When dealing with distributed Enterprise Resource Planning (ERP) systems, the failure condition isn't just a dropped frame or a 404 page—it's operations halting at a branch location while an operator waits for a spinning loader to resolve.
High-Concurrency Operations
In my experience migrating legacy local SQL-based terminals to cloud-connected architectures, the most critical element has always been offline resilience.
We utilized CQRS (Command Query Responsibility Segregation). When the central database became unreachable, the local node would queue commands onto a hardened local MassTransit/RabbitMQ queue. The operator could continue processing transactions and generating documents. The moment the node re-established a heartbeat with the cloud hub, the queue would replay the exact ledger movements idempotently.
This allowed a 100% uptime SLA for the branch operator, circumventing any centralized database lock or outage.
Inventory Synchronization
Stock counts in a multi-location enterprise suffer from phantom reads. An order is placed online while someone is processing the exact item at a branch. By leveraging a strict event-sourcing layer instead of UPDATE inventory SET count = count - 1, we gained a perfect ledger to manually review discrepancies and ensure no ghost inventory was allocated.