Rethinking GraphQL Execution

Session description

To ease life for the roughly 1,000 developers who contribute regularly to our GraphQL interface, Airbnb has a highly opinionated developer API that we believe eases life for both developers and the operators who maintain the service. In the past, this API was implemented on top of a traditional, specification-based GraphQL engine, which supported agility as we evolved our opinionated approach to resolvers. As that approach matured, we saw opportunities to build a more efficient engine to support them. This talk describes our new GraphQL engine. Key elements in our design include: - Refactoring query execution into distinct resolution and completion phases. - A query planner that optimizes and orchestrates the execution of those phases. - A new data structure that is the intermediary between the two phases, a data structure that allows for principled communication between the many resolvers involved in executing a query