E-commerce Checkout

Designing a resilient checkout pipeline: cart, pricing, inventory, payments, and orders.

Requirements

• Cart and promo codes; taxes and shipping
• Atomic order placement with payment authorization/capture
• Inventory reservation and decrement

Non-functional Requirements

• High availability: 99.9%+ during peak sales
• Idempotency for checkout retries
• Observability for payment/inventory failures

High-Level Design

• Checkout orchestrator drives the saga across services
• Payment authorization before order confirmation; capture on shipment
• Inventory reservation with timeouts; eventual consistency to stock

Capacity & Sizing

• Peak TPS during sales events; orchestrator horizontal scale
• Reservation TTL sizing vs available inventory
• Ledger/event storage growth from orders and payments

Key Components

• Checkout Orchestrator (saga), Payment adapter(s)
• Inventory reservation and stock service
• Order service and events outbox

Data Model

• orders (order_id PK, user_id, status, total, currency, created_at)
• order_items (order_id, sku, qty, price)
• payments (payment_id PK, order_id, provider, status, auth_id, amount)
• inventory_reservations (reservation_id PK, sku, qty, expires_at, order_id)

APIs

• Add to cart: POST /api/cart/items with body { sku, qty }
• Checkout: POST /api/checkout with body { cartId, paymentMethod, address }
• Order by id: GET /api/orders/:id

Hot Path

1. Validate cart → reserve inventory → authorize payment
2. Create order → emit events → finalize reservations

Checkout Flow

1. Validate cart, promotions, taxes, and shipping
2. Reserve inventory per SKU with expiry
3. Authorize payment with idempotency key
4. Create order; emit events; release or adjust reservations on failure

Scaling

• Shard orders by order_id prefix; scale orchestrators horizontally
• Use queues for retries and outbox pattern for event delivery
• Rate limit external providers; circuit breakers and fallbacks

Caching & TTL

• Cache product/pricing lookups; invalidate on catalog changes
• Short TTL caches for tax/shipping quotes

Failure Modes & Mitigations

• Payment timeout → retry with idempotency; mark pending
• Inventory race → conditional updates and reservations with TTL
• Partial failures → saga compensation (refund/cancel)

Observability

• Track auth/capture success rates, fraud flags, and stockouts
• Trace across orchestrator, payment, and inventory services