Architecting for Zero-Downtime Deployment

Architecting for Zero-Downtime Deployment

In this lesson, you will learn how to implement connection draining to migrate WebSocket clients gracefully during server restarts, ensuring your users never experience a dropped connection.

When you deploy a new version of a Node.js application, simply killing the old process terminates every open WebSocket connection instantly. This is disruptive. To achieve zero-downtime, we must adopt a "Drain and Shift" strategy: we signal the server to stop accepting new connections while allowing existing ones to finish their work or migrate to a new instance.

Implementing the Drain Pattern

To execute this, your server needs to transition into a "draining" state upon receiving a shutdown signal (like SIGTERM). Here is how you can manage this lifecycle:

let isDraining = false;

// Stop accepting new traffic
process.on('SIGTERM', () => {
  isDraining = true;
  
  // 1. Tell your Load Balancer to stop sending new clients here
  // 2. Broadcast a "reconnect" signal to all connected clients
  wss.clients.forEach(socket => {
    socket.send(JSON.stringify({ type: 'RECONNECT_ADVISORY', delay: 5000 }));
  });

  // 3. Gracefully close connections after a short buffer
  setTimeout(() => {
    wss.close(() => process.exit(0));
  }, 5000);
});

// Middleware to reject new connections
server.on('connection', (socket) => {
  if (isDraining) {
    socket.close(1001, 'Server is restarting');
    return;
  }
  // handle normal connection logic...
});

The Workflow

1. Signal: The process receives SIGTERM from your orchestrator (like Kubernetes).
2. Flag: Set isDraining = true. Your load balancer should detect that this instance is no longer "ready" and redirect incoming requests to a newly spawned instance.
3. Notify: Broadcast a message to connected clients suggesting they reconnect within a staggered window to avoid a "thundering herd" effect on your database or authentication service.
4. Clean Exit: Once the clients have had time to move to the new instance, close the remaining sockets and shut down the server process.

By decoupling the process shutdown from the client's session lifecycle, you ensure the user experience remains seamless even during continuous delivery.

Key Takeaway: Zero-downtime deployment relies on signaling existing clients to reconnect to new instances while blocking new incoming connections, allowing the old process to clear its state before exiting.