific namespace.
• List Kubernetes either namespaced or non-namespaced resources.
• Update a Kubernetes resource given specific name & namespace.
• Delete a Kubernetes resource given specific name & namespace.

Features

• Cobra
• Function calling
• client-go

Architecture

Cobra

k8s-copilot
├── ask
│   ├── chatgpt
└── analyze
    ├── event

Demo

Prerequisite

An out-of-box Kubernetes cluster environment, try kind. 👈

Install Golang.

Build

$ go build -o k8s-copilot

Setup ENV

Try APIYI 👈 if you had difficulty acquiring OpenAI API Key from mainland China.

# wsl env to win
$ export API_KEY="api_key"
$ export BASE_URL="base_url"

If you're using WSL, add below to /etc/wsl.conf then export the ENV.

[automount]
options = "metadata"

$ export WSLENV=API_KEY/w:BASE_URL/w

Run

Help

$ ./k8s-copilot -h

Ask

$ ./k8s-copilot ask chatgpt

A greeting prompt will show up.

Greetings, I'm a Copilot for Kubernetes, you require my assistant?
>

Type your queries:

Note: open another terminal to run kubectl cmd for checking.

> create a deploy named nginx, image is nginx:latest, replica is 2

# check
$ kubectl get deploy

> ls all pods
> ls all pods in kube-system
> ls all services in kube-system
> ls all namespaces

> update deploy named nginx replica to 3

# check
$ kubectl get deploy

> add label env=test to deploy named nginx

# check
$ kubectl get po --show-labels

> remove label env=test from deploy named nginx

# check
$ kubectl get po --show-labels

> update image of deploy named nginx to nginx:1.26.2

# check
$ kubectl get deploy nginx -o yaml | grep image:

> delete deploy named nginx

$ kubectl get deploy

> exit

Implementation

See more in UML. 👈

Limitation

• Sometimes the update query is not strictly idempotent due to returned response from LLM, please try multiple times.
• The update query will create a new ReplicaSet if you try to add a label to Deployment.

Operation and Maintenance

N/A

Troubleshooting

N/A

Q&A

N/A

Reference

N/A

TODO

• Add a flag to select LLM.
• Replace stdin with GNU-Readline.
• Add event analyzer feature.
• Fine tune system prompt to LLM to improve robustness & Idempotence.