Taskfile.yml

# yaml-language-server: $schema=https://taskfile.dev/schema.json
version: '3'

vars:
  # Default version to upgrade to - can be overridden with VERSION=x.x.x
  TALOS_UPGRADE_VERSION: v1.10.6

tasks:
  default:
    cmds:
      - task --list

  up:
    desc: Brings up the entire system in order.
    cmds:
      - task: 00-bootstrap
      - task: 01-proxmox
      - task: 02-system
      - task: 03-base
    vars:
      STEPS: >-
        00-bootstrap
        01-proxmox
        02-system
        03-base

  teardown:
    desc: Tears down the entire system.
    cmds:
      - task: 03-base-teardown
      - task: 02-system-teardown
      - task: 01-proxmox-teardown
      - task: 00-bootstrap-teardown

  rebuild:
    desc: Rebuilds the entire system from scratch.
    cmds:
      - task: teardown
      - task: up

  00-bootstrap:
    desc: Sets up the initial bootstrap configuration.
    cmds:
      - task: terraform-up
        vars:
          DIR: 00-bootstrap

  01-proxmox:
    desc: Deploys the Proxmox-related configuration.
    cmds:
      - task: terraform-up
        vars:
          DIR: 01-proxmox

  02-system:
    desc: Applies the system-level configuration.
    cmds:
      - task: terraform-up
        vars:
          DIR: 02-system
      - until talosctl health; do echo 'Waiting for system to be healthy...'; sleep 5; done

  03-base:
    desc: Deploys the base configuration for the cluster.
    cmds:
      - task: terraform-up
        vars:
          DIR: 03-base

  00-bootstrap-teardown:
    desc: Tears down the bootstrap configuration.
    cmds:
      - task: terraform-down
        vars:
          DIR: 00-bootstrap

  01-proxmox-teardown:
    desc: Tears down the Proxmox-related configuration.
    cmds:
      - task: terraform-down
        vars:
          DIR: 01-proxmox

  02-system-teardown:
    desc: Tears down the system-level configuration.
    cmds:
      - task: terraform-down
        vars:
          DIR: 02-system

  03-base-teardown:
    desc: Tears down the base configuration for the cluster.
    cmds:
      - task: terraform-down
        vars:
          DIR: 03-base

  init-macos:
    desc: Initialize a macOS workstation with dependencies
    cmds:
      - brew install {{.DEPS}} {{.CLI_ARGS}}
    preconditions:
      - sh: command -v brew
        msg: |
          Homebrew is not installed.
    vars:
      DEPS: >-
        go-task/tap/go-task
        kubectl
        kustomize
        helm
        terraform
        talosctl
        k9s
        yq
        jq

  terraform-up:
    internal: true
    dir: "{{.DIR}}"
    cmds:
      - terraform init
      - terraform apply -auto-approve

  terraform-down:
    internal: true
    dir: "{{.DIR}}"
    cmds:
      - terraform destroy -auto-approve

  # Internal helper tasks
  get-controller-ips:
    silent: true
    cmds:
      - cd 01-proxmox && terraform output -json controller_nodes | jq -r 'to_entries[].value.address_ipv4'

  get-worker-ips:
    silent: true
    cmds:
      - cd 01-proxmox && terraform output -json worker_nodes | jq -r 'to_entries[].value.address_ipv4'

  get-all-ips:
    silent: true
    cmds:
      - task: get-controller-ips
      - task: get-worker-ips

  get-upgrade-image:
    silent: true
    cmds:
      - |
        VERSION="{{.VERSION | default .TALOS_UPGRADE_VERSION}}"
        SCHEMATIC_ID=$(cd 01-proxmox && terraform output -raw talos_image_factory_schematic | cut -d'/' -f3 | cut -d':' -f1)
        echo "factory.talos.dev/nocloud-installer/${SCHEMATIC_ID}:${VERSION}"

  upgrade-node-from-ip:
    requires:
      vars: [IP]
    cmds:
      - task: upgrade-node
        vars:
          NODE: "{{.IP}}"

  upgrade-node:
    desc: "Upgrade a specific Talos node"
    requires:
      vars: [NODE]
    cmds:
      - echo "Upgrading node {{.NODE}}..."
      - |
        IMAGE=$(task get-upgrade-image)
        echo "Using image: $IMAGE"
        talosctl upgrade --nodes {{.NODE}} --image "$IMAGE"

  upgrade-controllers:
    desc: "Upgrade all controller nodes sequentially"
    cmds:
      - echo "Upgrading controller nodes..."
      - |
        task get-controller-ips | while read -r ip; do
          echo "Upgrading controller node: $ip"
          task upgrade-node-from-ip IP="$ip"
        done

  upgrade-workers:
    desc: "Upgrade all worker nodes sequentially"
    cmds:
      - echo "Upgrading worker nodes..."
      - |
        task get-worker-ips | while read -r ip; do
          echo "Upgrading worker node: $ip"
          task upgrade-node-from-ip IP="$ip"
        done

  upgrade-all:
    desc: "Upgrade all nodes sequentially (controllers first, then workers)"
    cmds:
      - task: upgrade-controllers
      - task: upgrade-workers

  check-upgrade-status:
    desc: Check the current Talos version on all nodes
    cmds:
      - |
        echo "🔍 Checking Talos versions on all nodes..."
        cd 01-proxmox && terraform output -json | jq -r '
          (.controller_nodes.value // {}) as $controllers |
          (.worker_nodes.value // {}) as $workers |
          ([$controllers, $workers] | add | to_entries[] | .value.address_ipv4)
        ' | sort -u | while IFS= read -r node; do
          if [ -n "$node" ]; then
            echo "Checking node $node..."
            talosctl version --nodes "$node" || echo "Failed to get version for $node"
          fi
        done

  health-check:
    desc: Perform comprehensive health check of the cluster
    cmds:
      - echo "🏥 Performing cluster health check..."
      - echo "Checking Talos cluster health..."
      - talosctl health
      - echo "Checking Kubernetes cluster info..."
      - kubectl cluster-info
      - echo "Checking node status..."
      - kubectl get nodes -o wide
      - echo "Checking system pods..."
      - kubectl get pods -n kube-system

  check-versions:
    desc: Check current and available Talos versions
    cmds:
      - |
        echo "🔍 Current Talos version on nodes:"
        cd 01-proxmox && terraform output -json | jq -r '
          (.controller_nodes.value // {}) as $controllers |
          (.worker_nodes.value // {}) as $workers |
          ([$controllers, $workers] | add | to_entries[] | .value.address_ipv4)
        ' | sort -u | while IFS= read -r node; do
          if [ -n "$node" ]; then
            echo "Node $node:"
            talosctl version --nodes "$node" --short 2>/dev/null || echo "  Failed to get version"
          fi
        done
      - echo ""
      - echo "📦 Configured upgrade version{{":"}} {{.TALOS_UPGRADE_VERSION}}"
      - echo ""
      - echo "🌐 Latest Talos releases (check https://github.com/siderolabs/talos/releases):"
      - curl -s https://api.github.com/repos/siderolabs/talos/releases | jq -r '.[0:5][] | "\(.tag_name) - \(.published_at | split("T")[0])"' || echo "Failed to fetch latest releases"