Metric Naming and Labels
You'll learn how to name metrics consistently and use labels effectively for powerful drill-down analysis without creating a maintenance nightmare.
The Core Idea
Metric names and labels are like a filing system for your operational data. A metric name identifies what you're measuring (e.g., http_requests_total), while labels (also called tags or dimensions) add who, where, or how context (e.g., method=GET, status=200, endpoint=/api/users). Good naming creates queryable patterns; smart labeling enables dimensional analysis without exploding the number of unique metrics.
Naming Conventions
Use clear, hierarchical names that group related metrics:
http_requests_total✓http_request_duration_seconds✓RequestCount✗ (inconsistent)
Common patterns:
- Prefix by subsystem:
http_,db_,cache_ - Suffix by unit:
_total,_seconds,_bytes - Use underscores, not camelCase: easier to read in dashboards
Labels: Power and Peril
Labels let you slice one metric many ways:
http_requests_total{method="GET", status="200", endpoint="/users"}
http_requests_total{method="POST", status="500", endpoint="/login"}
Cardinality matters: Each unique label combination creates a new time series. High-cardinality labels (user IDs, timestamps) can explode storage costs. Keep cardinality bounded—labels should represent dimensions you'll actually query, not infinite values.
Think of labels as spreadsheet columns: method with 5 values and status with 10 values = 50 combinations. Add user_id with 1 million values? Now you have a problem.
Key Takeaway:
Name metrics with consistent, hierarchical patterns and use low-cardinality labels for dimensional filtering—high-cardinality values belong in logs, not metric labels.