make-series

This page explains how to use the make-series operator in APL.

Introduction

The make-series operator transforms event data into array-based time series. Instead of producing one row per time bucket, make-series encodes the values and corresponding timestamps into arrays stored in table fields. This makes it possible to apply series_* functions for advanced manipulations such as moving averages, smoothing, anomaly detection, or other time-series computations.

You find this operator useful when you want to:

Turn event data into array-encoded time series for further analysis.
Apply series_* functions (for example, series_fir, series_stats) to aggregated data.
Postprocess and then expand arrays back into rows with mv-expand for visualization or downstream queries.

Unlike summarize, which produces row-based aggregations, make-series is designed specifically for creating and manipulating array-based time series.

Usage

Syntax

APL

make-series [Aggregation [, ...]]
    [default = DefaultValue]
    on TimeField
    [in Range]
    step StepSize
    [by GroupingField [, ...]]

Parameters

Parameter	Description
`Aggregation`	One or more aggregation functions (for example, `avg()`, `count()`, `sum()`) applied to each time bin, producing arrays of values.
`default`	A value to use when no records exist in a time bin.
`TimeField`	The field containing timestamps used for binning.
`Range`	An optional range expression specifying the start and end of the series (for example, `from ago(1h) to now()`).
`StepSize`	The size of each time bin (for example, `1m`, `5m`, `1h`).
`GroupingField`	Optional fields to split the series by, producing parallel arrays for each group.

Returns

The operator returns a table where each aggregation produces an array of values aligned with an array of time bins. Each row represents a group (if specified), with arrays that encode the entire time series.

Use case examples

You want to create an array-based time series of request counts, then compute a rolling average using a series_* function, and finally expand back into rows for visualization.

Query

APLRun in Playground

['sample-http-logs']
| make-series count() on _time from now()-24h to now() step 5m
| extend moving_avg_count=series_fir(count_, dynamic([1, 1, 1, 1, 1]))
| mv-expand moving_avg_count to typeof(long), count_ to typeof(long), time to typeof(datetime)
| project-rename _time=time
| summarize avg(moving_avg_count), avg(count_) by bin(_time, 5m)

Output

_time	count_	moving_avg_count
2025-09-29T10:00:00	120	118
2025-09-29T10:05:00	130	122
2025-09-29T10:10:00	110	121

The query turns request counts into arrays, applies a smoothing function, and then expands the arrays back into rows for analysis.

You want to analyze span durations per service, storing them as arrays for later manipulation.

Query

APLRun in Playground

['otel-demo-traces']
| make-series avg(duration) on _time from ago(2h) to now() step 10m by ['service.name']

Output

service.name	avg_duration	time
frontend	[20ms, 18ms, 22ms, 19ms, ...]	[2025-09-29T08:00, ...]
checkout	[35ms, 40ms, 33ms, 37ms, ...]	[2025-09-29T08:00, ...]

The query produces array-encoded time series per service, which you can further process with series_* functions.

You want to analyze the rate of HTTP 500 errors in your logs per minute.

Query

APLRun in Playground

['sample-http-logs']
| where status == '500'
| make-series count() default=0 on _time from ago(30m) to now() step 1m

Output

count_	_time
[1489, 1428, 1517, 1462, 1509, ...]	["2025-09-30T09:08:14.921301725Z", "2025-09-30T09:09:14.921301725Z", ...]

The query generates a time series of HTTP 500 error counts as an array-based time series for further analysis with series_* functions.

extend: Creates new calculated fields, often as preparation before make-series. Use extend when you want to preprocess data for time series analysis.
mv-expand: Expands arrays into multiple rows. Use mv-expand to work with the arrays returned by make-series.
summarize: Aggregates rows into groups but doesn't generate continuous time bins. Use summarize when you want flexible grouping without forcing evenly spaced intervals.
top: Returns the top rows by a specified expression, not time series. Use top when you want to focus on the most significant values instead of trends over time.