# Qualify predictive scale-out without removing reactive autoscale

> Which workload and control conditions must hold before enabling predictive autoscale for a scale set?

- Canonical URL: https://update.dsesecurity.com/updates/dse-20260909-241-qualify-predictive-scale-out-without-removing-reactive-autoscale/
- Publisher: Detection Systems & Engineering (DSE Security)
- Author: DSE Security Editorial Team
- Published: 2026-09-10T00:27:55+00:00
- Modified: 2026-09-10T01:20:46+00:00
- Last reviewed by DSE: 2026-09-09
- Resource type: Guide
- DSE priority: Information
- Topics: Business Continuity, Networks & Infrastructure
- Reading time: 2 minutes

## What you need to know

Which workload and control conditions must hold before enabling predictive autoscale for a scale set?

## Potentially affected

Azure Commercial virtual machine scale sets with cyclical CPU demand.

## DSE recommendation

Evaluate forecast-only behavior against workload evidence while preserving reactive scale-out and an explicit scale-in policy.

## Article

## Source facts

Predictive autoscale supports Azure Commercial virtual machine scale sets with cyclical CPU demand. Its supported signal is Percentage CPU aggregated as Average. It only adds instances; standard autoscale must handle scale-in. Predictions require at least seven days of history, while the sampling window extends to fifteen days. [Microsoft Learn](https://learn.microsoft.com/en-us/azure/azure-monitor/autoscale/autoscale-predictive).

Reactive autoscale conditions must be configured before enabling either prediction-based scaling or forecast-only mode. Forecast-only displays predictions without performing predictive scaling. Standard rules remain the fallback for unexpected demand or unavailable predictive data. [Microsoft Learn](https://learn.microsoft.com/en-us/azure/azure-monitor/autoscale/autoscale-predictive).

## Applicability

Limit this decision to the documented scale-set and CPU scenario. Microsoft directs monthly or annual workload patterns toward scheduled or metric-based scaling instead. Do not treat a newly created scale set’s training period as a completed forecast evaluation. [Microsoft Learn](https://learn.microsoft.com/en-us/azure/azure-monitor/autoscale/autoscale-predictive).

## DSE recommendation

DSE recommends beginning with forecast-only and a written activation decision. Record the recurring load pattern, acceptable lead time and approved capacity bounds. Keep responsibility for scale-in explicit. Ask the workload owner to identify exceptional demand that should be handled reactively rather than assumed predictable.

## Verification

Compare predicted and observed demand across representative cycles before approving activation. Recheck the reactive rule and scale-in configuration separately from forecast accuracy. Preserve the comparison and any rejected forecast periods, then verify actual instance behavior after an approved limited enablement. Do not label model availability alone as application readiness.

## Official references

[Microsoft Learn: Predictive autoscale](https://learn.microsoft.com/en-us/azure/azure-monitor/autoscale/autoscale-predictive).

## Primary reference

- Name: Use predictive autoscale to scale out before load demands in virtual machine scale sets - Azure Monitor | Microsoft Learn
- Authority: Microsoft Learn
- URL: https://learn.microsoft.com/en-us/azure/azure-monitor/autoscale/autoscale-predictive
- Source publication date: Not stated by the source

## Citation and use

Preferred citation: “Qualify predictive scale-out without removing reactive autoscale,” DSE Security, https://update.dsesecurity.com/updates/dse-20260909-241-qualify-predictive-scale-out-without-removing-reactive-autoscale/
Publishing principles: https://update.dsesecurity.com/updates/dse-updates-editorial-methodology/
Usage and citation policy: https://update.dsesecurity.com/usage/
Copyright © 2026 Detection Systems & Engineering. All rights reserved.