E-Commerce Platform

Catalog, search, cart, checkout and orders

Design an online store with a large catalog, faceted search, a cart, stock held in several warehouses, and a checkout that never sells stock it does not have or charges a card twice.

Overview

An online store lets shoppers find products, put them in a cart and buy them. Behind that are two very different workloads. Browsing (product pages, search, category lists) is almost all reads: with the numbers assumed here, about 625 page views and searches for every order. Buying (checkout, stock, payment) is a small share of traffic, but it moves money and physical goods, so it must be correct.

A strong answer keeps the two apart. Browsing is served from caches, replicas and a search index that may be a second or two behind. Buying goes to a transactional store that decides, one row at a time, whether stock is really there. The interviewer usually probes the catalog model, how search stays fresh, how a price is fixed at checkout, how stock in several warehouses is reserved, and what happens when checkout fails halfway.

This topic is the generic store. The Amazon E-commerce topic goes deeper on an always-writable cart and one very hot item, and the Flash Sale System topic on a sale where a crowd arrives at one instant.

Questions to ask first

How many shoppers and orders? — 50 million daily active shoppers; 2 million orders a day
How big is the catalog? — 100 million SKUs (each size and color of a product is its own SKU)
Where is the stock? — Several warehouses; an order ships from one, or is split
Can we oversell? — No. A confirmed order has stock reserved behind it
When is stock held? — At checkout, for 15 minutes; never while an item only sits in a cart
Payments? — An external payment processor: authorize at checkout, capture at shipment
How spiky? — Sale days reach 10 times the average

What you will be able to do

Model products, variants and attributes so both the product page and filters work.
Keep a search index fresh from the catalog without losing updates.
Fix a cart's price at checkout, in whole cents, and explain what the shopper sees when it changes.
Reserve stock across several warehouses without overselling, and release it when a checkout is abandoned.
Run checkout as a saga with authorize-then-capture payments, and publish order events reliably with an outbox.

Key concepts

Products and SKUs
Search indexing pipeline
Facets
Price snapshot
Available to promise
Stock reservation
Checkout saga
Authorize and capture
Transactional outbox

Functional requirements

Show a product page with its variants, price and availability
Search the catalog by text, with filters (brand, size, price) and counts per filter value
Add, change and remove cart items; show an up-to-date total with promotions and tax
Check out: reserve stock, authorize payment, and return an order number
Show order status from confirmed to shipped and delivered; let the shopper cancel before shipment

Non-functional requirements

Product pages and search answer in under 200 ms at p99 (a target we chose)
Browsing stays up when checkout or payments are degraded
Never confirm an order without stock reserved for it
A retried checkout never creates a second order or a second charge
Catalog changes appear in search within about a minute (a target we chose)

Interview questions

How do you prevent overselling?
Walk me through checkout. What happens if payment fails halfway?
How does search stay in sync with the catalog?
Why store money as integer cents?
Where do you keep the cart, and does it hold stock?
Why authorize at checkout and capture at shipment?
How do other systems learn that an order was placed, without slowing checkout?
How would you scale the transactional store?