Zero-downtime deployments with rolling restarts and health checks

Deploying new code requires restarting the server, causing a window where requests fail or get dropped.

Implement a deploy script that validates before switching:

#!/bin/bash
set -e  # Exit on any error

APP_DIR=/opt/myapp
NEW_DIR=/opt/myapp-new

# 1. Pull and build new version in separate directory
git clone $REPO $NEW_DIR
cd $NEW_DIR && npm ci && npm run build

# 2. Validate the build starts correctly
NODE_ENV=production timeout 10 node $NEW_DIR/dist/server.js --check 2>&1 | grep 'OK'

# 3. Copy new files and reload (pm2 cluster reload is atomic)
cp -r $NEW_DIR/dist $APP_DIR/
pm2 reload myapp

# 4. Wait and verify health
sleep 5
curl -f http://localhost:3000/health || (pm2 reload myapp --revert && exit 1)

# 5. Cleanup
rm -rf $NEW_DIR
echo 'Deploy complete'

pm2 reload in cluster mode sends SIGINT to one worker, waits for it to finish serving requests, starts a new worker, then proceeds to the next — achieving zero-downtime rolling restart.