Skip to main content

​
Supported Data Types

📊 Metrics

​
Supported Metric Types

Metric TypeSupport LevelNotes
Gauge✅ Full
Sum (Counter)✅ Full
Histogram🧪 Alpha
Summary🧪 Alpha
ExponentialHistogram🧪 Alpha

​
Configuring the Sawmills Aggregate Metric Labels Processor

The Sawmills Aggregate Metric Labels Processor reduces metric cardinality by aggregating datapoints across specified labels. It removes selected labels and combines metric values using an aggregation function, resulting in fewer unique time series.

​
Configuration Components

​
1. Name

  • Description: Identifier for your processor. Use a unique and descriptive name to differentiate between multiple processors.

​
2. Metric Names

  • Description: List of metrics to apply label aggregation to. Select from available metrics in your pipeline.
  • Functionality: Only the specified metrics will have their labels aggregated. All selected metrics use the same aggregation function.

​
3. Aggregation Function

  • Description: The mathematical function used to combine datapoint values when labels are removed.
  • Options:
    • Sum: Add all values together
    • Average: Calculate the mean of all values
    • Max: Keep the maximum value
    • Min: Keep the minimum value
  • Use Cases:
    • Use Sum for counters and cumulative metrics (e.g., request counts, bytes transferred)
    • Use Average for gauge-type metrics (e.g., CPU utilization, memory percentage)
    • Use Max for capturing peak values (e.g., maximum latency, peak connections)
    • Use Min for capturing floor values (e.g., minimum available capacity)

​
4. Labels to Aggregate

  • Description: Select which labels to remove and aggregate across. Available labels are fetched based on the selected metrics.
  • Functionality: Datapoints that differ only by these labels will be combined into a single datapoint using the selected aggregation function.
  • Select All: Option to select all available labels for maximum cardinality reduction.

​
Use Cases

  • Cardinality Reduction: Remove high-cardinality labels like pod_name, instance_id, or container_id to reduce the number of unique time series.
  • Cost Optimization: Lower storage and query costs by reducing metric volume while preserving aggregate insights.
  • Environment Rollup: Aggregate metrics across replicas or instances to get environment-level totals.
  • Simplified Dashboards: Create cleaner visualizations by removing unnecessary dimensional detail.

​
Example

Before aggregation: 
http_requests_total{service="api", instance="pod-1"} = 100
http_requests_total{service="api", instance="pod-2"} = 150
http_requests_total{service="api", instance="pod-3"} = 120
After aggregating label instance with Sum: 
http_requests_total{service="api"} = 370

​
Implementation Notes

  • Label aggregation operates on 5-second windows to batch and combine datapoints.
  • Choose the aggregation function that matches your metric’s semantic meaning (counters → sum, gauges → avg).
  • Removing all labels results in a single aggregated value per metric name.