Idempotency

Run a Function Only Once

Wrapping a function so it runs at most once is the right tool for one-shot initialisers, lazy connection setup, and event handlers that must not double-fire. This snippet covers the canonical `once`, an async-aware version that caches the in-flight Promise, and a `resetOnce` variant for tests and feature flags. Drop it next to your event listeners and stop guarding with ad-hoc booleans.

Design a Notification Service

Design a multi-channel notification service that delivers 10B push, email, and SMS notifications per day across three independent provider networks (APNs, FCM, SendGrid, Twilio) with priority queues, per-user rate limits, and idempotent retries. The interview centerpiece is the fan-out from a single application event to multiple channels and providers, each with its own rate limits, failure modes, and delivery semantics. We cover priority queues for transactional vs marketing traffic, retry policies with exponential backoff, deduplication of duplicate triggers, user preference enforcement, and the device token lifecycle that quietly invalidates tens of millions of tokens per day.

Design an E-Commerce Platform (Amazon)

Design an Amazon-scale e-commerce platform that lets 200M monthly users browse 100M SKUs, add items to a cart, check out, and have orders fulfilled from regional warehouses. The interview centerpiece is the order lifecycle: how to reserve inventory atomically while a customer is on the checkout page, how to chain cart-to-payment-to-fulfillment as a saga with compensating actions, and how to make checkout idempotent so a flaky network never charges a customer twice. We also cover catalog browse at scale, multi-warehouse fulfillment routing, and the asymmetric read/write workload that makes aggressive catalog caching the right call.