Skip to content

stddev

Input Stack:
expr: TimeSeriesExpr
 ⇨ Output Stack:
TimeSeriesExpr

Compute the standard deviation across multiple time series from a grouped expression. This statistical measure shows how much variation exists from the average value within each group, useful for understanding data spread and identifying outliers.

Parameters

  • expr: A grouped time series expression (typically the result of :by)

Behavior

  • Group processing: Operates on multiple time series within each group
  • Statistical calculation: Computes population standard deviation across series
  • Variance measure: Shows spread around the mean value for each timestamp
  • NaN handling: Excludes NaN values from calculation

Algorithm

For each timestamp, calculates the standard deviation across all non-NaN values from the different time series in the group using the computational formula:

σ = √((N * sum(x²) - sum(x)²) / N²)

Where:

  • N = number of non-NaN values at that timestamp
  • sum(x) = sum of all time series values at that timestamp
  • sum(x²) = sum of squared time series values at that timestamp

This formula is mathematically equivalent to the standard definition but is computed using aggregated statistics rather than storing individual samples.

Examples

Computing standard deviation for request rates grouped by cluster:

BeforeAfter
name,sps,:eq,
(,nf.cluster,),:by

name,sps,:eq,
(,nf.cluster,),:by,
:stddev

Alternative for Distribution Metrics

For timer and distribution summary metrics, use :dist-stddev instead, which calculates standard deviation from the recorded samples rather than across multiple time series.

  • :dist-stddev - Standard deviation for timer/distribution metrics (recommended for those types)
  • :avg - Mean value across time series (center measure)
  • :max / :min - Range measures (spread indicators)
  • :by - Group time series for statistical analysis
  • :sum - Total aggregation across series

Since: 1.6