Google system design interview questions

Which levels get a design round

Design questions candidates report

• Design Google Docs (collaborative editing)

  A LADDER walkthrough of a collaborative document editor: requirements, scale math for millions of concurrent editors, the API and an operation-log data model, then the hard parts: why concurrent edits conflict, operational transformation versus CRDTs, one session server per document, durability with an op log and snapshots, failover, offline edits, live cursors, version history and permissions.

  Full walkthrough: Google Docs

• Design search autocomplete / typeahead

  Design the typeahead box that suggests the top five queries as you type. Requirements and scale (over a hundred thousand requests a second at peak), the suggest API, a trie with top-K cached at every node, an offline pipeline that rebuilds it from query logs, a small real-time layer for trending terms, replication versus sharding, client-side tricks and the trade-offs.

  Full walkthrough: Search autocomplete

• Design Google Maps

  A senior-level design for a maps and navigation service like Google Maps: requirements, tile and probe math, the tile, search and route APIs, pre-built vector tiles on a zoom pyramid served from a CDN, why plain Dijkstra is too slow on a continent and how preprocessing such as contraction hierarchies fixes it, live traffic from phone speed probes and map matching, ETA prediction, rerouting, and the trade-offs.

  Full walkthrough: Google Maps

• Count YouTube video views

  How to design a long-form video platform like YouTube: requirements, upload and egress math, resumable upload sessions, a transcoding pipeline built as a DAG of parallel jobs, adaptive bitrate streaming with HLS or DASH, CDN tiers and origin shielding for the long tail, counting billions of views with dedupe and batching, metadata storage, and the trade-offs.

  Full walkthrough: YouTube

• Design a web crawler

  Design a crawler that fetches a billion pages a month without hammering any site. Requirements and scale, the control interfaces and URL table, the fetch-parse-enqueue loop, then the hard parts: a URL frontier with priority and per-host politeness queues, URL and content deduplication with Bloom filters and fingerprints, spider traps, recrawl scheduling, partitioning by host, and the trade-offs.

  Full walkthrough: Web crawler

• Design a distributed job scheduler

  How to design a shared scheduler that runs millions of one-off and cron jobs on time: a jobs table indexed by next run time, partitioned schedulers that claim due work with leases, a unique run key so each tick fires once, a queue and worker pool with heartbeats, retries, timeouts and fencing tokens, and the midnight thundering herd, priorities and per-team quotas.

  Full walkthrough: Job scheduler

Also reported

• Design a load balancer

How to answer any design question: the LADDER method · How the design round is scored by level

Questions people ask