
Monitoring, Metrics and Degradation Tiers

Live streaming at world-record scale with Ashutosh Agrawal (ex-Jio / Disney+ Hotstar)
Topic 5 of 11
Published Feb 12, 2025 · Clip 27:30–32:58 (5:28)
The guest frames metrics as leading indicators (e.g., buffer time per minute, play failure rate) that get priority processing within 30-60 seconds to alert before users complain, versus trailing data used for post-mortem analysis on dashboards. Client-side metric upload frequency is an engineering trade-off - every-second uploads for 30M concurrent users don't scale - so intervals, sampling, and degradation frameworks change at runtime as load grows. He describes tiering of services (P0 through P2) so capacity planning reserves resources for critical services first, leaving P2 services to operate in residual capacity.