การทดสอบซากาแบบควบคุมจากศูนย์กลาง

An orchestration saga coordinates multiple services through a central orchestrator. A bug in the orchestrator can leave the system in an inconsistent state across services.

Testing sagas is harder than testing a single function because you must verify both the happy path and every compensation path.

• Did each step execute in order?
• Did failures trigger the right compensations?
• Is the final state consistent?

Apply the classic testing pyramid:

• Unit tests — the orchestrator's decision logic in isolation.
• Integration tests — the orchestrator talking to real or stubbed services.
• End-to-end tests — the full saga across all services.

Most tests should be unit tests because they are fast and deterministic.

The orchestrator is often a state machine. You can unit test its transitions by feeding events and asserting the next command.

In unit tests, replace real service calls with mocks or stubs. This lets you control exactly what each step returns.

For example, force the payment service to return a failure so you can assert the orchestrator issues a compensation command.

The riskiest part of any saga is compensation. For each forward step, write a test that:

• Executes steps up to a chosen point.
• Forces a failure.
• Asserts that every prior step was compensated in reverse order.

Integration tests run the orchestrator against real dependencies in disposable containers (databases, message brokers).

Tools like Testcontainers spin up a real broker (e.g. Kafka or RabbitMQ) so message flow is exercised exactly as in production.

To test resilience, inject failures: make a downstream service return errors, time out, or crash mid-saga.

• HTTP 500 from a step
• Timeout (no response)
• Duplicate event delivery

Each scenario should leave the system consistent.

Orchestrators often set a timeout for each step. If a step does not respond, the saga should trigger compensation.

The most important assertion is that the system ends in a valid state. After a failed saga, no money should be charged and no stock reserved.

Query each service and verify the invariants hold across them.

Because messages can be delivered more than once, replay the same event twice in your tests and assert the saga state is unchanged the second time. This verifies idempotency at the orchestrator level.

Add assertions on emitted logs, metrics, and trace spans. A well-tested saga should produce a clear audit trail so that, when a real failure happens in production, you can reconstruct exactly what occurred.

Which test type is the best place to verify the orchestrator's transition logic quickly and deterministically?

You learned how to test orchestrated sagas:

• Use the testing pyramid: many unit tests, fewer integration and E2E tests.
• Mock services to force happy and failure paths.
• Always test compensation order and final state consistency.
• Replay events to confirm idempotency, and inject timeouts and failures.

Thorough saga testing is what keeps distributed transactions trustworthy.