<![CDATA[Live commerce fulfilment is characterized by simultaneous surges in orders, rapid conversion from viewer engagement to purchase, and transactions involving multiple sellers. Conventional order-centric fulfilment architectures may produce partially completed orders when inventory, payment, carrier, or notification processes fail independently across sellers and supporting services. The study proposes a package-centric fulfilment architecture in which the package aggregate represents a single seller-to-buyer shipment unit and functions as the consistency boundary and failure-isolation scope. The architecture is implemented as six bounded microservices that communicate through domain events using Apache Kafka, with topic partitioning by package identifier to preserve per-package event ordering without synchronous inter-service coupling. The outbox pattern maintains consistency between local service state and published events, while idempotent consumer registries prevent duplicate state changes under at-least-once event delivery. A choreography-based saga defines bounded compensating paths for four representative failure scenarios: inventory exhaustion, carrier rejection, payment failure, and notification delivery loss. Scenario-based evaluation assesses each scenario for compensation scope, state-preservation guarantee, and idempotency safety under concurrent and asynchronous operating conditions. The evaluation shows that all four failure scenarios have explicit compensating paths and that compensation remains confined to the affected package rather than propagating across unrelated packages or sellers. Local state transitions remain protected against duplicate event processing, while package-level consistency is preserved during compensation. These results establish a structural mechanism for isolating partial-fulfilment failures while maintaining a coherent package state across distributed services. The proposed architecture provides a foundation for subsequent empirical assessment of recovery performance, event-processing latency, throughput, and operational reliability in live commerce fulfilment environments]]>
Copyrights © 2026
