Axiom
Overview

percentiles_array

You can use percentiles_array to:

  • Analyze latency or duration metrics across requests or operations.
  • Identify performance outliers.
  • Visualize percentile distributions in dashboards.

Usage#

Syntax#

percentiles_array(Field, Percentile1, Percentile2, ...)

Parameters#

  • Field is the name of the field for which you want to compute percentile values.
  • Percentile1, Percentile2, ... are numeric percentile values between 0 and 100.

Returns#

An array of numbers where each element is the value at the corresponding percentile.

Use case examples#

Use percentiles_array to understand the spread of request durations per HTTP method, highlighting performance variability.

Query

['sample-http-logs']
| summarize percentiles_array(req_duration_ms, 25, 50, 95) by method

Run in Playground

Output

method P25 P50 P95
GET 0.3981 ms 0.7352 ms 1.981 ms
POST 0.3261 ms 0.7162 ms 2.341 ms
PUT 0.3324 ms 0.7772 ms 1.341 ms
DELETE 0.2332 ms 0.4652 ms 1.121 ms

This query calculates the 25th, 50th, and 95th percentiles of request durations for each HTTP method. It helps identify performance differences between different methods.

Use percentiles_array to analyze the distribution of span durations by service to detect potential bottlenecks.

Query

['otel-demo-traces']
| summarize percentiles_array(duration, 50, 90, 99) by ['service.name']

Run in Playground

Output

service.name P50 P90 P99 P99
recommendationservice 1.96 ms 2.965 ms 3.477 ms 3.477 ms
frontendproxy 3.767 ms 13.101 ms 39.735 ms 39.735 ms
shippingservice 2.119 ms 3.085 ms 9.739 ms 9.739 ms
checkoutservice 1.454 ms 12.342 ms 29.542 ms 29.542 ms

This query shows latency patterns across services by computing the median, 90th, and 99th percentile of span durations.

Use percentiles_array to assess outlier response times per status code, which can reveal abnormal activity or service issues.

Query

['sample-http-logs']
| summarize percentiles_array(req_duration_ms, 50, 95, 99) by status

Run in Playground

Output

status P50 P95 P99
200 0.7352 ms 1.981 ms 2.612 ms
201 0.7856 ms 1.356 ms 2.234 ms
301 0.8956 ms 1.547 ms 2.546 ms
500 0.6587 ms 1.856 ms 2.856 ms

This query helps identify whether requests resulting in errors (like 500) are significantly slower than successful ones.

  • avg: Returns the average value. Use it when a single central tendency is sufficient.
  • percentile: Returns a single percentile value. Use it when you only need one percentile.
  • percentile_if: Returns a single percentile value for the records that satisfy a condition.
  • percentiles_arrayif: Returns an array of percentile values for the records that satisfy a condition.
  • sum: Returns the sum of a numeric column.

Other query languages#

Splunk SPL users

In Splunk, you typically calculate percentiles one at a time using the perc function. To get multiple percentiles, you repeat the function with different percentile values. In APL, percentiles_array lets you specify multiple percentiles in a single function call and returns them as an array.

Splunk example

... | stats perc95(duration), perc50(duration), perc25(duration) by service

APL equivalent

['otel-demo-traces']
| summarize percentiles_array(duration, 25, 50, 95) by ['service.name']
ANSI SQL users

Standard SQL typically lacks a built-in function to calculate multiple percentiles in a single operation. Instead, you use PERCENTILE_CONT or PERCENTILE_DISC with WITHIN GROUP, repeated for each desired percentile. In APL, percentiles_array simplifies this with a single function call that returns all requested percentiles as an array.

SQL example

SELECT
  service,
  PERCENTILE_CONT(0.25) WITHIN GROUP (ORDER BY duration) AS p25,
  PERCENTILE_CONT(0.50) WITHIN GROUP (ORDER BY duration) AS p50,
  PERCENTILE_CONT(0.95) WITHIN GROUP (ORDER BY duration) AS p95
FROM traces
GROUP BY service

APL equivalent

['otel-demo-traces']
| summarize percentiles_array(duration, 25, 50, 95) by ['service.name']

Updated

Was this page helpful?