Enterprise Integration Architecture
Integration styles (file, shared DB, RPC/API, messaging, events, CDC), ESB vs iPaaS vs event mesh, canonical models and integration governance in large enterprises.
Integration styles (from Enterprise Integration Patterns, Hohpe and Woolf)
| Style | Pros | Cons | Modern use |
|---|---|---|---|
| File transfer | Simple, batch-friendly | Latency, error handling | Partner/bank file exchanges (SFTP) |
| Shared database | Easy, immediate | Tight coupling, no ownership | Avoid between applications |
| Remote procedure / API | Synchronous answers | Temporal coupling | APIs via gateway |
| Messaging | Decoupled, reliable | Eventual consistency | Commands/work queues |
| Events + streaming | Many consumers, replay | Governance of schemas | Domain events, CDC |
Platform options
| Platform | Strength | Watch out |
|---|---|---|
| ESB (e.g. legacy IBM IIB/ACE, Oracle SB) | Central mediation, protocols | Bottleneck team, logic accumulation |
| iPaaS (MuleSoft, Boomi, Azure Integration Services, Workato) | Connectors, low-code, SaaS integration | Cost, vendor lock-in |
| API management + event streaming (Kafka) | Decentralized, scalable | Requires platform engineering |
| Event mesh (Solace, cloud event buses) | Cross-environment event routing | Another layer to govern |
flowchart LR SAP[(SAP ERP)] -- CDC/connector --> K[[Event streaming platform]] CRM[(Salesforce)] -- iPaaS connector --> K K --> MS1[Order services] K --> LAKE[(Lakehouse)] MS1 -- APIs --> APIM[API management] APIM --> PARTNERS[Partners & channels] BANK[Bank SFTP files] --> FT[Managed file transfer] --> K
Canonical data models
A canonical model standardizes message formats across systems. Use them within a domain or for stable core entities; enterprise-wide canonical models tend to become huge and slow to change. Prefer bounded-context-specific contracts with translation at boundaries.
Governance
Integration catalog (APIs, events, files) with owners; schema registries; SLAs per interface; monitoring of flows end to end; security (mTLS, OAuth, encryption) for all interfaces.
Interview questions and answers
How would you integrate SAP with 30 cloud-native services?
Expose SAP capabilities through system APIs (via SAP Integration Suite or an API layer) for synchronous needs, publish SAP business events via CDC or SAP event mesh to the streaming platform for asynchronous consumers, and keep translation in an anti-corruption layer so SAP structures do not leak into every service.
Is the ESB dead?
The centralized ESB as an all-purpose hub with business logic has largely fallen out of favour; its useful functions (mediation, connectivity, transformation) live on in API gateways, iPaaS and streaming platforms with decentralized ownership.