Scaling an Application

This page explains how to scale a deployed application in Google Kubernetes Engine.

Overview

When you deploy an application in GKE, you define how many replicas of the application you'd like to run. When you scale an application, you increase or decrease the number of replicas.

Each replica of your application represents a Kubernetes Pod that encapsulates your application's container(s).

Before you begin

To prepare for this task, perform the following steps:

• Ensure that you have enabled the Google Kubernetes Engine API.
Enable Google Kubernetes Engine API
• Ensure that you have installed the Cloud SDK.
• Set your default project ID:

  gcloud config set project [PROJECT_ID]

• If you are working with zonal clusters, set your default compute zone:

  gcloud config set compute/zone [COMPUTE_ZONE]

• If you are working with regional clusters, set your default compute region:

  gcloud config set compute/region [COMPUTE_REGION]

• Update gcloud to the latest version:

  gcloud components update

Note: You can override these default settings in gcloud commands using the --project, --zone, and --region operational flags.

Inspecting an application

Before scaling your application, you should inspect the application and ensure that it is healthy.

To see all applications deployed to your cluster, run kubectl get [CONTROLLER]. Substitute [CONTROLLER] for deployments, statefulsets, or another controller object type.

For example, if you run kubectl get deployments and you have created only one Deployment, the command's output should look similar to the following:

NAME                  DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE
my-app                1         1         1            1           10m

The output of this command is similar for all objects, but may appear slightly different. For Deployments, the output has six columns:

• NAME lists the names of the Deployments in the cluster.
• DESIRED displays the desired number of replicas, or the desired state, of the application, which you define when you create the Deployment.
• CURRENT displays how many replicas are currently running.
• UP-TO-DATE displays the number of replicas that have been updated to achieve the desired state.
• AVAILABLE displays how many replicas of the application are available to your users.
• AGE displays the amount of time that the application has been running in the cluster.

In this example, there is only one Deployment, my-app, which has only one replica because its desired state is one replica. You define the desired state at the time of creation, and you can change it at any time by scaling the application.

Inspecting StatefulSets

Before scaling a StatefulSet, you should inspect it by running kubectl describe statefulset my-app.

In the output of this command, check the Pods Status field. If the Failed value is greater than 0, scaling might fail.

If a StatefulSet appears to be unhealthy, run kubectl get pods to see which replicas are unhealthy. Then, run kubectl delete [POD], where [POD] is the name of the unhealthy Pod.

Attempting to scale a StatefulSet while it is unhealthy may cause it to become unavailable.

Scaling an application

The following sections describe each method you can use to scale an application. The kubectl scale method is the fastest way to scale. However, you may prefer another method in some situations, like when updating configuration files or when performing in-place modifications.

kubectl scale

kubectl scale lets your instantaneously change the number of replicas you want to run your application.

To use kubectl scale, you specify the new number of replicas by setting the --replicas flag. For example, to scale my-app to four replicas, run the following command, substituting [CONTROLLER] for deployment, statefulset, or another controller object type:

kubectl scale [CONTROLLER] my-app --replicas 4

If successful, this command's output should be similar to deployment "my-app" scaled.

Next, run kubectl get [CONTROLLER] my-app. The output should look similar to the following:

NAME                  DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE
my-app                4         4         4            4           15m

Note: It may take several minutes for scaling to complete.

kubectl apply

You can use kubectl apply to apply a new configuration file to an existing controller object. kubectl apply is useful for making multiple changes to a resource, and may be useful for users who prefer to manage their resources in configuration files.

To scale using kubectl apply, the configuration file you supply should include a new number of replicas in the replicas field of the object's specification.

The following is an updated version of the configuration file for the example my-app object. The example shows a Deployment, so if you use another type of controller, such as a StatefulSet, change the kind accordingly. This example works best on a cluster with at least three Nodes.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: my-app
spec:
  replicas: 3
  selector:
    matchLabels:
      app: app
  template:
    metadata:
      labels:
        app: app
    spec:
      containers:
      - name: my-container
        image: gcr.io/google-samples/hello-app:2.0

In this file, the value of the replicas field is 3. When this configuration file is applied, the object my-app scales to three replicas.

