Sagas: Choreography and Orchestration

Long transactions without distributed locks

A business process like place order → reserve stock → charge payment → arrange shipping spans several services, each with its own database. Two-phase commit across them is rarely practical. A saga is a sequence of local transactions, each publishing an event or message that triggers the next step; if a step fails, earlier steps are undone by compensating actions (release the stock, refund the payment). Compensations are business operations, not database rollbacks, and must themselves be idempotent. In choreography, services react to each other's events with no central coordinator: simple for short flows, but the overall process is implicit and hard to follow. In orchestration, an orchestrator (a dedicated service or a workflow engine such as Temporal, AWS Step Functions or Azure Durable Functions) tells each participant what to do and tracks state: easier to understand, monitor and change, at the cost of a central component.

An orchestrated order saga

Each step either succeeds or triggers compensations for the steps already done.

def order_saga(order_id):
    done = []
    try:
        inventory.reserve(order_id);     done.append("reserved")
        payments.charge(order_id);       done.append("charged")
        shipping.schedule(order_id);     done.append("scheduled")
        orders.mark_confirmed(order_id)
    except StepFailed as failure:
        if "charged" in done:
            payments.refund(order_id)        # compensation
        if "reserved" in done:
            inventory.release(order_id)      # compensation
        orders.mark_cancelled(order_id, reason=str(failure))

# in production the orchestrator persists its state after every step
# (a workflow engine does this for you) so it can resume after a crash

Planning a group trip

Choreography is a group chat where everyone books their own part after seeing the previous message. Orchestration is one friend with a checklist who books each item and, if the hotel falls through, cancels the train tickets.

Quick check: In a saga, what is a compensating action?

• A database rollback across services
• A business operation that semantically undoes an earlier completed step, such as a refund
• A retry of the same step
• A schema migration