Orchestration VS Choreography in Serverless

In the world of modern application development, where distributed systems are the norm, how do you keep your applications running smoothly and efficiently? Should you opt for a central coordination approach with orchestration, or let each component handle its own tasks through choreography?

In this talk, we’ll explore these two strategies, diving into their benefits and drawbacks. Using hands-on examples with AWS serverless services, you’ll see how to implement both approaches and gain practical tips to help you decide which fits your needs best. Whether you’re new to these concepts or looking to refine your skills, you’ll leave with clear insights to guide your decision-making.

Transcript

1. Orchestration VS Choreography in Serverless AWS User Group - Torino, 23/11/2023
2. Microservices anyone?
3. of organizations use microservices 85% Sources: CNCF Cloud Native Survey 2021, DORA State of DevOps 2021 reduction in mean time to recovery 60%
4. 󰗞 I am Francesco Lerro Cloud and Serverless computing expert, based in Turin (Italy) Currently riding the Software Architect elevator at I love solving the Rubikʼs cube and eating buffalo mozzarella Linkedin profile
5. Microservices collaboration A central coordinator controls the execution flow and the interactions between the various services ORCHESTRATION CHOREOGRAPHY Decentralised approach, each service may be asynchronously triggered by the operation of other services
6. Orchestration implementation
7. Choreography implementation
8. Letʼs try it live!
9. Orchestration VS Choreography ▪ Business process control ▪ Operation coupling ▪ Service failure and error handling
10. What about serverless?
11. Choreography, everywhere
12. Events, unexpected Sync invocation
13. How to mitigate choreography downsides?
14. Adopt distributed tracing
15. A service map in AWS X-Ray
16. Design for failure On Success On Error SQS Queue DynamoDB For example, leveraging Lambda Destinations
17. Use idempotency FTW! INVOCATION RETRY BEHAVIOUR API GATEWAY Synch None – returns error AMAZON S3 Asynch Until configured timeout, 2 times AMAZON SNS Asynch In case of unavailability DYNAMODB STREAMS Synch (from Poller) Until data expiration (24h) KINESIS Synch (from Poller) Until data expiration (24h-7d) SQS Synch (from Poller) Until retention period exp. or DLQ
18. What about orchestration in serverless?
19. Letʼs use Lambda…
20. Letʼs use Lambda… Please donʼt. Cost, tight coupling, error handling….
21. Letʼs use Lambda… Please donʼt. Try Step Functions, instead!
22. Coordinated transactions via AWS Step Functions Let’s start with the first set of slides Credits: Yan Cui, The SAGA pattern
23. Let’s start with the first set of slides Booking SAGA The happy path
24. Booking SAGA Error handling Let’s start with the first set of slides
25. Orchestration or Choreography?
26. THANKS! Any questions? Francesco Lerro