k8s-operator

README

k8s-operator

K8s operator for the Eidolon project

Description

Three k8s operators are implemented in this project:

• Eidolon Machine Operator: This operator is responsible for managing the Eidolon Machine resources.
• Eidolon Agent Operator: This operator is responsible for managing Eidolon Agent resources.
• Eidolon Reference Operator: This operator is responsible for managing Eidolon Reference resources.

Getting Started

Prerequisites

• go version v1.20.0+
• Access to a Kubernetes v1.11.3+ cluster.
• OLM. The instructions for our operator on operatorhub.io will install OLM for you.

Installation

TODO - once we are accepted on operatorhub.io

Local machine deployment

We recommend minikube for local development. You can find the minikube installation instructions here.

We develop on a Mac, so we have the following instructions for Mac users. If you are on a different OS, please refer to the minikube documentation.

• Install minikube

  brew install minikube
  minikube start --driver qemu --network socket_vmnet  —disk 100g
  alias kubectl="minikube kubectl --"   
  eval $(minikube docker-env)
  

• Install your secrets

  A secret with the needed credentials for your Eidolon deployment. The name of the secret is eidolon.

  See the Eidolon documentation for more information on what keys you need. However, typically this will be a secret with the following keys:

  • OLLAMA_URL - the URL for your Ollama deployment if you are using ollama.
  • OPENAI_API_KEY - your OpenAI API key if you are using OpenAI.
  • MISTRAL_API_KEY - your Mistral API key if you are using Mistral.
  • ANTHROPIC_API_KEY - your Anthropic API key if you are using Anthropic.

  If you already have a .env file with these keys, you can create the secret with the following command:

  kubectl create secret generic eidolon --from-env-file=<location of your env file>/.env
  

  ⚠️ To use the secret in your deployment, you need to add the following to your deployment yaml file's spec:

  envFrom:
    - secretRef:
        name: eidolon
  

• Install the eidolon operator

  helm repo add eidolon-operator https://eidolonai.com/charts 
  helm install eidolon-operator eidolon-operator/eidolon-operator-chart --namespace eidolon-operator-system --create-namespace
  

  If you already have added the repo and you want to update it:

  helm repo update
  helm uninstall eidolon-operator -n eidolon-operator-system                  
  helm install eidolon-operator eidolon-operator/eidolon-operator-chart --namespace eidolon-operator-system --create-namespace
  

Development

Now you can deploy your resources in your k8s cluster. To apply your machine resource:

kubectl apply -f <your machine location>.<your machine>.yaml

Just like any other k8s resource, you can apply the machine to any namespace. Use the -n flag on kubectl to specify the namespace.

NOTE: Only one machine can run per namespace

Now you can apply your agent resource:

kubectl apply -f <your agent location>.<your agent>.yaml

Make sure to apply the agent to the same namespace as the machine.

Exposing a machine locally

There are many ways to expose a minikube service locally. One way is to create a load balancer service which will then be exposed on your local computer.

To create a load balancer service for the eidolon server:

kubectl expose deployment eidolon-deployment --type=LoadBalancer --name=eidolon-server --port 8080

To get the IP address of the eidolon server service:

minikube service eidolon-server --url

You can use this URL to access your eidolon server.

Running the web UI

To start the webui, you can use the following command:

kubectl apply -f config/samples/webui.yaml

To expose the webui service we need to create a load balancer service, get the URL, and update the nextjs config map with the URL.:

# Get the URL from minikube
API_URL=$(minikube service eidolon-webui --url)

# Update the ConfigMap
kubectl create configmap eidolon-webui-config --from-literal=NEXT_PUBLIC_API_URL=$API_URL -o yaml --dry-run=client | kubectl apply -f -

# Restart the deployment to pick up the new ConfigMap value
kubectl rollout restart deployment eidolon-webui-deployment

echo "Web UI is available at $API_URL"