AI & Technical question

You're the CTO of a company. Design a program that provides unique IDs upon requests from a client. This program will be used by Facebook and Google and needs to scale.

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Distributed systems design for a unique ID generator at massive scale, a classic system design problem.

Clarify requirements: IDs must be unique, roughly sortable by time for indexing, and generated at very high throughput with low latency.
Rule out a single centralized counter, since it becomes a bottleneck and single point of failure at this scale.
Propose a Snowflake-style approach: each ID combines a timestamp, a worker ID, and a per-machine sequence number, generated independently per node.
Explain why this scales: no cross-node communication is needed per request, so throughput scales linearly with node count.
Address clock skew risk: handle cases where a node's clock moves backward, which can cause collisions if unhandled.
Note operational needs: worker ID assignment and monitoring for clock drift across the fleet.

Correctly rules out a centralized counter as unscalable and names the real distributed pattern used at this scale.
Explains the mechanism precisely, timestamp plus worker ID plus sequence, rather than vaguely saying "generate random IDs."
Names clock skew as a specific, real failure mode and proposes handling it, showing depth beyond the happy path.
Addresses worker ID assignment as an operational detail, not just the ID format.

Proposing a single centralized ID generator, which cannot scale to this throughput.
Ignoring clock skew and node coordination edge cases.
Not addressing how worker IDs get assigned without collision.

How would you handle a node's clock moving backward?
How would you assign and manage worker IDs across a large fleet?
What happens if two clients request an ID for the same logical event simultaneously?