Google system design interview questions

Sample questions with short answers

Tap a question to reveal the short answer and key points.

1. 1Design a search autocomplete service like Google's suggestion box.HardGoogle-specificShow short answer

   Short answer

   Serve suggestions from a trie or prefix index precomputed from query logs, storing the top-K completions at each node so lookups are O(prefix). Update rankings offline via a batch pipeline on aggregated logs, and cache hot prefixes at the edge. Discuss a p99 latency budget of tens of milliseconds, personalisation as a re-rank layer, and filtering offensive queries.

   Interviewers listen for

   • trie or prefix index with top-K per node
   • offline aggregation of query logs
   • tight latency budget with caching
2. 2How would you design a global notification system (push, email, SMS)?HardGoogle-specificShow short answer

   Short answer

   Accept events via an API into a durable queue, fan out per channel with dedicated workers, and keep templates and user preferences in a config store. Ensure idempotency with notification IDs, retries with backoff, and per-provider rate limits. Track delivery status for analytics and let users opt out per channel. Google expects you to quantify queue throughput and talk about regional deployment.

   Interviewers listen for

   • durable queue with per-channel workers
   • idempotency and retries
   • user preferences and opt-out
3. 3Design a URL shortener. Walk through the key decisions.MediumAsked everywhereShow short answer

   Short answer

   Clarify scale (e.g. 100M new URLs/month, 10:1 read:write) and features (custom aliases, expiry, analytics). Generate short keys with base-62 encoding of a counter or a hash with collision checks. Store the mapping in a key-value store, put a cache in front for hot links, and make redirects 301/302 depending on whether you need click analytics. Discuss sharding by key and rate-limiting abuse.

   Interviewers listen for

   • clarify scale and read/write ratio
   • base-62 key generation
   • cache for hot redirects
   • sharding the key-value store
4. 4What is the CAP theorem and how does it shape real system choices?MediumAsked everywhereShow short answer

   Short answer

   During a network partition a distributed system must choose between consistency (every read sees the latest write) and availability (every request gets a response). Banking ledgers lean CP; social feeds and shopping carts lean AP with eventual consistency. In interviews, name the partition scenario explicitly and justify which side the product can tolerate.

   Interviewers listen for

   • partition forces consistency vs availability
   • banking leans CP, feeds lean AP
   • justify choice from product tolerance
5. 5How would you design a rate limiter for an API gateway?HardAsked everywhereShow short answer

   Short answer

   Choose an algorithm — token bucket is the usual default (allows bursts, simple to reason about); sliding window log is exact but memory-heavy. Keep counters in Redis keyed by client and window, using atomic INCR/EXPIRE or a Lua script to avoid races. For multiple gateway nodes, centralise counters or accept slight over-admission with local buckets. Return 429 with Retry-After headers.

   Interviewers listen for

   • token bucket or sliding window
   • Redis counters with atomic operations
   • distributed nodes and 429 responses