URL Shortener

Overview

A learning project for distributed systems and system design, built around a short URL service with shared state across multiple application replicas.

Architecture

• Client → NGINX → 3 Flask replicas → Redis + PostgreSQL
• PostgreSQL is the source of truth.
• Redis holds cache and rate-limit state.

Key technical decisions

• Base62 short codes keep generated URLs compact.
• Cache-aside reads use Redis before falling back to PostgreSQL.
• NGINX provides load balancing across the Flask replicas.
• Docker service discovery keeps service-to-service connections independent of fixed container addresses.

Local k6 observations

Local Docker Desktop measurements, not production benchmarks:

• 18,457 requests at approximately 612.55 requests per second.
• p95 latency of approximately 31.68 ms.
• 0% failed requests.
• All 3 of 3 Flask replicas were observed handling requests.

Rate-limit finding

Sequential requests produced 10 × 201 responses and 2 × 429 responses. Concurrent requests produced 13 × 201 responses and 7 × 429 responses. The full GET → decision → INCR flow is not atomic, exposing a race condition under concurrency.

Lessons learned

• Cache-aside reads and cache invalidation boundaries.
• Shared state across replicas.
• Reverse proxies and load balancing.
• Docker service discovery.
• Load testing can expose correctness and concurrency issues, not only latency.