Enabling user-defined custom metrics for Horizontal Pod autoscaling

This topic describes how to configure user-defined metrics for Horizontal Pod autoscaling (HPA) in Google Distributed Cloud.

Enabling Logging and Monitoring for user applications

The configuration for Logging and Monitoring is held in a Stackdriver object named stackdriver.

  1. Open the stackdriver object for editing:

    kubectl --kubeconfig=USER_CLUSTER_KUBECONFIG --namespace kube-system edit stackdriver stackdriver
    

    Replace USER_CLUSTER_KUBECONFIG with the path of your user cluster kubeconfig file.

  2. Under spec, set both enableStackdriverForApplications and enableCustomMetricsAdapter to true:

    apiVersion: addons.gke.io/v1alpha1
    kind: Stackdriver
    metadata:
    name: stackdriver
    namespace: kube-system
    spec:
    projectID: project-id
    clusterName: cluster-name
    clusterLocation: cluster-location
    proxyConfigSecretName: secret-name
    enableStackdriverForApplications: true
    enableCustomMetricsAdapter: true
    scalableMonitoring: true
    enableVPC: stackdriver-enable-VPC
    optimizedMetrics: true
    
  3. Save and close the edited file.

Once these steps are done, all the user application logs are sent to Cloud Logging.

The next step is to annotate the user application for metrics collection.

Annotate a user application for metrics collection

To annotate a user application to be scraped and the logs sent to Cloud Monitoring, you must add corresponding annotations to the metadata for the service, Pod, and endpoints.

  metadata:
    name: "example-monitoring"
    namespace: "default"
    annotations:
      prometheus.io/scrape: "true"
      prometheus.io/path: "" - Overriding metrics path (default "/metrics")
  

Deploy an example user application

In this section, you deploy a sample application with both logs and prometheus-compatible metrics.

  1. Save the following Service and Deployment manifests to a file named my-app.yaml. Notice that the Service has the annotation prometheus.io/scrape: "true":
  kind: Service
  apiVersion: v1
  metadata:
    name: "example-monitoring"
    namespace: "default"
    annotations:
      prometheus.io/scrape: "true"
  spec:
    selector:
      app: "example-monitoring"
    ports:
      - name: http
        port: 9090
  ---
  apiVersion: apps/v1
  kind: Deployment
  metadata:
    name: "example-monitoring"
    namespace: "default"
    labels:
      app: "example-monitoring"
  spec:
    replicas: 1
    selector: 
      matchLabels:
        app: "example-monitoring"
    template: 
      metadata: 
        labels:
          app: "example-monitoring"
      spec:
        containers:
        - image: gcr.io/google-samples/prometheus-example-exporter:latest
          name: prometheus-example-exporter
          imagePullPolicy: Always
          command:
          - /bin/sh
          - -c
          - ./prometheus-example-exporter --metric-name=example_monitoring_up --metric-value=1 --port=9090
          resources:
            requests:
              cpu: 100m
  1. Create the Deployment and the Service:

    kubectl --kubeconfig USER_CLUSTER_KUBECONFIG apply -f my-app.yaml
    

Use the custom metrics in HPA

Deploy the HPA object to use the metric exposed in the previous step. See Autoscaling on multiple metrics and custom metrics for more advanced information about different type of custom metrics.

apiVersion: autoscaling/v2beta2
kind: HorizontalPodAutoscaler
metadata:
  name: example-monitoring-hpa
  namespace: default
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: example-monitoring
  minReplicas: 1
  maxReplicas: 5
  metrics:
  - type: Pods
    pods:
      metric:
        name: example_monitoring_up
      target:
        type: AverageValue
        averageValue: 20

The Pods type metric has a default metric selector for the labels of the target Pods, which is how kube-controller-maneger works. In this example, you cna query the example_monitoring_up metric with a selector of {matchLabels: {app: example-monitoring}} as they are available in the target Pods. Any other selector specified is added to the list. To avoid the default selector, you canremove any labels on the target Pod or use the Object type metric.

Check that the user-defined application metrics are used by HPA

Check that the user defined application metrics are used by HPA:

kubectl --kubeconfig=USER_CLUSTER_KUBECONFIG describe hpa example-monitoring-hpa

The output will look like this:

Name:                             example-monitoring-hpa
Namespace:                        default
Labels:                           
Annotations:                      CreationTimestamp:  Mon, 19 Jul 2021 16:00:40 -0800
Reference:                        Deployment/example-monitoring
Metrics:                          ( current / target )
  "example_monitoring_up" on pods:  1 / 20
Min replicas:                     1
Max replicas:                     5
Deployment pods:                  1 current / 1 desired
Conditions:
  Type            Status  Reason              Message


AbleToScale True ReadyForNewScale recommended size matches current size ScalingActive True ValidMetricFound the HPA was able to successfully calculate a replica count from pods metric example_monitoring_up ScalingLimited False DesiredWithinRange the desired count is within the acceptable range

Costs

Using custom metrics for HPA does not incur any additional charges. Users are charged only for application metrics and logs. See Google Cloud's operations suite pricing for details. The Pod for enabling custom metrics consumes an extra 15m CPU and 20MB memory.