Python System Design Interview Questions

Define CPU-bound as compute-heavy (use processes/C extensions) and I/O-bound as waiting on network/disk (use threads or asyncio). Design the hot path to avoid blocking the event loop and isolate heavy compute.

Use asyncio for many concurrent I/O tasks, threads for simple I/O concurrency or library compatibility, and multiprocessing for parallel CPU work. It’s common to mix: async API + background process pool.

Keep handlers non-blocking, cap concurrency with semaphores/queues, use connection pooling/timeouts, and run multiple workers per host. Add health probes, graceful shutdown, and protect the loop with backpressure.

Wrap awaits with asyncio.wait_for (timeouts), catch CancelledError to clean up, and use Tenacity (async-aware) for bounded retries with jitter. Make operations idempotent before retrying.

Enforce per-key limits with token/leaky buckets (often Redis + Lua). For idempotency, store key → result/lock and return the original result on retries to prevent duplicates.

FastAPI: lightweight, async-first, great for microservices. Django: batteries-included (ORM, admin, auth) for product backends. Many teams pair Django for core app and FastAPI for high-QPS services.

Use REST/JSON for public/web clients and caching; use gRPC/Protobuf for low-latency, strongly typed service-to-service calls and streaming. Some stacks expose REST externally and gRPC internally.

Place gateway/ingress (Envoy/Kong/API Gateway) in front to handle auth, quotas, and rate limits, then enforce per-tenant limits and idempotency in the Python service for defense in depth.