Safe Network API Migration Tactics
summary
Uber migrates high‑QPS JSON HTTP endpoints to gRPC with a requirement of zero downtime and QPS increase under 2% and uses Retrofit/OKHttp, gRPC over HTTP3, GCP/Cloudflare. Migration builds shadow calls at the gateway with an async cache to compare requests and responses, a client‑side transformer, and load tests, yielding 45% payload reduction and roughly a 30% latency drop. Rollout uses gradual percentage exposure, client randomization (50/50 split) and a local rollback library that flips after three errors per device.
AI-written summary · about
chapters (14)
- 0:00API migration challenges
- 2:36HTTP to gRPC migration
- 4:00Three phases of safe migration
- 5:05Pre-flight with shadow calls
- 9:42Latency and payload reduction
- 12:52Cloud provider incompatibilities
- 14:01Timeout and API behavior
- 16:31Shadow call results
- 17:13Inflight gradual rollout
- 18:38Circuit breakers and randomization
- 20:47Local rollback
- 23:38Cleanup and API retirement
- 25:05Production vs shadow results
- 26:30Summary
related talks
40mgRPC in Apps: A Migration Story
29mThe Story of Why We Migrate to gRPC and How We Go About It
32mDeploying HTTP/3 at Scale on Mobile: A 17% Startup Latency Gain Despite Implementation Constraints
17mBoost API Performance: gRPC vs REST vs Async
37mgRPC Intro
31mRPC Frameworks Overview
46mA Different Kind of Networking with gRPC and Protocol Buffers
39mResilient REST APIs: The case for Parallel Change