Enterprise Integration Patterns
Overview
Enterprise Integration Patterns (EIP), catalogued by Gregor Hohpe and Bobby Woolf, define a vocabulary of 65 patterns for designing robust, asynchronous messaging systems. The patterns describe how to connect applications, transform data in flight, route messages intelligently, and manage messaging infrastructure -- all while keeping systems loosely coupled and independently deployable.
The book organises patterns around the pipes-and-filters architectural style: messages flow through a series of processing steps (filters) connected by channels (pipes).
Pattern Categories
code
┌─────────────────────────────────────────────────────────────────┐ │ Enterprise Integration Patterns │ ├───────────────┬───────────────┬───────────────┬─────────────────┤ │ Messaging │ Message │ Message │ Message │ │ Channels │ Construction │ Routing │ Transformation │ │ │ │ │ │ │ How messages │ How messages │ How messages │ How messages │ │ travel │ are built │ are directed │ are reshaped │ ├───────────────┴───────────────┴───────────────┴─────────────────┤ │ Messaging Endpoints │ System Management │ │ │ │ │ How applications connect │ How to monitor, test, and │ │ to the messaging system │ control the messaging system │ └───────────────────────────────┴─────────────────────────────────┘
Pipes-and-Filters Architecture
code
┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐
│ Source │───>│ Filter A │───>│ Filter B │───>│ Sink │
│ (Producer)│ │(Transform│ │ (Route) │ │(Consumer)│
└──────────┘ └──────────┘ └──────────┘ └──────────┘
pipe pipe pipe
(channel) (channel) (channel)
Each filter performs a single processing step (validate, enrich, transform, route). Each pipe is a message channel connecting one filter to the next. This architecture provides:
- •Composability -- combine simple filters into complex workflows.
- •Replaceability -- swap a filter without touching others.
- •Scalability -- scale individual filters independently.
- •Testability -- test each filter in isolation.
When to Use Messaging vs. Direct Calls
| Concern | Direct / Synchronous Calls | Messaging / Asynchronous |
|---|---|---|
| Coupling | Caller must know callee's address and API | Sender only knows the channel |
| Availability | Callee must be running | Callee can be offline; messages queue |
| Latency | Immediate response required | Eventual response acceptable |
| Throughput | Limited by slowest participant | Buffer with queues; scale consumers |
| Error handling | Caller handles errors in real time | Dead-letter channels, retries, compensation |
| Complexity | Simple for 1:1 interactions | Worth it when >2 participants or reliability matters |
Rule of thumb: Start synchronous. Move to messaging when you need temporal decoupling, load levelling, reliable delivery, or fan-out to multiple consumers.
Pattern Selection Guide
| Problem | Pattern Category | Key Patterns |
|---|---|---|
| How do I send a message from A to B? | Messaging Channels | Point-to-Point Channel, Channel Adapter |
| How do I notify many subscribers? | Messaging Channels | Publish-Subscribe Channel |
| What goes inside a message? | Message Construction | Command Message, Event Message, Document Message |
| How do I correlate request and reply? | Message Construction | Correlation Identifier, Return Address |
| How do I route to the right consumer? | Message Routing | Content-Based Router, Recipient List |
| How do I split and reassemble? | Message Routing | Splitter, Aggregator |
| How do I orchestrate a multi-step flow? | Message Routing | Process Manager, Routing Slip |
| How do I reshape data between systems? | Message Transformation | Content Enricher, Normalizer, Canonical Data Model |
| How do I reduce message size? | Message Transformation | Claim Check, Content Filter |
| How do I consume messages reliably? | Messaging Endpoints | Competing Consumers, Idempotent Receiver |
| How do I monitor and debug? | System Management | Wire Tap, Message Store, Control Bus |
Reference Implementations
| Technology | Language / Platform | Strengths |
|---|---|---|
| Apache Camel | Java / JVM | Broadest connector ecosystem; DSL routes map 1:1 to EIP |
| Spring Integration | Java / Spring | Deep Spring ecosystem integration; annotation-driven |
| MassTransit | C# / .NET | First-class saga support; RabbitMQ & Azure Service Bus transports |
| NServiceBus | C# / .NET | Commercial-grade; strong tooling and monitoring |
| MediatR | C# / .NET | In-process mediator; great for CQRS without infrastructure |
| Azure Service Bus | Cloud (Azure) | Managed broker; topics, subscriptions, sessions |
| RabbitMQ | Any (AMQP) | Lightweight broker; exchanges map to routing patterns |
| Apache Kafka | Any | Log-based streaming; high throughput; replay capability |
Best Practices
- •Learn the pattern language before choosing a framework -- the vocabulary transcends any single implementation.
- •Prefer idempotent message handlers; at-least-once delivery is the norm.
- •Design messages as immutable, self-describing contracts with schema versioning.
- •Keep channels focused on a single data type or purpose (Datatype Channel).
- •Use Dead Letter Channels for every queue -- never silently drop messages.
- •Instrument messaging with Wire Taps and Message Stores from day one; debugging async flows without observability is painful.
- •Start with the simplest topology (point-to-point) and evolve toward pub-sub or content-based routing only when the need is proven.