Integration Patterns API Rules

x-aid: integration-patterns:rules
x-name: Integration Patterns Usage Rules
description: >-
  Operational rules and guardrails for applying Enterprise Integration
  Patterns (EIP) to real systems. These rules cover when to use messaging
  versus request/response, channel and endpoint design, routing and
  transformation, error handling, and the operational concerns of system
  management patterns in modern microservice and event-driven environments.
x-modified: '2026-04-28'
rules:
  - id: name-the-pattern
    category: Vocabulary
    statement: >-
      Use canonical EIP names (e.g., Content-Based Router, Message Filter,
      Aggregator, Wire Tap) in design documents and code reviews so that
      teams share a precise, technology-independent vocabulary.
  - id: messaging-vs-rpc
    category: Style
    statement: >-
      Reach for asynchronous messaging when components must be loosely
      coupled, survive partial failure, or scale independently; reach for
      synchronous request/response when a caller genuinely needs an
      immediate answer.
  - id: channel-per-purpose
    category: Messaging Channels
    statement: >-
      Define one channel per message type or business intent rather than a
      single firehose channel; this makes Datatype Channel and
      Publish-Subscribe semantics explicit.
  - id: durable-channels
    category: Messaging Channels
    statement: >-
      Use Guaranteed Delivery (durable, persistent) channels for any
      message whose loss would damage the business; reserve volatile
      channels for telemetry-class data.
  - id: command-vs-event
    category: Message Construction
    statement: >-
      Distinguish Command Messages (do this) from Event Messages
      (this happened) in both naming and routing; never overload a single
      message type to mean both.
  - id: correlation-ids
    category: Message Construction
    statement: >-
      Stamp every message with a correlation ID so that Request-Reply,
      Aggregator, and Saga-style flows can be traced end to end.
  - id: idempotent-consumers
    category: Endpoints
    statement: >-
      Make every message endpoint idempotent so that duplicate delivery
      (which is normal in at-least-once messaging) does not corrupt state.
  - id: poison-message-handling
    category: Reliability
    statement: >-
      Route un-processable messages to a Dead Letter Channel after a
      bounded number of redelivery attempts rather than letting them block
      a queue or loop forever.
  - id: content-based-routing
    category: Routing
    statement: >-
      Use a Content-Based Router or Message Filter when only a subset of
      consumers should see a message; do not push routing logic into every
      consumer.
  - id: transformer-isolation
    category: Transformation
    statement: >-
      Implement Message Translator and Content Enricher as dedicated
      components so that schema differences between systems are visible
      and testable, not buried inside business logic.
  - id: orchestration-vs-choreography
    category: Process
    statement: >-
      Use Process Manager / orchestration when a single business process
      needs central control, audit, and timeouts; use choreography
      (event-driven, no central coordinator) when independence and
      autonomy of services matter more than central oversight.
  - id: wire-tap-for-observability
    category: System Management
    statement: >-
      Use Wire Tap and Message History patterns to add observability to
      live channels without altering production message flow.
  - id: control-bus
    category: System Management
    statement: >-
      Manage configuration, feature flags, and operational commands for
      messaging components through a Control Bus, not by ad hoc redeploys
      or direct database writes.
  - id: schema-evolution
    category: Governance
    statement: >-
      Version message schemas explicitly and make changes additive when
      possible; coordinate breaking changes via Message Translator and a
      planned consumer migration rather than in-place schema mutation.
  - id: pattern-not-product
    category: Strategy
    statement: >-
      Choose a pattern first and a product (broker, iPaaS, agent
      framework) second; do not let a vendor's primitives dictate which
      EIP patterns are available to your architecture.