To apply an updated configuration file, run the following command:

kubectl apply -f config.yaml

Next, run kubectl get [CONTROLLER] my-app. The output should look similar to the following:

NAME                  DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE
my-app                3         3         3            3           15m

Console

To scale a workload in Google Cloud Platform Console, perform the following steps:

1. Visit the Google Kubernetes Engine Workloads menu in GCP Console.

   Visit the Workloads menu

2. Select the desired workload from the menu.

3. Click Actions, then Scale.
4. From the Replicas field, enter the desired number of replicas.
5. Click Scale.

Autoscaling Deployments

You can autoscale Deployments based on CPU utilization of Pods using kubectl autoscale or from the GKE Workloads menu in GCP Console.

kubectl autoscale

kubectl autoscale creates a HorizontalPodAutoscaler (or HPA) object that targets a specified resource (called the scale target) and scales it as needed. The HPA periodically adjusts the number of replicas of the scale target to match the average CPU utilization that you specify.

Note: Your object's Pod template can include a resources field that specifies a CPU utilization request for each Pod. To learn more about making resource requests for Pods and containers, refer to Managing Compute Resources for Containers

When you use kubectl autoscale, you specify a maximum and minimum number of replicas for your application, as well as a CPU utilization target. For example, to set the maximum number of replicas to six and the minimum to four, with a CPU utilization target of 50% utilization, run the following command:

kubectl autoscale deployment my-app --max 6 --min 4 --cpu-percent 50

In this command, the --max flag is required. The --cpu-percent flag is the target CPU utilization over all the Pods. This command does not immediately scale the Deployment to six replicas, unless there is already a systemic demand.

After running kubectl autoscale, the HorizontalPodAutoscaler object is created and targets the application. When there a change in load, the object increases or decreases the application's replicas.

Note: The maximum number of replicas is limited by the cluster's resources. If the cluster has a static number of available nodes, the cluster might run out of resources and prevent some replicated Pods from running. To learn about scaling your cluster, refer to Resizing a Cluster and Cluster Autoscaling.

To see a specific HorizontalPodAutoscaler object in your cluster, run:

kubectl get hpa [HPA_NAME]

To see the HorizontalPodAutoscaler configuration:

kubectl get hpa [HPA_NAME] -o yaml

The output of this command is similar to the following:

apiVersion: v1
items:
- apiVersion: autoscaling/v1
  kind: HorizontalPodAutoscaler
  metadata:
    creationTimestamp: ...
    name: [HPA_NAME]
    namespace: default
    resourceVersion: "664"
    selfLink: ...
    uid: ...
  spec:
    maxReplicas: 10
    minReplicas: 1
    scaleTargetRef:
      apiVersion: apps/v1
      kind: Deployment
      name: [HPA_NAME]
    targetCPUUtilizationPercentage: 50
  status:
    currentReplicas: 0
    desiredReplicas: 0
kind: List
metadata: {}
resourceVersion: ""
selfLink: ""

In this example output, the targetCPUUtilizationPercentage field holds the 50 percentage value passed in from the kubectl autoscale example.

To see a detailed description of a specific HorizontalPodAutoscaler object in the cluster:

kubectl describe hpa [HPA_NAME]

You can modify the HorizontalPodAutoscaler by applying a new configuration file with kubectl apply, using kubectl edit, or using kubectl patch.

To delete a HorizontalPodAutoscaler object:

kubectl delete hpa [HPA_NAME]

Console

To autoscale a Deployment, perform the following steps:

1. Visit the Google Kubernetes Engine Workloads menu in GCP Console.

   Visit the Workloads menu

2. Select the desired workload from the menu.

3. Click Actions, then Autoscale.
4. Fill the Maximum number of pods field with the desired maximum number of Pods.
5. Optionally, fill the Minimum number of pods and Target CPU utilization in percent fields with the desired values.
6. Click Autoscale.

Autoscaling with Custom Metrics

Beta

This is a Beta release of autoscaling Deployments with custom metrics. This feature is not covered by any SLA or deprecation policy and might be subject to backward-incompatible changes.

Starting with GKE version 1.9, you can scale your Deployments based on custom metrics exported from Stackdriver Monitoring.

To learn how to use custom metrics to autoscale deployments, refer to the Autoscaling Deployments with Custom Metrics tutorial.

Note: Stackdriver Monitoring is a Google Cloud Platform service separate from GKE. To use scaling based on custom metrics, you need to have a paid Stackdriver service account associated with your GCP Console project. For more information, refer to the Stackdriver Monitoring documentation.

What's next

• Learn about exposing your application externally.