Autoscaler Policies for ASG, ECS, VMSS and MIG
Scaling policies drift, get copied between teams, or were never set. ZopNight treats them as something to manage, with a way back.
What you can manage
ZopNight autoscaler policies cover four targets: AWS Auto Scaling groups, AWS ECS services, Azure VM scale sets and GCP managed instance groups. A policy sets minimum and maximum instances, a cooldown, the scaling mode and its triggers. Triggers can be CPU, memory, request count, latency, error rate, queue depth, custom metrics from CloudWatch, Cloud Monitoring or Azure Monitor, or step scaling, depending on what each target supports. Time-based triggers come with presets for business hours, peak hours and weekend scale-down.
Quick Setup fills in min, max, target and cooldown from the resource’s own metrics (average, spread and high percentiles of CPU), so a first policy is one form.
Monitor, recommend, autopilot
The mode follows the cloud account’s permission. A read-only account gets monitor mode: ZopNight observes and suggests. A read + write account can run recommend mode, where you apply suggestions by hand, or autopilot, where changes are executed for you.
Existing policies are not overwritten
When a target already has scaling configuration in the cloud, you choose explicitly:
- Adopt records the existing policy and makes no cloud change.
- Replace captures the full existing configuration, then writes ZopNight’s policy. Removing the ZopNight policy later restores what was there.
Replace is refused for configurations that could not be restored exactly, such as predictive scaling or step policies with many steps, and those are shown as not editable. Editing an adopted policy converts it to a managed one and prompts you to re-apply.
Every policy can be applied, paused, resumed and removed, and each action is logged with its status.
Findings that point here
Autoscaling-related recommendations include ASG Has No Scaling Policies on AWS, VMSS Autoscale Setting Not Configured, Scaling Target Too High and Scaling Cooldown Too Short on Azure scale sets, and ASG Floor Lock Detected, where the minimum equals desired capacity so the group cannot scale in during quiet hours. Several apply in one click with a configure or tune action.
Related: event readiness for planned spikes, the service pages for AWS Auto Scaling groups and Azure VM scale sets, and background in target tracking scaling.