FleetAutoScaler scalingLimited is reporting the incorrect state

[Dinhh1]

What happened:
It seems as though the scalingLimited attribute of an autoscaler is incorrectly reporting it's state.

What you expected to happen:
I expect scalingLimited = false when the number of gameservers < the maxReplicas

How to reproduce it (as minimally and precisely as possible):

1. Configure an autoscaler with: max - 50, min - 10, buffer - 5
2. Allocate 1 or 2 GameServers
3. Observer how scalingLimited is set to true even though we're below our max replica threshold
   

Anything else we need to know?:

Environment:

• Agones version: 1.26
• Kubernetes version (use kubectl version): 1.24
• Cloud provider or hardware configuration: AWS EKS
• Install method (yaml/helm): helm
• Troubleshooting guide log(s):
• Others: Original Slack Thread: https://agones.slack.com/archives/C9DGM5DS8/p1668793997712099

[oniku-2929]

I also met the same situation in my EKS and Minikube environment.
Then I did a little research and test about it.

In conclusion, I guess it's expected behavior.
Because this flag is exposed as a "scale-in & scale-out" Limited flag.
I misunderstood it's a "scale-out" Limited flag.

In pkg/fleetautoscalers/fleetautoscalers.go func applyBufferPolicy

agones/pkg/fleetautoscalers/fleetautoscalers.go

Lines 193 to 226 in 9f430b8

	func applyBufferPolicy(b *autoscalingv1.BufferPolicy, f *agonesv1.Fleet) (int32, bool, error) {
	var replicas int32
	if b.BufferSize.Type == intstr.Int {
	replicas = f.Status.AllocatedReplicas + int32(b.BufferSize.IntValue())
	} else {
	// the percentage value is a little more complex, as we can't apply
	// the desired percentage to any current value, but to the future one
	// Example: we have 8 allocated replicas, 10 total replicas and bufferSize set to 30%
	// 30% means that we must have 30% ready instances in the fleet
	// Right now there are 20%, so we must increase the fleet until we reach 30%
	// To compute the new size, we start from the other end: if ready must be 30%
	// it means that allocated must be 70% and adjust the fleet size to make that true.
	bufferPercent, err := intstr.GetValueFromIntOrPercent(&b.BufferSize, 100, true)
	if err != nil {
	return 0, false, err
	}
	// use Math.Ceil to round the result up
	replicas = int32(math.Ceil(float64(f.Status.AllocatedReplicas*100) / float64(100-bufferPercent)))
	}
	limited := false
	if replicas < b.MinReplicas {
	replicas = b.MinReplicas
	limited = true
	}
	if replicas > b.MaxReplicas {
	replicas = b.MaxReplicas
	limited = true
	}
	return replicas, limited, nil
	}

This limited flag is exposed as a Prometheus metric "agones_fleet_autoscalers_limited" and then we can see it as a "Scaling Limited" flag on Grafana.

In the above situation,

1.Configure an autoscaler with: max - 50, min - 10, buffer - 5
2.Allocate 1 or 2 GameServers

Initially, the replicas count is 0(no allocation) + 5 = 5.
then replicas fix to 10(MinReplicas) and will be limited for avoiding scale-in.

still, the same adjustment happens after 1 or 2 GameServers allocations.
So "Scaling Limited" is True.

on "Metrics" Page.
https://agones.dev/site/docs/guides/metrics/

agones_fleet_autoscalers_limited The fleet autoscaler is capped (1)

I think it would be helpful to add "scale-in or scale-out is limited based on the fleet autoscaler's setting" in the description for avoiding such a misunderstanding.

I couldn't see the Slack Thread that is mentioned above.
So I don't understand any details about it on Slack.

[atgane]

Hi, may I handle this issue? I'll create PR soon!

[markmandel]

Reading above, I'm not 100% sure if this is actually an issue 🤔

That being said, please investigate and let us know if you can confirm either way.

[markmandel]

Or to be clearer - are you looking to update the code or the docs? (Sounds like the docs do need an update!)

[atgane]

I will update docs!