Interservice Protocol Selection REST gRPC GraphQL
Use When
- Service APIs need protocol standardization
- Existing protocol mix causes complexity or performance problems
Workflow
- Classify interactions by sync/async, payload size, and fan-out.
- Choose protocol per use case with explicit rationale.
- Define schema contracts, error models, and compatibility rules.
- Implement observability and resiliency controls per protocol.
- Validate developer ergonomics and operational overhead.
Required Checks
- Protocol choice aligns with latency and operability targets
- Schema governance includes backward compatibility policy
- Gateway translation overhead is understood and bounded
Decision Framework
- Prefer REST for public, cache-friendly CRUD; gRPC for low-latency internal sync; GraphQL when consumers need shaped reads with strict query cost limits.
- Document error models, schema evolution rules, and breaking-change policy per protocol.
- Measure gateway translation overhead when mixing protocols at the edge.
- Avoid protocol proliferation without governance—standardize defaults per interaction class.
Common Rationalizations And Rebuttals
- "GraphQL everywhere simplifies clients." -> Unbounded queries and N+1 resolver paths hurt reliability; use where fan-out shaping pays off.
- "gRPC is always faster." -> Operational complexity (schema, proxies, browser incompatibility) has cost; use internally unless justified externally.
- "REST versioning in URLs is enough." -> Explicit compatibility tests and deprecation timelines still required.
Evidence Pack
- Protocol decision record per interaction type with latency and operability rationale
- Schema governance doc with backward-compatibility rules
- Load or latency comparison when gateway translates between protocols
- Observability config per protocol (trace propagation, error taxonomy)
Exit Criteria
- Protocol usage is intentional, documented, and maintainable
- Service communication failures are easier to debug and control