You are designing a new telemetry ingestion platform for connected devices that send real-time events from many regions. The system needs to accept continuous event streams, stay available during regional failures, and support both operational monitoring and downstream analytics. You want a design that can handle high write volume, replay data when needed, and keep bad records from contaminating trusted datasets.
How would you design a highly available, global ingestion pipeline for real-time telemetry data from 500,000 active vehicles?