Design a Metrics Monitoring & Alerting System
Medium · System Design · about 45 minutes
Asked at NVIDIA, Intel, Cisco, Oracle, Google, Roblox, AMD, Waymo, Broadcom, Arista Networks, Epic Games, Unity, SpaceX, Dell Technologies, Micron, Nutanix.
Design a monitoring system like Prometheus or Datadog that collects metrics from ~500k hosts or GPU nodes (~10M data points per second), stores them as time series for years, serves dashboard queries in under a second, and pages engineers within about a minute when an alert condition holds — even during a major outage.