Design a pub-sub replay system
Company: Google
Role: Software Engineer
Category: System Design
Difficulty: hard
Interview Round: Technical Screen
Assume you have a distributed publish-subscribe messaging system with topics, publishers that write messages, and subscribers that consume them. Design a **replay feature** that lets a subscriber re-read historical messages after a failure, a deployment, or when a new subscriber joins.
Your design should cover:
- replay from a timestamp, message ID, or offset
- replay for an individual subscriber or a subscriber group
- message ordering guarantees within a partition
- storage and retention of historical messages
- high throughput for normal delivery while also supporting replay traffic
- failure recovery, duplicate handling, and operational trade-offs
Discuss the APIs, data model, storage architecture, indexing strategy, scaling approach, and how replay works end to end.
Quick Answer: This question evaluates a candidate's skills in distributed systems and streaming architectures, including pub-sub replay semantics, message ordering, storage and indexing strategies, scalability, and operational reliability.