FastAPI High-Performance Social Media API

A robust, production-ready backend for a social media platform built with FastAPI, PostgreSQL, and Docker.

This project focuses on high-concurrency optimization, successfully scaling to handle 1,000 concurrent users.

---

Setup & Installation

1. Prerequisites

• Python 3.10+
• Docker Desktop (Recommended)
  OR
• PostgreSQL 15+

---

2. Environment Configuration

Create a .env file in the root directory:

DATABASE_HOSTNAME=db
DATABASE_PORT=5432
DATABASE_PASSWORD=your_password
DATABASE_NAME=fastapi
DATABASE_USERNAME=postgres

SECRET_KEY=your_secret_key_here
ALGORITHM=HS256
ACCESS_TOKEN_EXPIRE_MINUTES=30

---

3. Running with Docker (Recommended)

This method is optimized for high concurrency and avoids Windows networking limitations.

# 1. Build and start containers
docker-compose up --build -d

# 2. Run database migrations
docker-compose exec app alembic upgrade head

# 3. Create a test user
curl -X POST "http://localhost:8000/users/" \
-H "Content-Type: application/json" \
-d "{\"email\": \"hello124@gmail.com\", \"password\": \"password123\"}"

# 4. Verify login
curl -X POST "http://localhost:8000/login" \
-H "Content-Type: application/x-www-form-urlencoded" \
-d "username=hello124@gmail.com&password=password123"

---

4. Running Locally (venv)

# Create virtual environment
python -m venv venv

# Activate it
source venv/bin/activate
# Windows:
# venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Run the app
uvicorn app.main:app --reload

---

The Testing Journey: Scaling to 1,000 Users

As part of an engineering exploration at IIT Guwahati, this API was stress-tested using Locust to uncover real-world bottlenecks.

---

The Struggles & Bottlenecks

• 64 Connection Limit (Windows)
  Encountered:

  ValueError: too many file descriptors in select()
  

  Fixed by moving to Docker (Linux epoll)

• Connection Starvation
  Default SQLAlchemy pool size (5) caused 100% failures at 200 users

• CPU Saturation
  Argon2 password hashing maxed CPU → 34% failure rate at 1,000 users

---

The Breakthrough (12% Failure Rate)

Optimized system to handle 1,000 users (spawn rate = 2) with only 12% failures:

• Engine Tuning

  pool_size=50
  max_overflow=100

• PostgreSQL Scaling

  max_connections = 300
  

• Multi-Worker Scaling

  • 4 Uvicorn workers (Docker)

• Data Hygiene

  • reset_db.py script to truncate tables between tests

---

Testing Suite

Unit & Integration Testing

pytest

---

Load Testing

# Start Locust
locust -f locustfile.py

---

Future Roadmap: Path to 0% Failures

To push beyond current limits and scale to 5,000+ users:

Redis Caching

• Cache GET /posts/
• Store JWT user objects in memory
  Reduce DB load by ~80%

---

NGINX Load Balancing

• Reverse proxy
• SSL termination
• Smarter request distribution

---

Cloud Deployment

• AWS (ECS + RDS) or Render
• Public production endpoint

---