Page Comparison

...

When you schedule a pod to run on the Flightdeck platform, the autoscaler will find an existing node with enough remaining CPU and memory to run your pod’s containers. If you don’t have any remaining nodes with enough space, an additional EC2 instance will be provisioned and join the cluster. If the autoscaler detects unused space in your cluster, it will shut down EC2 instances which aren’t necessary.

...

...

Unlike some managed platforms, containers are not constrained to a predetermined size. When you deploy your application, you decide how much RAM and CPU to allocate to each container. This is done using resource requests in your pod manifests. Here is an example container which requests 512MiB of RAM and expects to use half a CPU core:

spec:
  containers:
  - name: main
    image: example:latest
    resources:
      requests:
        memory: 512Mi
        cpu: "500m"

We strongly recommend adding resource requests to all your containers. This will tell the cluster how much space it should expect to allocate to run your container.

Vertical Autoscaling

Many applications experience fluctuation in memory or CPU requirements as the application scales. Rather than manually tracking the resources your containers require and updating your resource requests as the application grows, you can tell the cluster to automatically adjust the requirements using the vertical pod autoscaler, which is deployed as part of the Flightdeck platform. Here is an example which will automatically adjust the size of the pods for a deployment, but will never scale it beyond 2 CPU cores or 4GiB of memory:

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: example
spec:
  targetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: example
  updatePolicy:
    updateMode: Auto
  resourcePolicy:
    containerPolicies:
    - containerName: "main"
      maxAllowed:
        cpu: 2000m
        memory: 4Gi