CURATED CODEX

System Design & Architecture Interview Questions

Q: How do you architect end-to-end API idempotency across distributed microservices and database tiers?

In distributed systems, transient network failures, client timeouts, and automatic retry loops guarantee that mutating HTTP requests (e.g., payments, order submissions) will arrive more than once. End-to-end idempotency guarantees that executing the

Q: How do retries, exponential backoff with jitter, and circuit breakers prevent thundering herd problems during downstream outages?

When a downstream microservice or database degrades, simultaneous retries from thousands of upstream clients create a 'thundering herd' (retry storm), amplifying the load by 10x-100x and preventing the service from recovering. High-scale architecture

Q: How do you design a high-throughput webhook delivery system to reliably process 1,000,000+ webhook deliveries per day?

Processing 1,000,000+ daily webhook deliveries (inbound or outbound) requires an asynchronous, fault-tolerant architecture capable of handling slow receiver endpoints, third-party downtimes, signature verification, and unpredictable traffic bursts wi

Q: How do you debug and isolate whether a downstream dependency is slow vs. down, and how do you protect upstream systems from each failure mode?

In microservice architectures, a downstream dependency that is 'slow' (high p99 latency) is far more dangerous than one that is completely 'down' (immediate TCP connection refused / 500 error). Slow services hold onto upstream worker threads, databas

Curated real-world System Design interview questions spanning rate limiting for AI agents, end-to-end API idempotency, WhatsApp presence engine, Uber surge pricing, multi-region CRDT counters, prompt caching, 100TB sharding, and database decomposition.

Codex Volume19 Questions

Preview Mode Active · 5 Free Sample Questions

Review verified answers and code patterns below. The remaining 14 questions in this codex are unlocked with the Interview Pass or Full Access.

Unlock Codex — ₹299

Q01HardFREE PREVIEW

How do you design a rate limiter that handles both human interactive traffic and AI agent traffic without causing cascading retry storms?

Traditional single-IP fixed-window rate limiters fail for AI agent workloads because autonomous agents generate rapid, multi-step, recursive execution loops that resemble volumetric attacks but are legitimate requests. A production rate limiter must differentiate human interactive users from AI agents, enforce both Request-Per-Minute (RPM) and Token-Per-Minute (TPM) limits, and implement adaptive backpressure to prevent cascading retry storms.

Key architectural pillars: 1. Dual-Dimension Quotas (RPM + TPM): Track both request count and LLM token velocity using Redis Sliding Window Logs or Token Bucket algorithms (CL.THROTTLE). 2. Granular Identity Fingerprinting: Rate limit by composite keys (Org ID + User ID + Agent ID + API Key) rather than raw IP addresses, isolating a rogue agent without impacting human users. 3. Runaway Loop Circuit Breakers: Automatically trip circuit breakers if an agent enters a runaway recursive loop (>50 calls/min on the same parent trace ID). 4. Standard Backpressure Contract: Return RateLimit-Limit, RateLimit-Remaining, RateLimit-Reset, and Retry-After headers with randomized exponential backoff hints for autonomous runtimes.

Verified Sources & Further Reading:

Designing Scalable Rate Limiters for Human & Agent Traffic Rate Limiting Architectures & Distributed Sliding Windows Preventing Agentic Loop Overload & Backpressure Management

Q02MediumFREE PREVIEW

How do you architect end-to-end API idempotency across distributed microservices and database tiers?

Q03MediumFREE PREVIEW

How do retries, exponential backoff with jitter, and circuit breakers prevent thundering herd problems during downstream outages?

Q04HardFREE PREVIEW

How do you design a high-throughput webhook delivery system to reliably process 1,000,000+ webhook deliveries per day?

Q05MediumFREE PREVIEW

How do you debug and isolate whether a downstream dependency is slow vs. down, and how do you protect upstream systems from each failure mode?

Unlock the remaining 14 System Design & Architecture Questions & Verified Answers

You have reached the end of the 5 free preview questions. Get unrestricted lifetime access to all 19 questions, model solutions, architectural blueprints, and complete DSA codex.

Unlock Interview Key — ₹299 Get All-Access Scholar Pass — ₹399

One-time settlement · Lifetime access · Zero recurring bills