launchpad/.tofu/scripts/setup-cluster.sh

#!/bin/bash
set -euo pipefail

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
TOFU_ROOT="$(dirname "$SCRIPT_DIR")"
PROJECT_ROOT="$(dirname "$TOFU_ROOT")"

# ─── Usage ────────────────────────────────────────────────────────────
usage() {
  cat <<EOF
Usage: $0 <cluster> [options]

  Provision a Kubernetes cluster using OpenTofu.
  Mirrors bootstrap.sh convention: cluster = <platform>-<env>

  Clusters:   aks-dev | aks-prod | eks-dev | eks-prod
              gke-dev | gke-prod | upc-dev | upc-prod
              <platform>-workload (for workload clusters)

  Options:
    --plan        Plan only, don't apply
    --destroy     Destroy the cluster (use teardown-cluster.sh instead)
    --auto        Skip confirmation prompts
    -h, --help    Show this help

  Examples:
    $0 aks-dev
    $0 eks-prod --plan
    $0 upc-dev --auto

  Prerequisites:
    - tofu, kubectl, helm installed
    - Platform credentials in .tofu/configs/<platform>.env
    - Cluster config in clusters/<cluster>.yaml

  After provisioning, run:
    ./bootstrap.sh <cluster>
EOF
  exit "${1:-0}"
}

# ─── Parse arguments ──────────────────────────────────────────────────
CLUSTER=""
PLAN_ONLY=false
DESTROY=false
AUTO_APPROVE=false

while [[ $# -gt 0 ]]; do
  case "$1" in
    --plan)       PLAN_ONLY=true; shift ;;
    --destroy)    DESTROY=true; shift ;;
    --auto)       AUTO_APPROVE=true; shift ;;
    -h|--help)    usage 0 ;;
    -*)           echo "Unknown option: $1"; usage 1 ;;
    *)
      if [[ -z "$CLUSTER" ]]; then
        CLUSTER="$1"
      else
        echo "Error: unexpected argument '$1'"
        usage 1
      fi
      shift
      ;;
  esac
done

[[ -z "$CLUSTER" ]] && { echo "Error: <cluster> argument required"; usage 1; }

# ─── Map cluster → platform + env ────────────────────────────────────
PLATFORM="${CLUSTER%%-*}"   # aks-dev → aks
ENV="${CLUSTER#*-}"         # aks-dev → dev

case "$PLATFORM" in
  aks|eks|gke|upc) ;;
  *) echo "Error: unknown platform '$PLATFORM'. Expected: aks, eks, gke, upc"; exit 1 ;;
esac

TOFU_DIR="$TOFU_ROOT/platforms/$PLATFORM/$ENV"
if [[ ! -d "$TOFU_DIR" ]]; then
  echo "Error: tofu directory not found: $TOFU_DIR"
  echo "Available environments for $PLATFORM:"
  ls -1 "$TOFU_ROOT/platforms/$PLATFORM/" 2>/dev/null | grep -v modules || echo "  (none)"
  exit 1
fi

echo "========================================="
echo "  Kubernetes Cluster Setup"
echo "========================================="
echo ""
echo "  Cluster:   $CLUSTER"
echo "  Platform:  $PLATFORM"
echo "  Env:       $ENV"
echo "  Tofu dir:  $TOFU_DIR"
echo ""

# ─── Prerequisites ────────────────────────────────────────────────────
echo "=== Checking Prerequisites ==="
command -v tofu    >/dev/null 2>&1 || { echo "Error: tofu is not installed."; exit 1; }
command -v kubectl >/dev/null 2>&1 || { echo "Error: kubectl is not installed."; exit 1; }
command -v helm    >/dev/null 2>&1 || { echo "Error: helm is not installed."; exit 1; }
echo "  tofu, kubectl, helm: OK"

# ─── Load platform credentials ────────────────────────────────────────
ENV_FILE="$TOFU_ROOT/configs/$PLATFORM.env"
if [[ -f "$ENV_FILE" ]]; then
  echo "  Loading credentials from configs/$PLATFORM.env"
  set -a
  # shellcheck disable=SC1090
  source "$ENV_FILE"
  set +a
else
  echo "  Warning: $ENV_FILE not found — using existing environment/CLI auth"
  echo "  Copy configs/$PLATFORM.env.example → configs/$PLATFORM.env to configure"
fi

# ─── Load cluster config (if exists) ──────────────────────────────────
CLUSTER_CONFIG="$PROJECT_ROOT/clusters/$CLUSTER.yaml"
if [[ -f "$CLUSTER_CONFIG" ]]; then
  echo "  Loading cluster config from clusters/$CLUSTER.yaml"
  if command -v yq >/dev/null 2>&1; then
    eval "$(yq -r 'to_entries[] | "export CLUSTER_\(.key)=\"\(.value)\""' "$CLUSTER_CONFIG")"
    echo "  Cluster name: ${CLUSTER_clusterName:-$CLUSTER}"
  else
    echo "  Warning: yq not installed — cluster config not loaded"
  fi
else
  echo "  Warning: $CLUSTER_CONFIG not found — using defaults"
fi
echo ""

