System Design: Scaling to 1 Million Users

Stage 1: The Monolith (0 - 1k Users)

#

One server. Database + App + Web Server all on the same machine.

• Pros: Simple deployment, cheap.
• Cons: Single point of failure.

Stage 2: Vertical Scaling (1k - 10k Users)

#

Upgrade the instance size (t3.micro -> m5.large). This works until it doesn’t. You eventually hit the limit of a single machine.

Stage 3: Horizontal Scaling (10k - 100k Users)

#

This is the big jump.

1. Load Balancer (ELB/Nginx): Distributes traffic across multiple app servers.
2. Stateless Apps: User sessions moved to Redis.
3. Database Separation: Master (Write) / Replica (Read) architecture.

graph TD
    User --> LB[Load Balancer]
    LB --> App1[App Server 1]
    LB --> App2[App Server 2]
    App1 --> Redis[Redis Cache]
    App2 --> Redis
    App1 --> DB_M[DB Master]
    App2 --> DB_R[DB Replica]
    DB_M -->|Replication| DB_R

Stage 4: 1M+ Users

#

Now we need serious optimizations.

• CDN (Cloudflare): Cache static assets at the edge.
• Database Sharding: Split the database by User ID (Users 1-1M on DB1, 1M-2M on DB2).
• Async Workers (Celery/Kafka): Move everything non-essential (email, PDF generation) to background queues.