Time-series database
Your task
Design a time-series database for infrastructure metrics ingesting 4 million data points/sec from 500k hosts, each emitting ~200 series. Queries are mostly recent-window dashboards (last 1–6 hours) with occasional 90-day rollups. Storage must stay affordable while keeping recent data fast. How do you model series, lay out storage, and handle retention and high-cardinality tags?
It covers
- Downsampling
- Retention
- Write amplification