Saga Orchestration

Patterns for managing distributed transactions and long-running business processes without two-phase commit.

Inputs and Outputs

What you provide:

• Service boundaries and ownership (which service owns which step)
• Transaction requirements (which steps must be atomic, which can be eventual)
• Failure modes for each step (transient vs. permanent, retry policy)
• SLA requirements per step (informs timeout configuration)
• Existing event/messaging infrastructure (Kafka, RabbitMQ, SQS, etc.)

What this skill produces:

• Saga definition with ordered steps, action commands, and compensation commands
• Orchestrator or choreography implementation for your chosen pattern
• Compensation logic for each participant service (idempotent, always-succeeds)
• Step timeout configuration with per-step deadlines
• Monitoring setup: state machine metrics, stuck saga detection, DLQ recovery

When to Use This Skill

• Coordinating multi-service transactions without distributed locks
• Implementing compensating transactions for partial failures
• Managing long-running business workflows (minutes to hours)
• Handling failures in distributed systems where atomicity is required
• Building order fulfillment, approval, or booking processes
• Replacing fragile two-phase commit with async compensation

Detailed section: Core Concepts

Moved to references/details.md.

Detailed section: Templates

Moved to references/details.md.

Best Practices

Do’s

• Make every step idempotent — Commands may be replayed on broker reconnect
• Design compensations carefully — They are the most critical code path
• Use correlation IDs — The saga_id must flow through every event and log
• Implement per-step timeouts — Never wait indefinitely for a participant reply
• Log state transitions — saga_id, step_name, old_state → new_state on every change
• Test compensation paths explicitly — Inject failures at each step index in integration tests

Don’ts

• Don’t assume instant completion — Sagas are async and may take minutes
• Don’t skip compensation testing — The rollback path is the hardest to get right
• Don’t couple services directly — Use async messaging, never synchronous calls inside a saga step
• Don’t ignore partial failures — A step that partially executed still needs compensation
• Don’t use a global timeout — Each step has different latency characteristics

Troubleshooting

Saga stuck in COMPENSATING state

A saga enters compensation but never reaches FAILED. This means a compensation handler is throwing an unhandled exception and never publishing SagaCompensationCompleted. Add dead-letter queue (DLQ) handling to compensation consumers and ensure every compensation action publishes a result event even when the underlying operation was already rolled back.

async def handle_release_reservation(self, command: Dict):
    try:
        await self.release_reservation(command["original_result"]["reservation_id"])
    except ReservationNotFoundError:
        pass  # Already released — treat as success
    # Always publish completion, regardless of outcome
    await self.event_publisher.publish("SagaCompensationCompleted", {
        "saga_id": command["saga_id"],
        "step_name"

Duplicate saga executions on restart

If your orchestrator service restarts mid-saga, it may replay events and re-execute already-completed steps. Guard every step action with an idempotency key — see Template 3 above.

Choreography saga losing events

In a choreography-based saga, a downstream service may miss an event if it was offline when published. Use a durable message broker (Kafka with replication, RabbitMQ with persistence) and store the current saga state in a dedicated saga_log table so you can replay from the last known good step.

Timeout firing before a slow-but-valid step completes

A step like create_shipment might take up to 15 minutes during peak load but your global timeout is 5 minutes, causing spurious compensation. Make step timeouts configurable per step type — see references/advanced-patterns.md for the TimeoutSagaOrchestrator implementation and the STEP_TIMEOUTS dict pattern.

Compensation order not matching execution order

When two steps both complete before a failure is detected, compensation must run in strict reverse order or you leave data in an inconsistent state. Verify that _compensate() iterates from current_step - 1 down to 0, and add an integration test that deliberately fails at each step index to confirm correct rollback order.

Advanced Patterns

The references/ directory contains production-grade implementations not needed for most sagas:

• references/advanced-patterns.md — Full SagaOrchestrator abstract base class, TimeoutSagaOrchestrator with per-step deadlines, detailed bank transfer compensating transaction chain, Prometheus instrumentation, stuck saga PromQL alerts, and DLQ recovery worker.

Related Skills

• cqrs-implementation — Pair sagas with CQRS for read-model updates after each step completes
• event-store-design — Store saga events in an event store for full audit trail and replay capability
• workflow-orchestration-patterns — Higher-level workflow engines (Temporal, Conductor) that build on saga concepts