Autoscaling you
can hand over
gradually.
Nobody sensible gives a new tool write access to production scaling on day one. Start in monitor mode and just watch. Move to recommend when the numbers look right. Switch to autopilot when you trust it.
Nothing is written until you say so.
Including the policies you already have.
Point at a target
An AWS Auto Scaling group, an AWS ECS service, an Azure virtual machine scale set or a GCP managed instance group. Quick Setup reads recent CPU and proposes min, max, target and cooldown from measured average, standard deviation and the P90, P95 and P99.
Adopt or replace
If the target already has cloud-side policies, you pick. Adopt observes the existing config with zero writes. Replace captures the whole thing first, and refuses outright on predictive scaling, step scaling beyond two adjustments, and customised metric specs it cannot rebuild. A refusal means zero mutation, not a half migration.
Choose how much rope
Monitor observes only. Recommend raises suggestions you apply by hand. Autopilot executes through the queue. Read-only cloud accounts get monitor mode automatically, because the permission level is detected rather than assumed.
Roll back exactly
The previous configuration is snapshotted on apply and restored byte for byte on Remove, including the opaque raw spec of every policy you had. An adopted policy is never written to, so there is nothing to undo.
Metrics, schedules and steps.
Availability is provider-aware, so the UI only offers what your target actually supports.
- Targets
aws:asgthrough EC2 Auto Scaling.aws:ecsthrough Application Auto Scaling, a separate API surface, with policies prefixed so Remove can identify them safely.azure:vmssthrough Autoscale Settings.gcp:migthrough the Autoscaler API.- Triggers
- CPU, memory, request count, latency, error rate, queue depth, custom CloudWatch, Cloud Monitoring and Azure Monitor metrics, plus step scaling with scale-out and scale-in adjustments.
- Schedules
- Time-based scaling with cron presets for business hours, peak hours and weekend scale-down, alongside the metric triggers rather than instead of them.
- Findings
- Five registered rules flag scaling groups with no policy at all, targets set above 90 percent or below 30 percent, and cooldowns under 120 seconds. Each one applies in a click.
Watch it for a month before it touches anything.
Monitor mode needs read-only access and writes nothing. Move up when the recommendations look right.