Pastebin

Problem statement (interviewer prompt)

Design a Pastebin: anyone can paste text, get a short URL, and others can read it. Support expiry, password-protect, syntax highlighting, edit/delete by owner, and view counters. Optimise for read-heavy (10:1) and 1M new pastes/day.

flowchart LR
  U([User])
  W([Web / API])
  ID([ID Generator])
  OBJ[(Object Storage<br/>paste body)]
  META[(Metadata DB<br/>id, owner, expiry)]
  CDN[CDN]
  U -->|POST /paste| W
  W --> ID
  W --> OBJ
  W --> META
  U -->|GET /:id| CDN --> W
  W --> META
  W --> OBJ

    classDef client fill:#dbeafe,stroke:#1e40af,stroke-width:1px,color:#0f172a;
    classDef edge fill:#cffafe,stroke:#0e7490,stroke-width:1px,color:#0f172a;
    classDef service fill:#fef3c7,stroke:#92400e,stroke-width:1px,color:#0f172a;
    classDef datastore fill:#fee2e2,stroke:#991b1b,stroke-width:1px,color:#0f172a;
    classDef cache fill:#fed7aa,stroke:#9a3412,stroke-width:1px,color:#0f172a;
    classDef queue fill:#ede9fe,stroke:#5b21b6,stroke-width:1px,color:#0f172a;
    classDef compute fill:#d1fae5,stroke:#065f46,stroke-width:1px,color:#0f172a;
    classDef storage fill:#e5e7eb,stroke:#374151,stroke-width:1px,color:#0f172a;
    classDef external fill:#fce7f3,stroke:#9d174d,stroke-width:1px,color:#0f172a;
    classDef obs fill:#f3e8ff,stroke:#6b21a8,stroke-width:1px,color:#0f172a;
    class U,W client;
    class CDN edge;
    class META datastore;
    class ID compute;
    class OBJ storage;

flowchart TB
  subgraph Client
    BR([Browser])
    CLI[CLI / API]
  end

  subgraph Edge
    DNS[DNS]
    CDN[CDN / Edge cache]
    LB[L7 LB]
    WAF[WAF + abuse]
  end

  subgraph App[App Tier]
    GW[API Gateway]
    CRT[Create Service]
    READ[Read Service]
    DEL[Delete / Expire]
    SYN[Syntax highlighter<br/>server-side]
    AUTH[Auth - opt]
  end

  subgraph ID[ID generation]
    SF[Snowflake / KGS]
    B62[Base62 encode<br/>8 chars]
  end

  subgraph Storage
    META[(Metadata SQL<br/>paste id, owner, expiry, lang)]
    BLOB[(Object store S3 / GCS<br/>raw paste body)]
    CACHE[(Redis hot pastes)]
  end

  subgraph Async
    Q[[Kafka events<br/>create / view / delete]]
    SC([Expiry sweeper<br/>cron])
    ANL[Analytics<br/>views, languages]
    AB([Abuse scanner<br/>secret detection])
  end

  subgraph Obs
    M[Metrics] 
    L[Logs] 
    T[Traces]
  end

  BR --> DNS --> CDN --> LB --> WAF --> GW
  CLI --> DNS
  GW --> CRT
  GW --> READ
  GW --> DEL
  CRT --> SF --> B62
  CRT --> META
  CRT --> BLOB
  READ --> CACHE
  CACHE -. miss .-> BLOB
  READ --> META
  READ --> SYN
  DEL --> META
  DEL --> BLOB
  SC -.expire.-> META
  SC -.delete.-> BLOB
  CRT --> Q
  Q --> ANL
  Q --> AB
  AB -.flag.-> META
  App -.metrics.-> M

    classDef client fill:#dbeafe,stroke:#1e40af,stroke-width:1px,color:#0f172a;
    classDef edge fill:#cffafe,stroke:#0e7490,stroke-width:1px,color:#0f172a;
    classDef service fill:#fef3c7,stroke:#92400e,stroke-width:1px,color:#0f172a;
    classDef datastore fill:#fee2e2,stroke:#991b1b,stroke-width:1px,color:#0f172a;
    classDef cache fill:#fed7aa,stroke:#9a3412,stroke-width:1px,color:#0f172a;
    classDef queue fill:#ede9fe,stroke:#5b21b6,stroke-width:1px,color:#0f172a;
    classDef compute fill:#d1fae5,stroke:#065f46,stroke-width:1px,color:#0f172a;
    classDef storage fill:#e5e7eb,stroke:#374151,stroke-width:1px,color:#0f172a;
    classDef external fill:#fce7f3,stroke:#9d174d,stroke-width:1px,color:#0f172a;
    classDef obs fill:#f3e8ff,stroke:#6b21a8,stroke-width:1px,color:#0f172a;
    class BR client;
    class DNS,CDN,LB,WAF,GW edge;
    class CLI,CRT,READ,DEL,SYN,AUTH,B62,ANL service;
    class SF,META datastore;
    class CACHE cache;
    class Q queue;
    class SC,AB compute;
    class BLOB storage;
    class M,L,T obs;

