cheapsk8s

Cheap Kuberernetes compute

Overview

Cheapsk8s hunts through cloud providers to find the cheapest compute for your workload.

Find compute based on resource requests

$ cks find --cpu 2 --mem 50kb --storage 100gi

select option:

  aws ec2 spot instance -- single node cluster (us-west-2) -- $0.10/hr
> gce preemptible vm -- single node cluster (us-central1) -- $0.12/hr
  azure spot instance -- single node cluster (west-us) -- $0.09/hr

You can also find based on the set requests and limits in your k8s deployment configurations.

$ cks find -f my-deployment.yaml

or

$ cks find -f my-folder/

Find compute based on runtime

Run a job that tests execution time vs cost on a variety of instance types

apiVersion: cheapsk8s.io/v1
kind: Experiment
metadata:
  name: my-experiment1
spec:
  providers:
    aws:
      regions: [us-west-2, us-west-1]
      instances: [m4-large, t2-small]
    gce:
      zones: [us-central1]
      intances: [e2, c2]
  job:
    template:
      spec:
        containers:
        - name: pi
          image: pytorch
          command: ["python", "classifier_test.py"]
        restartPolicy: Never
    backoffLimit: 4

Create the experiment

$ cks experiment -f my-experiment1.yaml

running job in 6 different environments...

Retrieve the outcomes

$ cks experiment status my-experiment1 -o yaml

status:
  provider:
    aws:
      "us-west-2/m4-large":
        cost: $2.40
        time: 400s
      "us-west-1/t2-small":
        cost: $1.40
        time: 700s
    gce:
      "us-central1/ec2":
        cost: $2.39
        time: 400s

Manage compute

List current clusters

$ cks compute view
NAME      PROVIDER     COST
aws-1     aws          $0.10/hr
gce-5     gce          $0.12/hr

Switch kubeconfigs

$ cks switch

select compute:

> aws-1
  gce-5