Deep Learnings Migrating legacy services to Federated GraphQL
summary
Booking.com replaces a single GraphQL proxy Adam proxy with a composed layer using Federation via Apollo, migrating legacy Perl services and new Java/Node subgraphs via traffic splits (10 → 30 → 50 → 100%) and an experimentation kill switch. Built a schema-publishing service, CI workflow, and scripts to add GraphQL and a semi-custom Java graphql library. Adoption rises from one subgraph every few weeks to ~one per week, traffic reaches billions/day with ~600% seasonal growth and a 10% latency overhead reduction.
AI-written summary · about
chapters (10)
- 0:00Modernizing legacy systems
- 1:08Data mediation layer (Atom)
- 4:06Transitioning to Federation
- 6:45Rolling out the federation
- 10:06Deconstructing Atom
- 13:23Tooling for adoption
- 17:12Scaling and performance
- 21:37Multi-cloud infrastructure
- 23:15Future goals and roadmap
- 24:22Q&A and closing
related talks
35mA Wild GraphQL Rollercoaster Ride a Federated GraphQL Adoption Story
06mEscape the GraphQL Monolith
23mBuilding the Graph at Booking.com
38mGraphQL Federation in the Wild
17mThe year ahead for GraphQL Federation: Combining APIs for the AI era.
26mKeynote: Apollo's Journey with GraphQL: Transforming Enterprise APIs for the Future
36mGraphQL Federation - The @key to Your Platform API Strategy
23mBuilding the Graph at Booking.com