DEV Community

Barberprostudio
Barberprostudio

Posted on

Building a Scalable Booking System: From Manual Scheduling to 24/7 Automation

#architecture #backend #laravel #performance

Building a Scalable Booking System: From Manual Scheduling to 24/7 Automation

The Problem

Last year, I was hired to build a booking system for a growing network of barbershops in Portugal. The initial requirements seemed straightforward: let customers book appointments online. But as the system scaled from 5 shops to 50+, we encountered challenges that pushed us to architect a much more complex solution than anticipated.

Initial Architecture: The MVP

We started with a traditional request-response model using Laravel and PostgreSQL:

Client → API → Database → Email Queue → Email Service
Enter fullscreen mode Exit fullscreen mode

Simple. Worked for small traffic. Fell apart under load.

The problem: every booking required:

  • Database write
  • Email notification to barber
  • Email confirmation to client
  • Calendar sync to 3 different integrations
  • SMS reminder scheduling

All synchronously. A single slow operation blocked the entire request.

The Scale Problem

By month 6, we had:

  • 50 barbershops
  • 3,000 daily bookings
  • Response times degrading from 200ms to 2-3 seconds
  • Customers receiving confirmations 30+ seconds after booking

The synchronous model wasn't viable anymore.

The Solution: Async Everything

We restructured the system using RabbitMQ for message queuing:

Client → API (Fast) ↓
         Queue → Workers (Parallel Processing)
              ├→ Email Worker
              ├→ SMS Worker
              ├→ Calendar Sync Worker
              ├→ Analytics Worker
              └→ Notification Worker
Enter fullscreen mode Exit fullscreen mode

Result: API response time dropped from 2.5s back to 180ms. Customers got instant confirmation.

Key Technical Decisions

1. Database Design for Concurrency

We implemented optimistic locking for appointment slots:

CREATE TABLE appointments (
    id BIGINT PRIMARY KEY,
    barber_id BIGINT,
    time_slot_id BIGINT,
    customer_id BIGINT,
    version INT DEFAULT 1,
    status ENUM('pending', 'confirmed', 'completed', 'cancelled'),
    created_at TIMESTAMP,
    updated_at TIMESTAMP,
    UNIQUE KEY(time_slot_id, version)
);
Enter fullscreen mode Exit fullscreen mode

Why? Double-booking prevention without expensive locks. The version column ensures atomic updates only succeed if no other process modified the record.

2. Handling Timezone Complexity

One issue that surprised us: international clients booking across timezones.

// Store everything in UTC internally
$appointment = Appointment::create([
    'scheduled_at' => $request->scheduled_at->setTimezone('UTC'),
    'barber_timezone' => $barber->timezone,
    'customer_timezone' => Auth::user()->timezone,
]);

// Return in client's timezone
return $appointment->scheduled_at
    ->setTimezone($customer->timezone)
    ->format('Y-m-d H:i');
Enter fullscreen mode Exit fullscreen mode

3. Queue Reliability

We learned the hard way: message queues need dead-letter handling.

// Retry failed jobs with exponential backoff
Queue::job(SendConfirmationEmail::class)
    ->tries(5)
    ->backoff([1, 5, 10, 30, 60]) // seconds
    ->retryUntil(now()->addHours(24))
    ->dispatch($appointment);
Enter fullscreen mode Exit fullscreen mode

Failed emails went to a dead-letter queue for manual review instead of silently failing.

Performance Metrics

After optimization:

Metric Before After Change
API Response Time 2.5s 180ms 93% faster
Booking Success Rate 94% 99.7% Double-booking eliminated
Customer Satisfaction 3.2/5 4.7/5 Faster confirmations
Concurrent Bookings 5/second 150/second 30x capacity

What We'd Do Differently

  1. Start with async from day 1 - We rebuilt the entire queue system mid-growth. Painful.
  2. Implement monitoring earlier - We discovered bottlenecks through customer complaints, not dashboards.
  3. Test edge cases more - 11 PM bookings, midnight hour transitions, daylight saving time were nightmares to debug in production.

The Lesson

A booking system is a concurrency problem disguised as a scheduling problem. The business logic is simple. The engineering is not.

The moment you have:

  • Multiple users competing for the same resource (appointment slots)
  • Time-sensitive operations (confirmations must be instant)
  • Distributed workers (multiple integrations)

...you need a system designed for concurrency, not convenience.

Have you built distributed systems with similar challenges? Share your approach in the comments — especially if you solved the timezone problem better than we did!

Top comments (0)