AWS Step Functions 101

Ref: https://learn.cantrill.io/courses/1820301/lectures/41301398

AWS Lambda Limitations

Reminder: Lambda is FaaS with 15-mins max execution time… → don't try to put a full app inside a Lambda function!
- Functions can be chained together (simulate stateful apps)… but generally a bad idea!
  - 👎 Scales very poorly!
  - Lambda's RTEs are stateless!
    - Data & environments cleaned after each function execution
❗ These Lambda limitations were designed purposefully!
- 💡 No product is a perfect fits-all, it's important to understand what a product was designed for, and where to use it

🔧 Manage long-running, serverless workflows
- Workflows defined with State Machines
- Addresses some limitations of Lambda (e.g. max execution time)
Example of a good fit: retail order in Amazon.com
- It can take hours or days to process an order, with lots of different things happening in between
- Lambdas can't wait for so long… we require a long-running workflow
Types of workflows:
1. Standard Workflow (Default)
  - Maximum duration = 1 year → long-running workflow
2. Express Workflow
  - Maximum duration = 5 minutes
  - Can rerun failed operations more easily
  - Use case: high volume & highly transactional event-processing workloads (need more processing guarantees)
    - e.g. data processing or ML workflows

🔧 State Machine = base entity of AWS Step Functions
- Serverless workflow: START → [STATES] → END
- States take in data, modify data, and output data
SM performs activity or flow
- Lots of individual components → SM maintains idea of data between states
- Complex workflows involving several AWS services
- IAM roles used for permissions → SM assumes role while running
SM can be started via APIGW, IoT rules, EventBridge, Lambda… even manually
- Generally used for Backend processing
Amazon States Language (ASL)
- Language used to create JSON templates that define SM state machines
- Create/export SMs