Course contentsShow
← System Design
Lesson 1208 of 1,91930. Observability: MetricsFree lesson

Choosing a Metrics System for Your Scale

Compare metrics systems based on scale, query patterns, retention needs, and operational complexity requirements.

Choosing a Metrics System for Your Scale

What you'll learn: How to select the right metrics system by evaluating scale, query patterns, retention needs, and operational complexity.

The Selection Challenge

Not every metrics system fits every use case. A startup with 10 servers has vastly different needs than a company running 10,000 containers. Choosing the wrong system creates technical debt—either overpaying for features you don't need or outgrowing your infrastructure as you scale.

Think of it like choosing a vehicle: a bicycle works for neighborhood trips, but you need a semi-truck for cross-country freight. Similarly, your metrics system must match your operational "load."

Key Evaluation Dimensions

Scale and Volume

How many metrics per second? StatsD with Graphite handles moderate loads well, while Prometheus excels at hundreds of thousands of time series but may need federation beyond single-datacenter deployments. InfluxDB offers strong write performance for IoT-scale ingestion. Cloud-native options (CloudWatch, Stackdriver) scale elastically but at higher cost.

Query Patterns

Do you run complex aggregations or simple dashboards? PromQL in Prometheus offers powerful ad-hoc queries, while Graphite provides simpler but less flexible querying. If you primarily view pre-built dashboards, even basic systems suffice.

Retention Needs

Short-term (days) or long-term (months/years)? Prometheus retains locally for weeks; longer retention requires long-term storage solutions or downsampling. Cloud systems handle retention transparently but increase costs over time.

Operational Complexity

Can you maintain clusters? Prometheus requires less operational overhead than distributed InfluxDB clusters. OpenTelemetry with vendor backends minimizes self-hosting burden. StatsD + Graphite needs more configuration but offers fine-grained control.

Decision Framework

Start simple (Prometheus for Kubernetes, CloudWatch for AWS). Grow into distributed systems only when cardinality exceeds millions or query latency degrades.

Key Takeaway: Choose metrics systems by matching your current scale, acceptable query complexity, retention duration, and team's operational capacity—not by chasing features you don't yet need.