Design Yelp (Nearby Places)

A proximity service answers "what's near me?": restaurants within 2 km, the top-rated coffee shops nearby, gas stations along the way. Yelp, Google Maps and food delivery apps all depend on one. Unlike ride-hailing, businesses rarely move, so this is a read-heavy system where a precomputed index shines.

Requirements#

Functional

• Given a location and radius, return businesses nearby, ranked by distance, rating or relevance.
• Show business details, photos and reviews.
• Owners can add or update a business; changes can take a few minutes to appear.

Non-functional

• Search results in under 200 ms.
• 100 million daily users, with search traffic far above business updates.
• Highly available; slightly stale results are acceptable.

Estimates#

The spatial index fits in memory on one machine, which simplifies the design a lot.

Choosing a spatial index#

Geohash is a good default to explain. A 6-character geohash is a cell about 1.2 km by 0.6 km.

Searching#

1. Pick a geohash precision based on the radius (bigger radius, shorter prefix).
2. Compute the user's cell and its 8 neighbours. Searching only the user's own cell would miss a business just across the boundary.
3. Fetch business IDs in those cells, compute the real distance, and drop those outside the radius.
4. Rank by a blend of distance, rating and popularity, and return a page of results.

If too few results come back, widen the search by dropping one character of precision.

Architecture#

• Search service: stateless; reads the geo index and business cache.
• Geo index: a table like (geohash, business_id) with an index on geohash, or an in-memory quadtree rebuilt periodically. Replicated widely since it's read-only for searches.
• Business service: handles owner edits, writes to the primary database, and triggers an index update.

Scaling#

• The geo index is small, so replicate rather than shard it: every search node can hold a full copy.
• Cache popular areas: downtown searches repeat constantly; cache results per geohash and filter set for a minute.
• Business details and reviews are much bigger than the index and can be sharded by business ID.

Ranking and filters#

Filters like "open now", price and cuisine are applied after the geo lookup on a few hundred candidates, which is cheap. Personalised ranking can run as a second pass over the filtered list.