Stripe System Design Interview Questions

In double-entry accounting, every transaction has two entries: a debit and a credit, each tied to an account. Store these in an immutable ledger table with transaction IDs, timestamps, and sequence numbers. The system should guarantee atomic writes of both entries to maintain balance integrity. Use append-only storage to preserve history, and provide reconciliation jobs to verify that debits and credits always net to zero.

A webhook service should be idempotent, retryable, and observable. Store pending deliveries in a durable queue, send HTTP POST requests to subscribed endpoints, and retry with exponential backoff for failures. Include delivery signatures (HMAC) for authentication. Log every attempt and allow users to view replay history. To prevent slow endpoints from blocking others, isolate retries in separate worker pools.

An international payout flow needs to support multi-currency settlements and comply with country-specific regulations. Maintain a currency ledger for each merchant, aggregate balances, and trigger payouts according to merchant preferences. Integrate with local payment rails (ACH, SEPA, SWIFT) through banking APIs. Handle FX conversions using either real-time or fixed-rate quotes. Track payout statuses and support recall flows in case of errors.

Ingest payment and refund events via an event bus (Kafka) into a stream processing layer that aggregates revenue, fees, and balances. Write results to a low-latency analytics store (like Druid or Pinot) for fast dashboard queries. Keep raw transactions in a data lake for historical reporting. Ensure per-merchant data isolation and permission enforcement before rendering data.

Integrations occur via secure, versioned APIs. OAuth2 is used for authorization, with scopes restricting data access. API requests are signed and rate-limited. Webhooks keep third-party apps updated with real-time changes. Sandbox environments allow safe testing before production rollout. Integrations are monitored for latency, error rates, and suspicious activity.

Kindle payments involve digital goods with regional restrictions. The payment flow should integrate with Stripe’s API for authorization and settlement, while enforcing region-based tax rules (VAT, GST). Support recurring payments for subscriptions and one-click purchases for eBooks. Track entitlements in a separate service so refunds or disputes automatically revoke access to purchased books.

Use a consistent hashing scheme to distribute keys across nodes. Each node maintains an LRU eviction policy locally, removing least-recently-used keys when full. Replicate keys to a small number of neighbors for fault tolerance. Use a gossip protocol to detect failures and rebalance partitions. Monitor hit/miss ratios to optimize cache sizing.

An APM collects metrics, traces, and logs from distributed services. Agents in each service send telemetry to a collector, which aggregates and stores it in a time-series database. Provide dashboards to visualize latency, throughput, and error rates. Sampling and compression reduce overhead. Anomaly detection can trigger alerts when performance deviates from baselines.

Use a token bucket or sliding window algorithm to enforce limits. Store counters in a distributed in-memory store (e.g., Redis) with atomic increment and expiry. Apply limits per API key, IP, or account. For multi-region deployments, keep local limits and sync periodically for global enforcement. Expose rate-limit headers so clients can self-throttle.