URL shortener end to end

The URL shortener (such as Bitly or TinyURL) is the quintessential system design interview question. While the problem statement sounds deceptively simple, it tests core competencies: calculating back-of-the-envelope capacity, choosing the right key generation strategy, managing heavy read-write asymmetries, and designing caching layers that withstand traffic spikes.

Here is how to apply the 4-step delivery framework end to end.

1. Requirements

Before writing architecture components, clarify the functional and operational scope with your interviewer:

2. Estimates and scale

State clear assumptions and perform back-of-the-envelope math:

• Write volume: 100 million new URLs created per month, roughly 40 writes per second.
• Read volume: 100 to 1 ratio implies 4,000 redirects per second average; 10,000 requests per second at peak.
• Storage math: Each record holds short code (7 bytes), original URL (2 KB), created timestamp (8 bytes), user ID (8 bytes), totaling roughly 2.5 KB per URL.
• 5-year storage: 100M * 12 months * 5 years = 6 billion URLs * 2.5 KB ≈ 15 TB.
• Cache sizing (80/20 rule): 20% of URLs generate 80% of daily redirects. Daily reads = 4,000 * 86,400 ≈ 345M redirects. Caching 20% requires roughly 170 GB of RAM, easily hosted in a modest Redis cluster.

3. API design and data model

Define compact REST endpoints:

• POST /v1/urls returns {"short_url": "https://sho.rt/aZ89k1q", "created_at": 1718000000}
• GET /:short_code returns HTTP 302 Found with Location: https://example.com/long/path

Data schema stored in a relational database (PostgreSQL) or distributed KV store (DynamoDB):

• id: 64-bit integer primary key.
• short_code: VARCHAR(7) UNIQUE (indexed).
• long_url: VARCHAR(2048).
• created_at: TIMESTAMP.

4. High-level design

Laying out the diagram…

Deep dive 1: Generating unique short codes

How do you turn a long URL into 7 Base62 characters ([0-9, a-z, A-Z])?

To generate unique 64-bit integers without centralized bottlenecks, workers claim blocks of 10,000 IDs from a central token range coordinator (e.g. ZooKeeper or Redis INCRBY) and increment local memory counters.

Deep dive 2: Handling cache stampedes on viral links

When a breaking news link is posted on social media, thousands of concurrent requests arrive simultaneously for a newly created short URL not yet in Redis.

If all 10,000 requests miss the cache at once, they all query the database simultaneously, causing a cache stampede:

• Use single-flight request coalescing: In the application layer, ensure that only one concurrent worker queries the database for a given short_code, while all other waiting threads share that single database response and populate Redis.
• Use probabilistic early expiration (XFetch) or pre-warming to refresh hot keys before their TTL expires.

Quick check

Quick check

Why should analytics click events be written to Kafka rather than directly to the primary relational database during a redirect?

AKafka encrypts clicks automaticallyBSynchronous database writes add latency to the redirect and risk taking down the database during viral click surgesCPostgreSQL does not support countingDRedirects cannot run without Kafka

To build this design step by step, with a load test at the end, open the Bitly lab.

Try it

URL Shortener HLD

Or design it freely in the practice workspace.