Duolingo Backend Architecture

How to make a server request fast and hard to break — and how to move a live system to a new shape without stopping it.

The through-line across every node: latency adds, availability multiplies, and hold time is the variable you control. Most of the wins below come from deleting work from the live path rather than making that work faster. The rest come from being honest about what a fix costs before you ship it.

Scope note: this skill owns latency and failure modes. Seeing, attributing and governing infrastructure spend is a different discipline and belongs to the cost-efficiency skill in this pack.

The request path