Schema Scoring: Ensuring Schema Excellence in GraphQL
summary
Booking.com builds an automated schema-quality system that scores four concrete metrics — nullability blast radius, stale fields (unused in last 30 days), linting (GraphQL ESLint with internal patches), and cyclical types — using production usage from Apollo and a central schema-publish pipeline. Scores are 1–100, combined via a weighted harmonic mean (linting + stale fields get ~35% each); warnings = 1 pt, errors = 2 pt. A dashboard and MR bot report exist; initial run across ~120–150 subgraphs shows average ~D, cyclical types A++, nullability A+, and stale fields failing badly.
AI-written summary · about
chapters (12)
- 0:00Introduction to booking.com
- 0:43Migration to federation
- 2:02Maintaining schema quality
- 3:29Designing automated scoring
- 5:20Key schema metrics
- 7:51Calculation methodology
- 9:41Final scoring and grading
- 11:33Initial assessment results
- 13:19Dashboard and observability
- 14:13Drawbacks and future roadmap
- 18:22Potential use of AI
- 19:10Q&A session
related talks
24mBrute Force Correctness
01h 17mSafely Evolve Your (Federated) GraphQL Schema with GraphQL Hive
23mBuilding the Graph at Booking.com
31mAccelerating GraphQL API Development with Schema Storming and Code-first Implementation
28mPanel Discussion: The GraphQL Production Roundtable
16mKeynote: GraphQL’s Next Chapter: Progress, Proposals, and Participation
22m8 Years of GraphQL: Unraveling the Trade-Offs
23mBuilding the Graph at Booking.com