Stock Trading App (Robinhood)

Brokerage orders and live prices

Design a retail brokerage app: millions of people watch live prices and place market and limit orders, which the broker checks against their buying power and routes to exchanges and market makers, then tracks to the last fill.

Overview

A retail brokerage app lets people watch stock prices and buy or sell shares. The broker is not the exchange: it does not match buyers with sellers. It holds customer accounts, checks that a customer can pay, sends each order to an exchange or a market maker (a firm that trades against incoming orders), and records what comes back. A clearing firm then settles the trades and holds the cash and shares; in the US, most stock trades settle one business day after the trade (T+1).

The design splits into two paths with opposite needs. The price path is huge and forgiving: millions of open screens, each watching a few dozen symbols, where a price that is 200 ms old is fine and a skipped tick is invisible. The order path is small and strict: a few thousand orders a second at the opening bell, where a lost order, a duplicate order or an overspent account is a real loss. Keep them apart, so a flood of quotes can never slow an order.

Questions to ask first

Are we the broker or the exchange? — The broker; orders go out to exchanges and market makers
Which products? — US stocks in a cash account; no options, crypto or margin
Which order types? — Market and limit, day orders, place and cancel
How fresh must prices be? — Under a second on screen; quotes are informational, never used to fill
How many users? — 2 million connected at the open, 5 million orders a day
Trading hours? — The regular session only, 9:30 a.m. to 4:00 p.m. Eastern

What you will be able to do

Fan live prices out to millions of clients with per-symbol subscriptions and conflation.
Make order entry safe to retry with a client order ID, end to end.
Check and hold buying power before an order leaves the broker, and release it correctly.
Drive each order's state from FIX execution reports, including partial fills and cancel races.
Reconcile with the clearing firm and plan for the opening bell.

Key concepts

Quote fan-out
Per-symbol subscriptions
Conflation
Client order ID
Buying-power hold
FIX execution reports
Order state machine
Positions read model
Clearing reconciliation
Opening-bell spike

Functional requirements

See live prices for a watchlist and a stock page
Place a market or limit order to buy or sell
Cancel an open order
See each order's status, fills and average price
See positions, cash, buying power and portfolio value

Non-functional requirements

An order is never lost and never placed twice, even when the client retries
No order can spend more than the account's buying power
Prices reach the screen within about 1 second of the feed (a target we chose)
Order acknowledgment in under 500 ms at p99, not counting the venue (a target we chose)
The order path stays up through the opening-bell spike; prices may degrade first
Every order and execution kept for the required retention period, unchangeable

Interview questions

How do you push live prices to millions of clients?
How do you make sure an order is never placed twice?
How do you stop a customer from spending money they do not have?
What happens when a user cancels an order that is filling?
How does the broker know what actually traded?
How do you handle the opening bell?
What does reconciliation with the clearing firm catch?
How is this different from designing the exchange?