Storage choice

• Body in object store (S3) - keyed by paste id; cheap, durable.
• Metadata in SQL (or DynamoDB) for indexing by owner, recent, expiry.
• Hot pastes cached in Redis as id → body (TTL 1 hr).

Expiry

• Either: (a) explicit cron sweeping expires_at < now, (b) S3 lifecycle policy delete.
• Soft-delete then GC for undo window.

Privacy modes

• Public / Unlisted / Private (password-protected).
• For private: store paste encrypted with key derived from URL fragment (zero-knowledge - server can't read).

Anti-abuse

• Scan content for API keys, PII; rate-limit per IP.
• Captcha + honeypot for anonymous create.

Glossary & fundamentals

Concepts referenced in this design. Each row links to its canonical page; the tag column shows whether it is a high-level (HLD) or low-level (LLD) concept.

Tag	Concept	What it is	Page
HLD	Load balancer / GSLB	L4/L7 traffic distribution and failover	load-balancer
HLD	CDN	edge caching for static assets	cdn
HLD	API gateway / BFF	single ingress, auth, rate limit, routing	api-gateway
HLD	Pub/Sub & message brokers	topics, consumer groups, delivery semantics	pub-sub-pattern
HLD	Observability	metrics, logs, traces, SLOs	observability
HLD	Search internals	inverted index, BM25, embeddings, ANN	search-internals

Quick reference

Functional

• Anonymous or authenticated paste creation.
• Short URL → text body retrieval.
• Optional: expiry, password, language tag, view counter.
• Edit / delete by owner.

Non-functional

• Read-heavy (10:1).
• p99 read < 200 ms.
• Durable: never lose body once written.

Capacity

• 1M new pastes/day → 12/s avg.
• 10M reads/day → 120/s avg.
• Avg paste 10 KB → 10 GB/day → ~3.6 TB/year.
• Hot working set 1 day: ~10 GB cache.

API

POST /v1/paste {text, lang, expiry, password?} -> {id, url}
GET  /v1/paste/{id}                            -> {text, lang, ...}
DELETE /v1/paste/{id}                          (owner / token)

Schema

pastes(id PK, owner_id, lang, expires_at, created_at, size, sha256, abuse_flag)

Body lives in S3 at s3://bucket/p/<id>.

Trade-offs

• Store body in DB (simple) vs S3 (cheap at scale). S3 wins beyond a few GB.
• Single-region simple; multi-region needs S3 replication + DB replication.
• Strong vs eventual read after write: read-after-write critical for newly created pastes - pin first read to same DC.
• Encryption client-side prevents server abuse but no server-side search.

Refs

• pastebin.com architecture posts, GitHub Gist design, ByteByteGo "Design Pastebin" video, Alex Xu Vol 1.

FAQ

How do you generate paste URLs?

Generate a base62 key from a random source or counter, check for collision in the metadata DB, and retry on conflict. Six to eight chars give ample address space.

Where do you store the paste body?

Object storage like S3 keeps unbounded text cheap, while a relational or KV metadata DB holds id, owner, expiry, and password hash. CDN caches the body for read scale.

How does TTL expiry work for pastes?

Store an expiresAt timestamp on the metadata row and let either lazy deletion on read or a daily janitor reclaim object storage and remove expired rows.

How do you scale a read-heavy paste service?

Front the GET endpoint with a CDN, cache hot pastes in Redis, shard the metadata DB by paste ID, and serve the body straight from object storage where possible.

How are password-protected pastes implemented?

Store a bcrypt hash on the metadata row. On read the user submits the password, the API verifies the hash, and only then returns the body from object storage.

Related Topics

• Object Storage: blob and object stores back the raw paste content at scale
• Caching Strategies: read-aside and CDN caching reduce latency for popular pastes
• CDN: edge delivery for static paste content worldwide

Video walkthrough

Design PasteBin (Systems Design Questions 3.0) : via Jordan has no life (Ex-Google SWE)