System design interview question
Design a Distributed Multi-Tenant Message Queue and Event Stream Processing System (like Kafka/Pulsar at Scale)
Design a globally distributed, multi-tenant message queue system that supports both durability-critical transactional workloads and high-throughput event streaming. The system must handle 500K messages/second across 10,000+ customer tenants with strict isolation guarantees, support variable retention policies (1 hour to 30 days), enable topic partitioning with consumer group semantics, and maintain ordering guarantees within partitions while allowing horizontal scaling across geographically distributed data centers. Include mechanisms for handling backpressure, rebalancing consumers gracefully, and preventing data loss during node failures or network partitions.
Practice this question step by step with an AI interviewer that reviews your requirements, your architecture and your trade-offs, and scores the design.