​
Overview

An autoscaling group manages a pool of machines instead of individual ones. You define a metric, thresholds, and how aggressively to react, and the platform adds or removes members within the bounds you set.
Navigation: Compute → AutoScaling Groups. Titled AutoScaling Groups, subtitled “Automatically scale VMs based on resource metrics”.
An autoscaling group requires an existing load balancer rule — it is the entry point traffic arrives on, and the wizard cannot complete without one. Load balancing is switched off in the current phase (see Public IPs), so on most deployments there will be no rule to select and this feature is effectively unavailable. Confirm load balancing is enabled for your account before planning around it.

​
Columns

Summary cards show Total Groups, Enabled, and Disabled. There’s a search box and a All AutoScale Groups ({n}) heading.
This screen has no filter dropdown and no sortable columns — unusual compared with the rest of Compute. Use the search box.

​
Row actions

Deleting asks for confirmation and spells out the consequence worth knowing: machines the group created are not deleted with it — they keep running and become unmanaged.
Disable is the safe way to stop unexpected scaling during an incident. It stops the platform reacting to the metric without destroying the group’s configuration, so you can re-enable once you understand what happened.

​
The detail page

A single page rather than tabs. Breadcrumb Compute / AutoScaling Groups / {name}.

​
How scaling decides

Four settings govern the behaviour, and they interact:
A short Duration with a short Cooldown produces flapping: the group adds a machine, the metric drops because load spread, the group removes it, load concentrates again, and it repeats. Keep scale-down Duration and Cooldown longer than scale-up — it’s better to shed capacity slowly than to thrash.

​
The loop over time

The decision above repeats on every interval, and the two timers are what stop it acting on every reading.

​
Creating a group

Five steps: Metric, Conditions, Policies, VM Profile, VM Group.
1

Metric

Metric (Counter) selects what to scale on, from the counters the platform offers. If counters can’t be loaded you can enter a counter identifier directly.
2

Conditions

Leave a wide gap between the two. Thresholds close together — say 60 and 55 — leave almost no stable band, so normal variation crosses both and the group never settles.
3

Policies

Under Scale Up Policy and Scale Down Policy:The asymmetric defaults are deliberate — react to load in two minutes, but wait five before shedding.
4

VM Profile

Zone, Compute Offering, and Template — the shape of each machine the group creates. Every member is identical.
Members are built from a template with no per-machine customisation step, so whatever the machine needs to do must already be in the template or must configure itself on boot. There’s no opportunity to log in and set a new member up by hand.
5

VM Group

Success confirms with AutoScale group "{name}" created.
Max VMs is the only thing bounding your spend. A metric that misbehaves will scale to the ceiling and hold there. Set it to what you can actually afford, and remember group members consume the same quota as machines you create by hand — hitting a quota ceiling stops scaling regardless of what Max says.

​
Worked example: a web tier that handles a daily peak

1

Sizing

Traffic needs 2 machines overnight and 6 at peak. Set Min VMs 2, Max VMs 8 — headroom above the known peak, capped well below runaway.
2

Thresholds

Scale-Up 70, Scale-Down 30. A 40-point band absorbs normal variation without triggering.
3

Timing

Interval 60. Scale-up Duration 120 with Cooldown 300; scale-down Duration 600 with Cooldown 600. Two minutes of sustained load adds capacity; ten minutes of quiet removes it.
4

Verify before trusting it

Watch VMs Now on the detail page across one full daily cycle. If the count changes more than a couple of times per direction per day, lengthen the durations.

​
Empty and error states

The detail page has no single-group endpoint behind it — it fetches the whole list and filters client-side. On an account with many groups the page may take a moment longer than expected.

Virtual Machines

Individually managed machines.

Quotas and limits

Scaling stops at your quota, whatever Max VMs says.

Public IPs

Why load balancer rules may not exist yet.

Templates

Members boot from a template with no manual setup.