Adjust HorizontalPodAutoscaler settings for workloads #2158
Description
Problem
We're running relatively small Knative setups and we have configured lower CPU & memory requests and limits to keep the overhead small. Currently, spikes in Knative usage cause the HorizontalPodAutoscalers to scale up very aggressively. This is because our CPU requests are low and it's very easy to reach the 100% averageUtilization that knative-operator currently configures to HPA objects. And because the default value for maxReplicas is high, we suddenly get a lot more pods for a smallish traffic spike.
The operator should expose a way to configure the HPA targets (averageUtilization) and maxReplicas so that the autoscaling behavior can be limited.
Persona:
System Operator
Exit Criteria
Knative operator exposes settings to configure averageUtilization and/or maxReplicas in the HPA objects created for Knative serving components.
Time Estimate (optional):
Additional context (optional)