200 Is Not OK: Strategies for Tracing Partial Responses with GraphQL Observability
Tuesday, September 10 / 10:50a.m. PDT - 11:20a.m. PDT
GraphQL is agnostic to the transport layer. Almost all out of the box observability tooling is tailored to REST/HTTP APIs. Major observability challenge with GraphQL over HTTP is the support of partial response and HTTP status with 2XX may have failed. GraphQL Gateway generally fans-out one incoming request to one or many outgoing subgraph requests. It is essential to monitor each subgraph and the GraphQL gateway’s metrics and also to trace a single request across the network stack. This talk will cover how Intuit is reducing Mean Time to Detect (MTTD) and Mean Time to Recover (MTTR) for GraphQL APIs, by capturing Failed Customer Interactions, Golden Signal to determine gateway and subgraph’s health, latency, error rate and other related metrics. This talk will also cover open tracing and logging for GraphQL APIs.