Design a Paste Bin

Design a web service like Pastebin or GitHub Gists where users can store and share plain text or code snippets via a unique URL. The system separates lightweight metadata (stored in a relational database) from potentially large paste content (stored in object storage like S3), making it an excellent study in storage-tier separation.

Scale Estimates

Metric	Value
New pastes created	5 million / month (~2 per second)
Paste reads	25 million / month (~10 per second)
Read-to-write ratio	5:1
Average paste size	10 KB
Max paste size	10 MB
Metadata per paste	~500 bytes
Content storage for 5 years	5M × 12 × 5 × 10KB ≈ 3 TB (S3)
Metadata storage for 5 years	5M × 12 × 5 × 500B ≈ 150 GB (DB)
Unique paste keys (8 chars Base62)	62⁸ ≈ 218 trillion

Non-Functional Requirements

• Low latency: Retrieve pastes in < 100ms (p99); CDN-cached public pastes in < 20ms
• High availability: 99.9% uptime; reads should always work (eventual consistency acceptable)
• Durability: Paste content must not be lost; S3 provides 99.999999999% (11 9's) durability
• Scalability: Object storage (S3) scales infinitely for content; DB scales via read replicas + sharding
• Security: Private pastes accessible only by owner; content scanning for malware/spam
• Cost-efficient: S3 is ~$0.023/GB/month — 3 TB costs ~$70/month; separate hot (Redis) vs cold (S3) tiers