$12 Hug Test 🥊

Can a $12/month VPS survive the Reddit Hug of Death?

This repo documents load-testing experiments on a $12 Linode box (1 CPU, 2 GB RAM, 50 GB storage) running FastAPI + SQLite.

We wanted to know: what happens if my indie side-project actually hits the Reddit front page?

🔧 Setup

• Server: Linode, 1 vCPU / 2 GB RAM / 50 GB storage
• Stack: FastAPI, SQLite (FTS5), Gunicorn + Uvicorn workers, Nginx (TLS via Certbot)
• Rules: no upgrades, no Postgres, no Kubernetes. Just the $12 box.

🐹 The App

Minimal journal app (MicroJournal):

• / → homepage (read entries)
• /search?q=term → full-text search
• /add → add entry (redirect)
• /add_raw → raw write endpoint (used for load tests)

🚦 Load Testing

We used bombardier to simulate traffic spikes:

• Page 4 Reddit (~5k/day): ~1 RPS
• Page 2 Reddit (~50k/day): ~10 RPS
• Page 1 Reddit (~500k/day): ~100 RPS (≈1k–2k simultaneous users)

Test cases:

• Read (/)
• Search (/search)
• Write (/add_raw)
• Mixed (90% reads, 10% writes)

Logs: see /logs (sanitized).

🐛 Bottlenecks Found

• /search with naive LIKE pegged CPU at 100%
• Concurrent writes caused database locked

✅ Fixes

• FTS5 full-text search → efficient indexing
• WAL mode → concurrent reads + writes
• Nginx micro-caching (1–5s TTL) → absorbed bursts

📊 Results

• Reads: stable at 100 RPS (~140ms latency, ~60% CPU)
• Search: from 5xx → clean 200s with FTS5
• Writes: from "DB locked" → stable 204s with WAL