GraphQL — Part 2: Architecture, Execution Engine & Production Patterns

What is this?

Back to the à la carte kitchen from Part 1. The diner handed in a card listing exactly what they want. Now we go behind the pass and watch the kitchen actually cook it. The card has to be read, checked against what the kitchen can make ("sorry, we're out of the duck"), turned into a plan, and then cooked station by station — and if the kitchen is sloppy, one table's order can send a runner sprinting to the walk-in fridge fifty times for the same tub of butter.

That sloppiness has a name in GraphQL — the N+1 problem — and taming it with batching is the single most important performance skill in this chapter.

The one-line idea: GraphQL execution is a parse → validate → execute pipeline that walks the query tree calling one resolver per field — fast and flexible, but naively it fires one data fetch per node, which you fix with batching (DataLoader), not with cleverer SQL.