deploy-node.yml

name: Deploy Game Nodes
on:
    workflow_dispatch:
        inputs:
            environment:
                type: environment
                description: Select the environment
                default: Production

jobs:
    deploy:
        runs-on: self-hosted
        environment: ${{ inputs.environment }}
        steps:
            - name: Checkout repo
              uses: actions/checkout@v3

            - name: Deploy Game Nodes
              run: |
                  ENV=$(echo "${{ github.event.inputs.environment }}" | tr '[:upper:]' '[:lower:]')

                  if [ "${{ github.event.inputs.environment }}" = "Production" ]; then
                    NAMESPACE="keyteki"
                    REPLICAS=2
                    SCALE_INCREMENT=2
                    ENV_SUBDOMAIN="production"
                  elif [ "${{ github.event.inputs.environment }}" = "Development" ]; then
                    NAMESPACE="keyteki-development"
                    REPLICAS=1
                    SCALE_INCREMENT=1
                    ENV_SUBDOMAIN="dev"
                  else
                    NAMESPACE="keyteki-$ENV"
                    REPLICAS=1
                    SCALE_INCREMENT=1
                    ENV_SUBDOMAIN="$ENV"
                  fi

                  echo "Deploying to $NAMESPACE ($REPLICAS replicas)"
                  DEPLOY_TIME=$(date +%s)
                  echo "Deployment timestamp: $DEPLOY_TIME"

                  cat > /tmp/node-secrets.yaml <<EOF
                  nodeConfig:
                    env: "production"
                    sentryDsn: "${{ secrets.SENTRY_DSN }}"
                    redisKeyPrefix: "keyteki:$ENV"
                    redisUrl: "${{ secrets.REDIS_URL }}"
                    gameNode:
                      host: null
                      socketioPort: 80
                    secret: "${{ secrets.SECRET }}"
                  EOF

                  helm upgrade --install game-node ./infrastructure/node \
                    --set namespace=$NAMESPACE \
                    --set replicaCount=$REPLICAS \
                    --set updatePartition=$REPLICAS \
                    --set environment=$ENV_SUBDOMAIN \
                    --set deployTime=$DEPLOY_TIME \
                    -f /tmp/node-secrets.yaml \
                    -n $NAMESPACE

                  echo "Starting graceful rollout..."

                  NEW_REPLICAS=$((REPLICAS + SCALE_INCREMENT))
                  echo "Scaling up to $NEW_REPLICAS replicas..."
                  kubectl scale statefulset node -n $NAMESPACE --replicas=$NEW_REPLICAS

                  for i in $(seq $REPLICAS $((NEW_REPLICAS - 1))); do
                    NEW_POD="node-$i"
                    echo "Waiting for $NEW_POD to become ready..."
                    kubectl wait --for=condition=ready pod/$NEW_POD -n $NAMESPACE --timeout=30m
                    echo "$NEW_POD is ready."
                  done

                  OLD_PODS=$(kubectl get pods -n $NAMESPACE -l app=game-node -o jsonpath='{.items[*].metadata.name}' | tr ' ' '\n' | sort | head -n $REPLICAS)
                  echo "Old pods: $OLD_PODS"

                  echo "Sending SIGTERM to all old pods to start draining..."
                  for POD in $OLD_PODS; do
                    echo "Signaling $POD..."
                    kubectl exec -n $NAMESPACE $POD -- /bin/sh -c "kill -TERM \$(pidof node)" || true
                  done

                  echo "Waiting for all old pods to drain..."
                  for POD in $OLD_PODS; do
                    echo "Monitoring $POD..."
                    ATTEMPTS=0
                    while [ $ATTEMPTS -lt 540 ]; do
                      GAMES=$(kubectl exec -n $NAMESPACE $POD -- wget -q -O- http://localhost:9000/health/games 2>/dev/null || echo "error")
                      if [ "$GAMES" = "error" ]; then
                        echo "$POD: Health check failed; assuming node shutting down."
                        break
                      fi
                      if [ "$GAMES" = "0" ]; then
                        echo "$POD: All games finished."
                        break
                      fi
                      echo "$POD: Still running $GAMES games... ($ATTEMPTS/540)"
                      ATTEMPTS=$((ATTEMPTS + 1))
                      sleep 10
                    done
                  done

                  echo "Updating old pods to new version..."
                  helm upgrade --install game-node ./infrastructure/node \
                    --set namespace=$NAMESPACE \
                    --set replicaCount=$NEW_REPLICAS \
                    --set updatePartition=0 \
                    --set environment=$ENV_SUBDOMAIN \
                    --set deployTime=$DEPLOY_TIME \
                    -f /tmp/node-secrets.yaml \
                    -n $NAMESPACE

                  echo "Waiting for rollout to complete..."
                  kubectl rollout status statefulset/node -n $NAMESPACE --timeout=30m

                  rm /tmp/node-secrets.yaml

                  echo "Scaling back down to $REPLICAS replicas..."
                  kubectl scale statefulset node -n $NAMESPACE --replicas=$REPLICAS

                  echo "Graceful rollout complete!"