Architecture Lab

High-volume payment processing

Design considerations for throughput, correctness and operability in payment-like systems.

High-volume payment-like systems force an explicit conversation about correctness under concurrency.

Useful design pressure points include idempotency keys, bounded concurrency, durable auditability, and failure isolation between validation, processing and reconciliation.

Sanitised system shape

                    API Gateway
                         │
                         ▼
                  Validation Layer
                         │
                         ▼
                    Orchestrator
                    /          \
                   ▼            ▼
               Event Bus      Sync Path
               /      \
              ▼        ▼
          Processor  Processor
              │        │
              └───┬────┘
                  ▼
            Durable Store
                  │
                  ▼
        Reconciliation / Ops

Decision

Prefer asynchronous processing for the hot path while keeping a constrained synchronous path only where business correctness requires immediate confirmation.

Trade-off

Async improves throughput and isolation, but increases operational complexity: ordering, retries, reconciliation and observability become first-class design work.

Failure mode

Duplicate processing, partial writes, poison messages, and silent lag between “accepted” and “settled” if idempotency and reconciliation are weak.

Why I chose this

Throughput problems are usually systems problems. Scaling machines without changing how work moves through the system buys temporary capacity, not durable headroom.

Throughput Idempotency Observability Failure isolation

Related: Engineering Decisions