Design a URL shortener

Question 1

Who calls this system?

Accepted Answer

Every design starts with the caller. Without one on the board there is no read path to reason about, and no place for the latency budget to start.

Question 2

What does the client actually connect to?

Accepted Answer

A gateway or load balancer is where TLS terminates, where rate limiting lives, and what lets you run more than one instance of anything behind it.

Question 3

Reads outnumber writes 100 to 1. Where do you absorb them?

Accepted Answer

The redirect path is the hot path: the same small rows, read constantly, changing almost never. That is the textbook case for a cache in front of the store — and without one, your database is sized for 19,000 reads a second it should never have seen.

Question 4

Where do the links themselves live?

Accepted Answer

The mapping has to survive a restart. A cache is not a store; it is a copy of one.

Question 5

Two people shorten a URL at the same instant. Who decides the short code?

Accepted Answer

Short codes have to be unique across every instance at once. Hashing collides, an auto-increment column serialises your writes — so this is usually its own concern: pre-issued key ranges, or a coordination service handing out blocks.

Question 6

Your client talks straight to the store. Who enforces anything?

Accepted Answer

A client with a direct line to the database is a client that can read rows it should not, ignore your rate limits, and pin your connection pool. Put a service in between — that service is where the rules live.

Question 7

Can a request actually get from the client to the store?

Accepted Answer

Boxes on a board are not a system until they are connected. If the store is unreachable from the caller, the design describes something that cannot serve a request.

Question 8

How many writes a second is 500M new URLs a month?

Accepted Answer

500,000,000 / 30 days / 86,400 seconds is about 193 writes a second. That is a small number, and knowing it is small is the point — it tells you the write path is not what needs the engineering.

Question 9

And how many reads a second?

Accepted Answer

At 100:1 against 193 writes a second, that is roughly 19,300 reads a second. This is the number that sizes the cache, the number of read replicas, and the whole redirect path.

Blueprint — system design