Application and enterprise integration
Integration makes independent systems cooperate without turning them into one fragile machine.
Four patterns
| Pattern | Use it when | AWS starting points |
|---|---|---|
| Request/response | caller needs an immediate answer | API Gateway, ALB, AppSync |
| Queue | work can wait and needs buffering | SQS, Amazon MQ |
| Publish/subscribe | many consumers need the same event | SNS, EventBridge |
| Stream | ordered continuous records must be processed | Kinesis, MSK |
Step Functions coordinates multi-step workflows. EventBridge routes events by content. SQS absorbs bursts and isolates failures. SNS fans messages out. Amazon MQ supports familiar broker protocols. MSK provides managed Apache Kafka.
Integration rules
- Prefer contracts and business events over shared databases.
- Make consumers idempotent because delivery can repeat.
- Set timeouts, retries with backoff and jitter, and dead-letter handling.
- Add correlation IDs, schema versions, ownership, and observability.
- Do not retry permanent validation errors forever.
- Protect synchronous chains from cascading failure.
SAP, Salesforce, and Workday
Use AppFlow for supported managed data transfers, including Salesforce and SAP OData. Use APIs and events for operational integration. Use DMS for supported database migration/change data capture, Transfer Family or DataSync for managed file patterns, and an integration platform when enterprise transformation and governance needs justify it.
Workday integration commonly uses Workday-supported APIs, reports, or files through a controlled integration layer. Confirm the tenant's supported interfaces; do not invent a native connector that does not exist.
Architect decision
Separate systems of record from copies. Define which system owns customer, product, employee, price, and order facts. Every replicated field needs an owner, freshness target, failure path, and reconciliation process.
Official sources: Amazon AppFlow, Salesforce connector, and SAP OData connector.