Labels and Dimensionality
What you'll learn: How to use Prometheus labels to transform flat metrics into multi-dimensional data that you can slice, filter, and aggregate in powerful ways.
The Core Idea
In Prometheus, labels are key-value pairs attached to metrics that add dimensions to your time-series data. Think of a basic metric like http_requests_total as a single number. Now imagine adding labels like {method="GET", status="200", endpoint="/api/users"} — suddenly that single metric becomes a rich, multi-dimensional dataset you can query from any angle.
Labels are what make Prometheus metrics flexible and powerful. Without them, you'd need a separate metric for every combination (like http_requests_get_200, http_requests_post_404, etc.). With labels, one metric name can represent thousands of unique time-series, each identified by its unique label combination.
How Dimensionality Works
Each unique combination of labels creates a distinct time-series. For example:
http_requests_total{method="GET", status="200"}
http_requests_total{method="POST", status="200"}
http_requests_total{method="GET", status="404"}
These are three separate time-series under one metric name. This dimensionality lets you:
- Filter: Show only
GETrequests - Group: Aggregate all
200statuses across methods - Pivot: Compare behavior across endpoints or regions
The Analogy
Think of labels like spreadsheet columns. A basic counter is just a single number. Labels turn it into a table where you can sort, filter, and create pivot tables — except Prometheus does this at massive scale over time.
Watch the Cardinality
Each label combination creates storage overhead. Labels like user_id or request_id with thousands of unique values create high cardinality, which can overwhelm Prometheus. Stick to labels with bounded, predictable values.
Key Takeaway: Labels transform metrics from simple numbers into multi-dimensional datasets, enabling flexible filtering and grouping — but choose dimensions wisely to avoid cardinality explosions.