# ─── Run OpenTofu ─────────────────────────────────────────────────────
cd "$TOFU_DIR"

echo "=== Initializing OpenTofu ==="
tofu init

echo ""
if $DESTROY; then
  echo "=== Planning Destruction ==="
  tofu plan -destroy -out=tfplan

  if ! $AUTO_APPROVE; then
    echo ""
    read -rp "DESTROY cluster $CLUSTER? This is irreversible. (yes/no) " REPLY
    [[ "$REPLY" == "yes" ]] || { echo "Cancelled."; exit 1; }
  fi

  echo "Destroying infrastructure..."
  tofu apply tfplan
  echo ""
  echo "=== Cluster $CLUSTER Destroyed ==="

elif $PLAN_ONLY; then
  echo "=== Planning Infrastructure ==="
  tofu plan
  echo ""
  echo "=== Plan complete (--plan mode, no changes applied) ==="

else
  echo "=== Planning Infrastructure ==="
  tofu plan -out=tfplan

  if ! $AUTO_APPROVE; then
    echo ""
    read -rp "Apply this plan for $CLUSTER? (y/n) " -n 1 REPLY
    echo
    [[ "$REPLY" =~ ^[Yy]$ ]] || { echo "Cancelled."; exit 1; }
  fi

  echo "Applying infrastructure..."
  tofu apply tfplan

  # ─── Save kubeconfig ──────────────────────────────────────────────
  KUBECONFIG_DIR="$PROJECT_ROOT/private/$CLUSTER"
  mkdir -p "$KUBECONFIG_DIR"
  KUBECONFIG_FILE="$KUBECONFIG_DIR/kubeconfig"

  echo ""
  echo "=== Saving Kubeconfig ==="

  case "$PLATFORM" in
    aks)
      if tofu output -raw kubeconfig > "$KUBECONFIG_FILE" 2>/dev/null; then
        echo "  Saved from tofu output"
      else
        echo "  Fetching from Azure CLI..."
        RG=$(tofu output -raw resource_group_name 2>/dev/null || echo "${CLUSTER_clusterName:-$CLUSTER}-rg")
        NAME=$(tofu output -raw cluster_name 2>/dev/null || echo "${CLUSTER_clusterName:-$CLUSTER}")
        az aks get-credentials --resource-group "$RG" --name "$NAME" --file "$KUBECONFIG_FILE" --overwrite-existing
      fi
      ;;
    eks)
      NAME=$(tofu output -raw cluster_name 2>/dev/null || echo "${CLUSTER_clusterName:-$CLUSTER}")
      REGION=$(tofu output -raw aws_region 2>/dev/null || echo "${AWS_REGION:-eu-west-1}")
      aws eks update-kubeconfig --name "$NAME" --region "$REGION" --kubeconfig "$KUBECONFIG_FILE"
      ;;
    gke)
      NAME=$(tofu output -raw cluster_name 2>/dev/null || echo "${CLUSTER_clusterName:-$CLUSTER}")
      REGION=$(tofu output -raw region 2>/dev/null || echo "${GCP_REGION:-europe-west4}")
      PROJECT=$(tofu output -raw project_id 2>/dev/null || echo "${GCP_PROJECT_ID:-}")
      gcloud container clusters get-credentials "$NAME" --region "$REGION" --project "$PROJECT" 2>/dev/null \
        && cp ~/.kube/config "$KUBECONFIG_FILE" \
        || echo "  Warning: could not fetch kubeconfig via gcloud"
      ;;
    upc)
      if tofu output -raw kubeconfig > "$KUBECONFIG_FILE" 2>/dev/null; then
        echo "  Saved from tofu output"
      else
        CLUSTER_ID=$(tofu output -raw cluster_id 2>/dev/null || echo "${UPCLOUD_CLUSTER_ID:-}")
        if [[ -n "$CLUSTER_ID" ]]; then
          upctl kubernetes config "$CLUSTER_ID" > "$KUBECONFIG_FILE"
        else
          echo "  Warning: could not determine cluster ID for kubeconfig"
        fi
      fi
      ;;
  esac

  if [[ -f "$KUBECONFIG_FILE" ]]; then
    chmod 600 "$KUBECONFIG_FILE"
    echo "  Kubeconfig: $KUBECONFIG_FILE"
  fi

  # ─── Wait for nodes ──────────────────────────────────────────────
  echo ""
  echo "=== Waiting for Cluster Nodes ==="
  export KUBECONFIG="$KUBECONFIG_FILE"
  if kubectl wait --for=condition=Ready nodes --all --timeout=300s 2>/dev/null; then
    echo "  All nodes ready"
  else
    echo "  Warning: nodes not ready within timeout — check cluster status"
  fi

  # ─── Summary ─────────────────────────────────────────────────────
  echo ""
  echo "========================================="
  echo "  Cluster $CLUSTER Provisioned"
  echo "========================================="
  echo ""
  echo "  Kubeconfig: $KUBECONFIG_FILE"
  echo ""
  echo "  Next steps:"
  echo "    export KUBECONFIG=$KUBECONFIG_FILE"
  echo "    ./bootstrap.sh $CLUSTER"
  echo ""
fi