Section 1: Engineering Challenges in Agent Collaboration 1.1 Communication Complexity: Link fragmentation, resource contention, and scheduling blind spots under concurrent inference. 1.2 Engineering Hazards: Disrupted communication links, resource contention and scheduling failure caused by transient bursts, and the difficulty of fault recovery following the loss of collaborative context.

Section 2: Lite-Topic: Efficient Routing and Scheduling for Massive Sessions 2.1 Dynamic Session Management: Building a TTL-based, second-level dynamic topology to achieve logical isolation of session spaces between agents. 2.2 ReadySet Scheduling Mechanism: Optimizing the delivery efficiency of massive agent events via distributed brokers, addressing “thundering herd” problems and latency jitter under ultra-high concurrency.

Section 3: RocketMQ-A2A Asynchronous Interaction Model: Decoupling and Deterministic Control 3.1 Asynchronous Paradigm Refactoring: Building a full-link asynchronous model for “task orchestration, trigger execution, and result return,” sinking collaboration logic into the RocketMQ foundation. 3.2 Deterministic Communication Orchestration: Defining standardized protocols for inter-agent communication to ensure the deterministic execution of complex business flows.

Section 4: Collaborative Consistency and Observability: Constructing Event-Driven Evidence Chains 4.1 Semantic Enhancement and State Recovery: Leveraging idempotency design and event logs to achieve atomic interruption replay and lossless state migration for agent collaboration flows. 4.2 Full-Link Collaborative Auditing: Creating “collaborative evidence chains” based on structured messages to transform fragmented agent behaviors into traceable and attributable system governance data.

Section 5: Large-Scale Cluster Performance and Practical Closed-Loop 5.1 Stability and fault recovery under high-pressure collaboration. 5.2 How AI gateways leverage the messaging foundation to achieve fault self-healing and collaborative task orchestration for large-scale multi-agent systems.

Speakers:

Zhou Li: aliyun

Messaging Expert at Alibaba Cloud, responsible for Alibaba’s core messaging middleware