Katkat
Back

Design a Sliding-Hour Top-Restaurants Service

DoorDashSoftware Engineer · Onsite
MediumSoftware EngineerOnsiteSystem Design

Last updated

This problem tests the ability to design a high-throughput, real-time analytics system that maintains a top-k leaderboard over a sliding time window.

Problem Statement

Design a service that maintains a list of the top 10 most-ordered restaurants globally over a trailing 60-minute window.

Functional Requirements

  • Provide a real-time getTopRestaurants() API that returns the top 10 restaurants.
  • The system must process incoming order events and update the leaderboard continuously.
  • Support handling late-arriving events (events where the timestamp is slightly behind the current wall clock).

Non-Functional Requirements

  • Scale to handle 100,000 orders per second globally.
  • The top 10 list must reflect the state of the trailing hour with a latency of no more than 5 seconds.
  • The system should be fault-tolerant.

Your Task

Provide a high-level design that covers:

  1. The data ingestion pipeline.
  2. Storage and aggregation strategy (consider trade-offs between precision and latency).
  3. How you handle time-window sliding (e.g., event-time vs. processing-time).
  4. How to scale this service to support per-city leaderboards.

Hints

Hint 1

Consider using a stream processing framework that supports windowing operations, such as Apache Flink or Spark Streaming.

Hint 2

For massive event throughput, using an approximate counting algorithm (like Count-Min Sketch) can significantly reduce memory overhead at the cost of slight accuracy loss.

Hint 3

To handle per-city leaderboards, think about partitioning the data stream by city ID before applying the windowing logic.

Follow-up questions

  • How would your design change if the top 10 list needed to be 100% accurate without using probabilistic data structures?
  • What would you do if a specific, high-volume restaurant causes a 'hot partition' issue in your distributed system?
  • How would you ensure that the top-ten calculation handles system restarts or process failures without losing data?