Microsoft System Design: Distributed Key-Value Store & Cache

You’re asked to design a distributed key-value system that acts both as a low-latency cache (sessions, profiles, product details) and as persistent storage for critical configuration data. The system will serve millions of global users across multiple data centers and must meet strict NFRs: high availability (99.9% uptime), horizontal scalability, low latency (<10ms typical), configurable consistency, durable persistence, and robust failure handling.

Core content

You should describe an architecture that includes: consistent hashing for automatic data partitioning and shard rebalancing; replication across nodes and across data centers for redundancy; a pluggable consistency model (strong vs eventual) using quorum reads/writes or consensus (Raft/Paxos) for strongly consistent items; cache layer features (TTL, LRU eviction, high cache-hit strategies); and persistence options (write-through, write-back, or async replication). Explain how you’ll handle keys up to 256 bytes and values up to 1MB, and how to avoid hot keys and size-related performance problems.

Flow / interview stages

Start with requirements and trade-offs, propose a high-level architecture, then dive into data partitioning, replication, consistency choices, failure detection and recovery, and operational concerns (monitoring, metrics, logging, and capacity planning). Be ready to sketch API semantics for get/put/delete/update and TTL behavior.

Skill signals

Demonstrate knowledge of distributed systems (consistent hashing, quorum, consensus), caching strategies (eviction policies, cache warming), replication and cross-DC design, failure modes and recovery, and performance optimizations (load balancing, batching, compression). Show you can reason about CAP trade-offs and give concrete mitigation strategies.