Design a Local Delivery Service (DoorDash / Instacart)
DifficultyEasy | HelloInterview: problem breakdown
Problem Statement
Design a local delivery service where customers order from local businesses (restaurants, grocery stores), dashers (delivery drivers) pick up and deliver orders, and customers track their delivery in real-time.
๐Real-world: DoorDash/Instacart blend two patterns you've seen elsewhere โ the geospatial matching of Uber (find nearby available dashers) and the state machine of an order moving through placed โ accepted โ picked-up โ delivered, with live location tracking over WebSocket. The interesting wrinkle versus Uber is the two-sided, multi-leg coordination: the dasher, the merchant (is the food ready?), and the customer all need consistent state, and demand is spiky around meal times. Security/integrity angles worth raising: payment fraud and promo abuse (fake accounts farming first-order discounts is a massive cost center for these apps), GPS spoofing by drivers to fake deliveries or game incentive zones, and PII minimization โ the customer's address and phone are sensitive, so mature designs mask phone numbers (proxy numbers) and limit how long the dasher can see the address. "How do you stop a driver from spoofing a delivery?" is a fun curveball.
Requirements
Functional
- Browse nearby businesses and menus
- Place order โ payment โ assigned to dasher
- Real-time order tracking (customer tracks dasher)
- Dasher app: accept/decline orders, navigate to pickup + delivery
Non-Functional
- 10M orders/day โ 115 orders/s average; 500/s peak
- Dasher location update: every 5s โ 500k active dashers ร 0.2/s = 100k location updates/s
- Order assignment latency: < 30 seconds
- Real-time tracking: < 5s location lag
Core Design
Dasher Matching
Similar to Uber: geohash-based location index in Redis.
On new order:
1. Get pickup location geohash
2. Query Redis for available dashers in geohash cell + neighbors
3. Filter: unoccupied dasher, correct vehicle type, rating >= threshold
4. Rank by: proximity + estimated pickup time
5. Send request to top 3 dashers (first to accept wins)
6. Timeout 15s: if no accept, expand radius, try next batchOrder State Machine
PLACED โ CONFIRMED (business accepts) โ READY_FOR_PICKUP โ
PICKED_UP โ EN_ROUTE โ DELIVEREDState transitions trigger:
- Push notifications to customer
- Business notifications
- Analytics events
Real-Time Tracking
Dasher app โ location update every 5s โ Location Service โ Redis (TTL 30s)
Customer app subscribes to dasher location for their active order:
- WebSocket:
subscribe order:{order_id}:dasher_location - Server polls Redis for dasher location, pushes updates to all subscribed customers
ETA: distance_to_destination / average_speed_for_area_and_time โ or integrate with Google Maps Directions API.
Architecture
Customer App โ Order Service โ Payment Service โ Order DB (Postgres)
โ
Dispatcher โ Redis (dasher locations)
โ
Notify Dasher App (push notification)
โ
Dasher Location Service โ Redis โ WebSocket โ Customer
โ
Business Notification ServiceSecurity Considerations
| Threat | Mitigation |
|---|---|
| Customer address privacy | Share address only with assigned dasher; never broadcast |
| Dasher impersonation | Unique per-delivery QR code for ID verification |
| Payment fraud | 3DS2; ML fraud scoring; velocity limits |
| Location spoofing | GPS validation; speed sanity checks; device attestation |
| Order injection | Only authenticated users can place orders; validate cart before payment |
๐ฐReal incident โ DoorDash (2019). An unauthorised party accessed a DoorDash database through a third-party service provider and took data on about 4.9 million customers, couriers, and merchants โ names, addresses, phone numbers, the last four digits of some cards, and driver's licence numbers for about 100,000 couriers. Delivery platforms hold precise addresses and movement data; vendor access to production data needs the same scrutiny as employee access.
๐ฏOn the job. Check how long courier and customer locations are kept and who can see them. Live courier location should be visible to a customer only while their order is active, and historical location trails should be aggregated or deleted after a set period โ location history identifies homes and routines.
Interview Tips
same geo-index, same matching algorithm. The difference is the multi-party coordination (customer + restaurant + dasher).
is explicit and simple โ draw it; shows structured thinking.
is a product differentiator โ mention the feedback loop (actual delivery time โ improve ETA model).