The API said 200 OK and sent zero flags

transcript

The API said 200 OK. And sent zero flags. It did more damage than an outage would have.

A bad deploy. For 11 minutes, the config API returns an empty ruleset. SDKs trust it. They replace their rules with nothing. Every flag falls back to its default. Everywhere. One customer's kill switch reverts. A broken payment flow comes back on.

The candidate spots it. If the API had been down, SDKs would keep their rules. A successful, empty answer did the damage.

So the candidate writes a requirement before any design. An SDK never replaces good rules with worse ones. Not empty, invalid or older.

Replay it. The SDK holds version 9013 and asks for changes. The bad deploy answers: 200 OK, zero flags. Before applying, it checks checksum, schema, version, and sudden emptiness. Rejected. It keeps 9013 and keeps evaluating. The kill switch holds.

The interviewer asks: what if a customer really deletes every flag? Then it's an explicit, signed operation. Never inferred from missing flags.

Down, and SDKs keep their rules. A 200, and they obey it. So never trust a successful bad response.

Full breakdown in the lecture.