Skip to main content

Orchestration vs Choreography

Coordinating multi-step workflows requires a pattern. Orchestration uses a central controller. Choreography relies on events and independent services.

Simple Explanation​

What it is​

Orchestration is a workflow engine telling each step what to do. Choreography is services reacting to events without a single coordinator.

Why we need it​

Multi-step workflows can fail in the middle. You need a pattern that keeps the system consistent and recoverable.

Benefits​

  • Orchestration is easier to monitor and debug.
  • Choreography is more decoupled and flexible.

Tradeoffs​

  • Orchestration can become a bottleneck if overused.
  • Choreography can be harder to trace end-to-end.

Real-world examples (architecture only)​

  • Order workflow -> Step Functions -> Payments -> Inventory.
  • Event stream -> Services react independently -> Final state emerges.

Orchestration flow Choreography flow

Orchestration Model​

Best for:

  • Complex workflows with clear step order
  • Strong audit requirements
  • Long-running business processes

AWS: Step Functions GCP: Cloud Workflows

Python example (start a workflow execution)

import json
import boto3

sfn = boto3.client("stepfunctions")


def start_workflow(order_id):
	return sfn.start_execution(
		stateMachineArn="arn:aws:states:REGION:ACCOUNT:stateMachine:OrderFlow",
		input=json.dumps({"orderId": order_id}),
	)

GCP Workflows call (Python)

import json
from google.cloud import workflows_v1
from google.cloud.workflows.executions_v1 import ExecutionsClient

workflows_client = workflows_v1.WorkflowsClient()
executions_client = ExecutionsClient()


def start_workflow(project_id, location, workflow_name, order_id):
	workflow = workflows_client.workflow_path(project_id, location, workflow_name)
	execution = executions_client.create_execution(
		parent=workflow,
		execution={"argument": json.dumps({"orderId": order_id})},
	)
	return execution.name

Choreography Model​

Best for:

  • High decoupling between services
  • Independent teams owning steps
  • Event-driven ecosystems

AWS: EventBridge, SNS/SQS GCP: Pub/Sub, Eventarc

Python example (event-driven)

import json
import boto3

eventbridge = boto3.client("events")


def emit_order_created(order_id):
	eventbridge.put_events(
		Entries=[
			{
				"Source": "orders",
				"DetailType": "OrderCreated",
				"Detail": json.dumps({"orderId": order_id}),
				"EventBusName": "default",
			}
		]
	)

Decision Matrix​

RequirementOrchestrationChoreography
Clear step order✅❌
Independent teams❌✅
Easy debugging✅❌
Loose coupling❌✅
Long-running workflows✅❌

Common Pitfalls​

  • Orchestration with too many tiny steps becomes hard to maintain.
  • Choreography without good observability becomes a black box.
  • Mixing both without clear boundaries leads to confusion.

Project​

Design a three-step workflow (payment, inventory, notification) in two ways:

  1. Orchestrated workflow
  2. Event-driven choreography

Deliverables:

  • Architecture diagrams for both approaches
  • Failure scenario and recovery plan
  • Which approach you recommend and why

Email your work to [email protected].

References​