Orchestration · Software component
Event Broker
Software componentOrchestrationOrchestration & ToolsVariation point (abstract)arc:EventBroker
An asynchronous infrastructure that decouples event publishers from subscribers so agents react to state-change events without direct addressing.
Responsibility. Distributes published events to subscribing agents asynchronously.
Also known as: Event bus, Message broker, Event-driven backbone
Variant of Agent Message Bus abstract
When to choose. Choose when high throughput is critical, agents run asynchronously at variable rates, resilience to temporary failures is essential, state changes must reach many subscribers, and eventual consistency is acceptable.
Variants
| Variant | When to choose |
|---|---|
| Content-Based Event Router | Choose when events in one stream must reach different specialised agents depending on classification metadata in the payload (e.g., financial vs legal documents). |
| Event Stream Log | Choose when events must be replayed to rebuild state or train models, analysed temporally, or retained as a permanent audit record. |
| Message Queue | Choose for FIFO buffering and temporal decoupling where queues must absorb traffic spikes and ordering matters. |
| Publish-Subscribe Bus | Choose when a single event must reach multiple subscribers simultaneously by topic for parallel downstream processing. |
Relationships
deployed on structural
writes dependency
receives data from dynamic
sends data to dynamic
- Worker Agent abstract Ch4.3
triggers dynamic
is monitored by assurance
alternative to variability
Design guidance
- MUST pair at-least-once delivery with idempotent subscribers.
- SHOULD use exactly-once semantics where duplicate processing creates monetary errors.
- MUST isolate poison messages to a dead letter queue after retry limits.
- SHOULD enforce retry limits, exponential backoff and jitter to avoid thundering herds.
- SHOULD track event ancestry to detect and break event loops.
- SHOULD decouple producing agents from consuming agents with an asynchronous broker so they scale independently and buffer workload spikes.
- SHOULD reserve events for naturally asynchronous or multi-step workflows rather than simple reads.
- SHOULD let producers remain unaware of consumers so new agent capabilities are added only by subscribing.
Quantitative guidance
As stated by the sources; verify before use.
- Suited to thousands of events per second (Ch1.3 selection criteria).
Classification
- Patterns
- Publish-subscribeAt-most-once deliveryAt-least-once deliveryExactly-once delivery
- Technologies
- Apache KafkaRabbitMQAWS SQSNATSRedis StreamsRedis
- Quality attributes
- Performance efficiency (ISO/IEC 25010)Maintainability (ISO/IEC 25010)Reliability (ISO/IEC 25010 | NIST AI RMF: valid and reliable)
- Risks mitigated
- Cascading failures from synchronous coupling
Sources
- Ch1.3: T. Nguyen, "Multi-Agent Systems," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 1.3. ISBN: 9798244538229.
- Ch4.1: T. Nguyen, "Introduction to AI Agent Deployment and Scaling," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 4.1. ISBN: 9798244538229.
- Ch4.2: T. Nguyen, "Deployment and Scaling," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 4.2. ISBN: 9798244538229.
- Ch4.3: T. Nguyen, "Container Orchestration and Edge Deployment," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 4.3. ISBN: 9798244538229.