pi8s cluster config

Deploy a kubernetes cluster across two or more raspberry pi's using Terraform. This repository also contains config and instructions for deploying:

1. on-cluster load balancer (metallb)
2. Argo CD continuous delivery tool
3. centralized logging (ELK stack)

The deployment strategy currently taken is to patch helm charts using helm install -f myvalues.yaml when it is convenient. Otherwise, explicit deployment yaml is used. Helm charts require patching because the x86 images need to be swapped out for corresponding arm64 images (as well as any other custom config that is required).

Requirements:

• Each raspberry pi has an SD card flashed with Ubuntu 20.04 ARM64
• The ubuntu user password is the same on each raspberry pi

Prepare nodes and create cluster using Terraform

Clone the terraform-raspberrypi-bootstrap repository and follow the instructions in the README to provision nodes and create a new k8s cluster.

$ git clone git@github.com:adfindlater/terraform-raspberrypi-bootstrap.git

After running terraform_pis.sh, verify the existence of the new k8s cluster by coping ~/terraform-raspberrypi-bootstrap/control-plane/kube_config to ~/.kube/config and then running

$ kubectl get pods --all-namespaces

Deploy bare-metal load balancer

Deploy a load balancer to the cluster using helm. Custom configuration values are provided by the myvalues.yaml file.

$ helm install -f pi8s-config/load-balancer/helm/myvalues.yaml metallb bitnami/metallb

myvalues.yaml

configInline: 
  address-pools:
    - name: default
      protocol: layer2
      addresses:
      - 192.168.1.125-192.168.1.130

speaker:
  image:
    registry: docker.io
    repository: metallb/speaker
    tag: v0.9.3

controller:
  image:
    registry: docker.io
    repository: metallb/controller
    tag: v0.9.3

Argo CD

$ helm install -f pi8s-config/argo-cd/helm/myvalues.yaml argo-cd argo/argo-cd

myvalues.yaml

global:
  image:
    # https://github.com/argoproj/argo-cd/issues/2167#issuecomment-584936692
    repository: phillebaba/argocd
    tag: v1.3.6

server:
  service:
    type: LoadBalancer

Centralized logging with the ELK stack

Elasticsearch

TODO: switch to helm chart

Create a persistent volume in the same namespace that the ELK stack will be deployed. Make sure that k8s can write/read to the host path /mnt/data on the worker nodes.

$ kubectl apply -f pi8s-config/elk/pv.yaml -n kube-system

Deploy Elasticsearch to the cluster.

$ kubectl apply -f pi8s-config/elk/es/yaml -n kube-system

Filebeat

Deploy using helm with customized values.

myvalues.yaml

image: "kasaoden/filebeat"
imageTag: "7.2.0-arm64"

filebeatConfig:
  filebeat.yml: |
    filebeat.inputs:
    - type: container
      paths:
        - /var/log/containers/*.log
      processors:
      - add_kubernetes_metadata:
          host: ${NODE_NAME}
          matchers:
          - logs_path:
              logs_path: "/var/log/containers/"
    output.logstash:
      hosts: ["logstash-service:5044"]

$ helm install -f pi8s-config/elk/filebeat/helm/myvalues.yaml filebeat elastic/filebeat -n kube-system

Logstash

TODO: switch to helm chart

kubectl apply -f pi8s-config/elk/logstash/yaml -n kube-system

Kibana

myvalues.yaml

image: "gagara/kibana-oss-arm64"
imageTag: "7.6.2"

resources:
  requests:
    cpu: "500m"
    memory: "1Gi"
  limits:
    cpu: "500m"
    memory: "1Gi"

elasticsearchHosts: "http://elasticsearch-master:9200"

service:
  type: LoadBalancer

$ helm install -f pi8s-config/elk/kibana/helm/myvalues.yaml kibana elastic/kibana -n kube-system