Variations
The Variations tab breaks down how label combinations affect a metric's volume and cardinality.
Use this view to identify which label sets consume the most storage and where to optimize.
Total labels
Lists all label keys found across the metric's variations.
Examples:
__name__, process_tags_k8s_pod_name, operation, grpcStatusCode, serviceVersion, transactionRoot.
- Shows the total number of labels (for example, 20 labels)
- Displays each label as a visual tag
- Includes both application and infrastructure labels (for example, Kubernetes
pod_name,deployment_name)
Note
High label diversity increases unique series, storage, and billing. Review regularly to optimize. Metrics with multiple variations (different label combinations) require extra care at query time. Always use specific label filters in your queries to target the correct variation and avoid unexpected or inconsistent results.
Variations table
Each row represents a unique label set (variation). Hover to view the full label list.
Columns:
- Variation: Label combination
- Labels: Labels used versus total (for example, 18/20)
- Usage: Data volume and units
- % Usage: Share of metric usage
- Samples: Datapoints ingested
- Cardinality: Unique series generated
- % Cardinality: Share of total series
Example:
- The main variation uses 18/20 labels, consumes 3.33 GB (99.4%), and generates 8,688 series.
- Two smaller variations use < 0.01 U each but still add thousands of series.
Note
Focus optimization on high-volume or high-cardinality variations.
Variation label overview
The Variations view highlights which labels are used in a selected variation.
- Highlighted labels: labels included in the selected variation
- Faded labels: labels that exist on the metric but don't apply to this variation
This helps you see how label sets differ across variations and quickly identify which labels contribute to extra cardinality.
Note
When you see multiple variations for a metric, always filter by label in your queries to get accurate results.
Use the Variations view to:
- Search by label key (for example,
namespace=prod) - Compare high-cardinality versus high-usage sets
- Drill down into label-level impact
- Optimize redundant or low-value combinations
Aim for one variation per metric to simplify queries and reduce cardinality.