Skip to content

Saga: Orchestration vs Choreography - Professional

A Saga is a durable state machine whose correctness depends on business invariants, not global isolation.

Real systems

  • Temporal persists workflow histories and deterministically replays workflow code.
  • AWS Step Functions stores orchestration state and retry/catch policy.
  • Camunda models BPMN compensation and incidents.
  • Kafka-based choreography relies on log retention, keys, and consumer idempotency rather than a central state machine.

At scale, history growth, retry storms, hot correlation keys, and version skew dominate. Dashboard state age, transition latency, compensation success, history size, and replay failures.

Design and operations checklist

  • State business invariants and compensability per step.
  • Persist transitions and messages atomically.
  • Version workflows for executions spanning deployments.
  • Provide search, pause, retry, and repair controls.
forward action: local commit
compensation: explicit semantic repair

Test yourself

  1. How do you change workflow code while histories are active?
  2. Which invariant cannot be restored by compensation?
  3. What evidence favors choreography over orchestration?

Further reading

  • Garcia-Molina and Salem, Sagas.
  • Temporal durable execution documentation.
  • Richardson, Microservices Patterns.