Scheduled Scaler

In order to use the ScheduledScaler you will need to install the CRD and deploy the Scaling Controller into your Kubernetes cluster.

Requirements

Kubernetes Version: 1.7+
Kubernetes Cluster Settings:
- "Legacy authorization": "Enabled"

Tested Environments

Google Kubernetes Engine
- Kubernetes Version: 1.9.3-gke.0, 1.7.15
- Docker Version: 1.12.5
- Golang Version: 1.9.4

Getting Started

Clone this repo

mkdir -p $GOPATH/src/k8s.restdev.com && \
git clone https://github.com/k8s-restdev/scheduled-scaler.git $GOPATH/src/k8s.restdev.com/operators && \
cd $GOPATH/src/k8s.restdev.com/operators

Install using Helm Chart

helm install scheduled-scaler artifacts/kubes/scaling/chart

Note: This uses the image stored at https://hub.docker.com/r/k8srestdev/scaling by default.

See chart README for detailed configuration options

Installation without Helm (and compiling binary yourself):

Install the CRD

kubectl create -f ./artifacts/kubes/scaling/crd.yml

Once you have the repo installed on your local dev you can test, build, push and deploy using make

Note: If you are just looking for a prebuilt image you can find the latest build here. Just add that image tag to the deployment yml in the artificats dir and apply to your kube-system namespace to get up and running without doing a fresh build :D

Using Make

The Makefile provides the following steps:

test - Run go unit tests
build - Build the go bin file and docker image locally
push - Push the built docker image to gcr (or another repository of your choice)
deploy - Deploy the updated image to your Kubernetes cluster

Each of these steps can be run in a single pass or can be used individually.

Examples

Do all the things (kubectl)

# This example will test, build, push and deploy using kubectl's currently configured cluster
make OPERATOR=scaling PROJECT_ID=my_project_id

Do all the things (kubernodes)

# This example will test, build, push and deploy using kubernodes
make OPERATOR=scaling PROJECT_ID=my_project_id DEPLOYBIN=kn KN_PROJECT_ID=my_kubernodes_project_id

Note: You only need to add KN_PROJECT_ID if it differs from PROJECT_ID

Just build the image

make build OPERATOR=scaling PROJECT_ID=my_project_id

Just push any image

make push IMAGE=myrepo/myimage:mytag

Just deploy any image (kubectl)

make deploy OPERATOR=scaling IMAGE=myrepo/myimage:mytag

Just deploy any image (kubernodes)

make deploy OPERATOR=scaling IMAGE=myrepo/myimage:mytag DEPLOYBIN=kn KN_PROJECT_ID=my_kubernodes_project_id

Now that you have all the resources required in your cluster you can begin creating ScheduledScalers.

Scheduled Scaler Spec

Note: This controller uses the following Cron Expression Format

HPA

apiVersion: "scaling.k8s.restdev.com/v1alpha1"
kind: ScheduledScaler
metadata:
  name: my-scheduled-scaler-1
spec:
  timeZone: America/Los_Angeles
  target:
    kind: HorizontalPodAutoscaler
    name: my-hpa
    apiVersion: autoscaling/v1
  steps:
  #run at 5:30am PST
  - runat: '0 30 5 * * *'
    mode: range
    minReplicas: 1
    maxReplicas: 5

Instance Group

apiVersion: "scaling.k8s.restdev.com/v1alpha1"
kind: ScheduledScaler
metadata:
  name: my-scheduled-scaler-2
spec:
  timeZone: America/Los_Angeles
  target:
    kind: InstanceGroup
    name: my-instance-group-name
    apiVersion: compute/v1
  steps:
  #run at 5:30am PST
  - runat: '0 30 5 * * *'
    mode: fixed
    replicas: 3

As you'll see above, you can target either instance groups (if you are on GKE) or hpa, but all the other options are the same.

Options

Option	Description	Required
spec.timeZone	Timezone to run crons in	False
spec.target.kind	Type of target (InstanceGroup/HorizontalPodAutoscaler)	True
spec.target.name	Name of the target resource	True
spec.target.apiVersion	API Version of the target	True
spec.steps	List of steps	True
spec.steps[].runat	Cronjob time string (gocron)	True
spec.steps[].mode	Type of scaling to run (fixed/range)	True
spec.steps[].replicas	Defined if mode is 'fixed'	False
spec.steps[].minReplicas	Defined if mode is 'range'	False
spec.steps[].maxReplicas	Defined if mode is 'range'	False