Design a Metrics Storage System

Requirements

• •Ingest high-frequency metric data points (metric name, value, timestamp, tags)
• •Store raw data with configurable retention (e.g., 7 days raw)
• •Automatically downsample data into rollup aggregates (1-min, 1-hour, 1-day)
• •Serve time-range queries with aggregation (avg, max, p99) efficiently
• •Support alerting hooks when a metric exceeds a threshold

Non-Functional Requirements

• •Ingest 1 million data points/second
• •Query latency < 500ms for 24-hour time ranges
• •90-day retention for raw data, 2-year for hourly rollups
• •Storage efficiency: target < 2 bytes/data point after compression

Constraints

• •Metrics are numeric time-series data (counters, gauges, histograms)
• •Write-heavy: metrics are ingested continuously from thousands of services
• •Queries are time-range based; no ad-hoc filtering by arbitrary tag combinations required in v1

Evaluation Criteria