Why

This service has been designed to be compatible with future autoscale-to-zero logic in kubernetes itself. Until then - you have either use something like KEDA, or implement it by yourself

Known Limitations

• Only Object and External metrics are supported
• Only Deployment is supported as HPA target
• Metric selector MatchExpressions is not supported
• HPA's stabilizationWindow is not being respected, scaling from/to 0 will be done asap

Architecture overview

sequenceDiagram
loop every minute
    HpaInformer->>+autoscaling.k8s.io: subscribe
    autoscaling.k8s.io-->>-HpaInformer: Updates
    par for every hpa
        HpaInformer->>Goroutines: new
        loop every minute
            par for every metric
                Goroutines->>+*.metrics.k8s.io: Give me actual state
                *.metrics.k8s.io-->>-Goroutines: result
            end   
            alt All metrics are 0 AND Replicas > 0
                Goroutines->>+autoscaling.k8s.io: scale to 0
                autoscaling.k8s.io-->>-Goroutines: ok
            else Any metric is not 0 AND Replicas == 0
                Goroutines->>+autoscaling.k8s.io: scale to 1
                autoscaling.k8s.io-->>-Goroutines: ok
            end
            autoscaling.k8s.io-->>HpaInformer: Hpa exists no more
            HpaInformer->>Goroutines: die
        end
    end
end

Loading

Available metrics

scale_to_zero_errors
scale_to_zero_events
scale_to_zero_panics

Howto test locally

First you will need a testing cluster

kind create cluster --config=./demo/kind-config.yaml

You have to install prometheus and prometheus-adapter

helm upgrade -i prometheus prometheus-community/kube-prometheus-stack
helm upgrade -i prom-adapter prometheus-community/prometheus-adapter -f ./demo/prom-adapter-values.yaml

Now install demo services from the ./demo folder

kubectl apply -f ./demo/app.yml -n default
kubectl apply -f ./demo/metric-generator.yml -n default

to have multiple deployments with multiple random metrics bouncing between 0 and 1

Now you can run service locally to scale services in a real kube cluster: go run ./cmd --kube-config "<path to your kube config>" --hpa-selector "scaleToZero.spscommerce.com/watch=true"