Metrics Monitoring System
Difficulty: Hard
What It Tests
Metric ingestion at scale, time-series storage, alerting, and query efficiency.
Topics Covered
- Time Series Databases
- Stream Processing
- Caching
- Big Data Architectures
- Distributed Counting / Top K
- Scaling Fundamentals
Hello Interview Breakdown
Read the full Hello Interview breakdown →
Approach Hints
- Agents push metrics to Kafka (pull model with Prometheus scrape is an alternative); stream processor aggregates 1-minute windows into a time-series DB
- Pre-aggregate at multiple granularities (1-min, 1-hour, 1-day) and apply retention policies (raw data: 7 days, hourly: 3 months, daily: 2 years)
- Alert rules evaluated continuously against rolling windows; alert pipeline deduplicates and groups related alerts to prevent alert storms
- Query engine uses columnar storage and time-range partitioning for scan efficiency; cache frequently accessed dashboards in Redis