Skip to content

node-avg

Input Stack:
query: Query
 ⇨ Output Stack:
TimeSeriesExpr

Compute an average per node by using the poller.asg.instance metric as the denominator. This specialized aggregation function accounts for the actual number of active instances rather than just averaging using the number of time series. It automatically restricts the scope using common infrastructure tags to ensure proper normalization.

Parameters

  • query: The query to compute the per-node average for

Purpose

This operator should be used instead of :avg when you need a true per-node average that accounts for the actual instance count. Regular :avg computes the mean across all matching time series, but :node-avg divides by the actual number of reporting instances.

The key difference: if you have 3 instances but each reports metrics with multiple dimensions (creating more time series), :avg would divide by the total number of time series while :node-avg divides by 3 (the actual instance count).

How It Works

  1. Executes the provided query to get the numerator values
  2. Finds the corresponding poller.asg.instance metrics using infrastructure tags
  3. Divides the query results by the instance count to get per-node averages
  4. Uses common tags (like nf.app, nf.cluster) to properly scope the calculation

Examples

Computing average requests per second per node:

name,sps,:eq,
nf.app,nccp,:eq,
:and,
:node-avg
  • :avg - Standard average aggregation (doesn't normalize by node count)
  • :eureka-avg - Average using nodes UP in Eureka for the denominator
  • :dist-avg - Average over the number of recorded samples