Every retry-safe design depends on idempotency, and most candidates cannot say how it actually works. Here is the full mechanism: who generates the key, what its scope is, where the record lives, how to make the check atomic, what to return on a replay, and how long to keep it.

The reference page for system design estimation: powers of two, time conversions, the latency ladder, throughput per node, sizes of common things, cloud cost ballparks, availability math, and the rules of thumb that turn a number into a design decision.

The job scheduler question is about delivery guarantees, leases, and what happens when the scheduler itself goes down. Here is what interviewers are probing, the gotchas around double execution and missed runs, and a structure that survives the follow-ups.

The failure question is a test of ownership, judgment, and whether you changed how you decide. Here is how to pick the right story, structure it in six beats, and handle the follow-ups that catch people blaming the team.

Everyone can name token bucket. The rate limiter interview question is really about atomicity in a distributed store, what happens when that store is down, who the client actually is, and the contract you give callers. Here is how to answer it like an architect.

The URL shortener question looks trivial, which is exactly why interviewers use it. Here is what they are really probing, the gotchas that separate senior candidates from junior ones, and a structure for an answer that holds up under follow-ups.

The monolith vs. microservices question is a trap for candidates who answer with a slogan. Here is what interviewers are actually probing, the gotchas, and a structure for an answer that sounds like an architect.

A study log for preparing for a technical software architecture role: one interview question or subject per post, the gotchas behind it, and how to answer it intelligently.