Skip to main content

OpenMetrics Cardinality

Info

Purpose of the OpenMetrics Cardinality Document

After storing metrics exposed in Prometheus format, you can directly calculate cardinality per metric using a simple script.
This allows you to see how many different label combinations each metric has in your environment.

What is Cardinality?​

Cardinality refers to the number of unique label combinations (excluding values) for a specific metric. Even if the metric name is the same, data with identical label composition is treated as one.

Example

00:00   http_requests_total{url="/hello"}                                         Value: 100
00:00   http_requests_total{url="/hello", name="whatap"}          Value: 93
00:01   http_requests_total{url="/hello", name="whatap"}          Value: 114
00:01   http_requests_total{url="/hello"}                                         Value: 5

Although there are four data points, based on label composition they are counted as two:

  • url="/hello" → 1
  • url="/hello", name="whatap" → 1

Thus, the cardinality of this metric is 2.

How to Check Cardinality​

To check the cardinality of actual metrics collected in your environment, calculate it based on metrics exposed in Prometheus format.

Step 1. Extract Prometheus Metric File​

Call the endpoint (/metrics) that exposes metrics in Prometheus format and save it as a text file.
Port 9100 is an example for Prometheus Node Exporter; use the endpoint address suitable for your environment.

curl -s http://localhost:9100/metrics > metrics.txt

Step 2. Check Metric Format​

The saved file (metrics.txt) usually looks like the following. Here, the basis for calculating cardinality is metric_name + label combination.

metric_name{label1="value1",label2="value2"} 123

Step 3. Calculate Cardinality​

Run the following command to calculate the number of unique label combinations for each metric (metric_name).

Download the shell script and execute the following commands. Download count_cardinality.sh 

chmod +x count_cardinality.sh
./count_cardinality.sh metrics.txt

If you cannot download and run the shell script in your environment, copy and execute the following commands directly:

grep -v '^#' metrics.txt \
  | awk -F'{' '{
      if (NF > 1) {
        name=$1
      } else {
        split($1, arr, " ")
        name=arr[1]
      }
      count[name]++
      total++
    }
    END {
      for (n in count) {
        print n, count[n]
      }
      print "Total", total
    }' | sort

Step 4. Interpret Results​

The execution result will look like the following. It shows the cardinality count of each metric and the total cardinality.

apiserver_flowcontrol_priority_level_request_utilization_bucket 44
go_gc_duration_seconds_count 1
go_gc_duration_seconds_sum 1
process_cpu_seconds_total 1
Total 47
  • apiserver_flowcontrol_priority_level_request_utilization_bucket: 44 unique label combinations
  • go_gc_duration_seconds_count: 1 unique label combination
  • go_gc_duration_seconds_sum: 1 unique label combination
  • process_cpu_seconds_total: 1 unique label combination
  • Total: 47 (overall cardinality count)
Last updated on