Edge telemetry aggregation

Design the telemetry pipeline for a global edge/CDN network: 300 points-of-presence, each handling 200k req/s, that need per-request metrics (status, latency, cache hit/miss, origin, customer) centralized for billing and analytics. Sending one event per request to a central system is 60M events/s, far too expensive. You still need accurate per-customer billing counters and the ability to investigate latency by customer/region/status. Design edge-side aggregation, what you keep raw vs aggregated, and how you bound central cardinality and cost.