Microservices Testing

Challenge

• 50+ services interdependent.
• Each team owns service.
• Versioning hell.
• Test environment expensive (run all 50 services?).

Test layers

• Unit — within service.
• Component — single service in isolation.
• Contract — consumer-provider agreement.
• Integration — couple service + DB.
• End-to-end — full system (sparingly).

Honeycomb model

Spotify model: focus on Integrated tests (cross-service), less Implementation Detail tests.

Consumer-driven contract (CDC)

• Consumer (frontend) writes contract.
• Producer (backend) verifies in CI.
• Backend break contract → consumer CI fail.
• Pact Broker stores contracts.

Service virtualization

Mock downstream service in test.

• Hoverfly, WireMock, Mountebank.
• Record real interactions → replay.

End-to-end in microservice

• Expensive: spin up 20 services.
• Use shared staging environment.
• Limit to 10-20 critical journeys.
• Run nightly, not per commit.

Versioning strategy

• Semantic versioning.
• Backward-compatible by default.
• Breaking change → new version, run both.
• Sunset old version after 6 months.

Observability

• Distributed tracing (Jaeger, Honeycomb).
• Service mesh (Istio) for service-level metric.
• Log aggregation (ELK, Datadog).

VN context

Tiki: 200+ microservices. Shopee VN: similar scale. MoMo: heavy microservice for payment. FPT product groups adopting microservice 2022+ — testing strategy evolving.

Common pitfall

• E2E everything → slow, flaky pipeline.
• No contract test → integration bug surprise.
• Shared DB between service → not really microservice.