Reduce replica size on low resources

I am using Google Kubernetes Engine and some of my deployments are more important than others (For example, the staging environment can run on less replicas or even stop if needed).

I want to dynamically change the amount of replicas of a deployment depending on the available resources, especially memory.

Ideally, I would set a default number of replicas which is used normally and if the cluster is running low on resources, it should reduce the number of replicas of that deployment.

This should happen for some deployments but not all of them.

apiVersion: apps/v1
kind: Deployment
metadata:
  labels:
    app: my-deployment
  name: my-deployment
  namespace: my-namespace
spec:
  replicas: 3 # This should be lower if the cluster is running on low resources (memory or CPU)
  selector:
    matchLabels:
      app: my-deployment
  template:
    metadata:
      creationTimestamp: null
      labels:
        app: my-deployment
    spec:
      containers:
      - image: my/image:version
        name: my-deployment

It should even be possible to reduce the number of replicas down to 0 for some deployments but not for others.

Note that my deployments are distributed across multiple namespaces (if that matters).

> I want to dynamically change the amount of replicas of a deployment
> depending on the available resources, especially memory.

You can use the Kubernetes HPA (Horizonatal POD autoscaling), which which dynamically changes the number of replicas based on the CPU/Memory utilization.

While you can also set the default number of replicas for your deployment for deployment you have 1 scaling to 3 for staging you have a minimum 3 running scaling up to 5 and for Prod minimum running you 4 and scaling to 10 etc.

Read more about the HPA : https://kubernetes.io/docs/tasks/run-application/horizontal-pod-autoscale/

Example

kubectl autoscale deployment <Deployment-name> --cpu-percent=50 --min=1 --max=10

Example ref : https://kubernetes.io/docs/tasks/run-application/horizontal-pod-autoscale-walkthrough/

If there is less traffic auto-scale down to zero is also possible and based on CPU usage it can scale up pods auto without user input.

If you not looking for scaling or just for templating you are facing issues ignore above answer and you can use the scaffold or helm to mange the YAML templating dynamically.

Deployment.yaml

replicas: {{ .Values.replicas}}

values-dev.yaml

replicas=3

values-staging.yaml

replicas=5

based on the environment you can pass the values-*.yaml replica to helm and it will create the template for you.