Axiom
Overview

series_fill_backward

You can use series_fill_backward when you have time series data with missing values and want to fill gaps using the last observed value. This is particularly useful for forward-looking analysis, forecasting scenarios, or when the most recent data point is the best estimate for missing earlier values. Typical applications include financial data analysis, sensor data processing, and performance monitoring where recent values are more relevant.

Usage#

Syntax#

series_fill_backward(array)

Parameters#

Parameter Type Description
array dynamic A dynamic array of numeric values that may contain null values.

Returns#

A dynamic array where null values are replaced by the last non-null value encountered when traversing the array backward.

Use case examples#

In log analysis, you can use series_fill_backward to fill missing request duration data using the most recent available values, which is useful for analyzing performance trends.

Query

['sample-http-logs']
| summarize durations = make_list(req_duration_ms) by id
| extend filled_durations = series_fill_backward(durations)

Run in Playground

Output

id durations filled_durations
u123 [null, 150, null, 200] [150, 150, 150, 200]
u456 [100, null, null, 300] [100, 300, 300, 300]

This query fills missing request durations with the most recent available values, useful for maintaining continuity in performance analysis.

In OpenTelemetry traces, you can use series_fill_backward to fill missing span duration data using the most recent observed values for better trace analysis.

Query

['otel-demo-traces']
| summarize durations = make_list(duration) by ['service.name']
| extend filled_durations = series_fill_backward(durations)

Run in Playground

Output

service.name durations filled_durations
frontend [null, 100ms, null, 200ms] [100ms, 100ms, 100ms, 200ms]
productcatalogservice [50ms, null, null, 150ms] [50ms, 150ms, 150ms, 150ms]

This query fills missing span durations with the most recent available values, useful for maintaining continuity in service performance analysis.

In security logs, you can use series_fill_backward to fill missing request duration data using the most recent values for consistent security analysis.

Query

['sample-http-logs']
| summarize durations = make_list(req_duration_ms) by status
| extend filled_durations = series_fill_backward(durations)

Run in Playground

Output

status durations filled_durations
200 [null, 150, null, 250] [150, 150, 150, 250]
500 [100, null, null, 400] [100, 400, 400, 400]

This query fills missing request durations with the most recent available values grouped by status code, useful for consistent security analysis across different response types.

  • series_fill_forward: Fills missing values by propagating the first known value forward. Use when you want to use the earliest available value to fill gaps.
  • series_fill_const: Fills missing values with a constant value. Use when you want to replace nulls with a specific default value.
  • series_fill_linear: Fills missing values using linear interpolation. Use when you want smooth transitions between known values.
  • series_equals: Compares each element to a specified value. Use for identifying specific values after filling operations.
  • series_greater: Returns elements greater than a specified value. Use for threshold analysis after filling missing data.

Other query languages#

Splunk SPL users

In Splunk SPL, filling missing values typically requires complex eval expressions with fillnull or custom logic using streamstats and filldown. The backward filling approach is less common and usually requires manual implementation. In APL, series_fill_backward provides a direct, efficient way to perform backward filling on dynamic arrays.

Splunk example

... | fillnull value=0 | streamstats window=5 current=f last(field) as filled_field

APL equivalent

datatable(values: dynamic)
[
  dynamic([null, null, 100, null, 200])
]
| extend filled_values = series_fill_backward(values)
ANSI SQL users

In SQL, filling missing values backward requires complex window functions with LAG() or custom logic using LAST_VALUE() with specific window specifications. Most SQL implementations focus on forward filling rather than backward filling. In APL, series_fill_backward simplifies this operation by directly handling backward propagation of values in arrays.

SQL example

SELECT LAST_VALUE(value) OVER (ORDER BY timestamp ROWS BETWEEN CURRENT ROW AND UNBOUNDED FOLLOWING) AS filled_value
FROM measurements;

APL equivalent

datatable(values: dynamic)
[
  dynamic([null, null, 100, null, 200])
]
| extend filled_values = series_fill_backward(values)

Updated

Was this page helpful?