Implement Distributed Time-Series Database for IoT Sensor Analytics

Design a horizontally scalable time-series database architecture for handling 500,000 IoT sensor readings per second across multiple geographic regions. Create a solution that supports automatic data sharding, handles late-arriving sensor data, implements adaptive compression strategies, and provides real-time aggregation capabilities. Include considerations for data retention policies, fault tolerance, and query performance optimization across distributed nodes.