Enabling user-defined custom metrics for Horizontal Pod autoscaling

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

This page is for Admins, Architects, and Operators who optimize systems architecture and resources to ensure the lowest total cost of ownership for their company or business unit, and plan capacity and infrastructure needs. To learn more about common roles and example tasks that we reference in Google Cloud content, see Common GKE Enterprise user roles and tasks.

Deploy Prometheus and Metrics Adapter

In this section you deploy Prometheus to scrape user-defined metrics and the prometheus-adapter to fulfil the Kubernetes Custom Metrics API with Prometheus as its backend.

Save the following manifests to a file named custom-metrics-adapter.yaml.

Manifest file content for Prometheus and Metrics Adapter

# Copyright 2018 Google Inc
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#      http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

apiVersion: v1
kind: ServiceAccount
metadata:
  name: stackdriver-prometheus
  namespace: kube-system
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: stackdriver-prometheus
  namespace: kube-system
rules:
- apiGroups:
  - ""
  resources:
  - nodes
  - services
  - endpoints
  - pods
  verbs:
  - get
  - list
  - watch
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: stackdriver-prometheus
  namespace: kube-system
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
  name: stackdriver-prometheus
subjects:
- kind: ServiceAccount
  name: stackdriver-prometheus
  namespace: kube-system
---
apiVersion: v1
kind: Service
metadata:
  name: stackdriver-prometheus-app
  namespace: kube-system
  labels:
    app: stackdriver-prometheus-app
spec:
  clusterIP: "None"
  ports:
    - name: http
      port: 9090
      protocol: TCP
      targetPort: 9090
  sessionAffinity: ClientIP
  selector:
    app: stackdriver-prometheus-app
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: stackdriver-prometheus-app
  namespace: kube-system
  labels:
    app: stackdriver-prometheus-app
spec:
  replicas: 1
  selector:
    matchLabels:
      app: stackdriver-prometheus-app
  template:
    metadata:
      labels:
        app: stackdriver-prometheus-app
    spec:
      serviceAccount: stackdriver-prometheus
      containers:
      - name: prometheus-server
        image: prom/prometheus:v2.45.0
        args:
        - "--config.file=/etc/prometheus/config/prometheus.yaml"
        - "--storage.tsdb.path=/data"
        - "--storage.tsdb.retention.time=2h"
        ports:
        - name: prometheus
          containerPort: 9090
        readinessProbe:
          httpGet:
            path: /-/ready
            port: 9090
          periodSeconds: 5
          timeoutSeconds: 3
          # Allow up to 10m on startup for data recovery
          failureThreshold: 120
        livenessProbe:
          httpGet:
            path: /-/healthy
            port: 9090
          periodSeconds: 5
          timeoutSeconds: 3
          failureThreshold: 6
        resources:
          requests:
            cpu: 250m
            memory: 500Mi
        volumeMounts:
        - name: config-volume
          mountPath: /etc/prometheus/config
        - name: stackdriver-prometheus-app-data
          mountPath: /data
      volumes:
      - name: config-volume
        configMap:
          name: stackdriver-prometheus-app
      - name: stackdriver-prometheus-app-data
        emptyDir: {}
      terminationGracePeriodSeconds: 300
      nodeSelector:
        kubernetes.io/os: linux
---
apiVersion: v1
data:
  prometheus.yaml: |
    global:
      scrape_interval: 1m
    rule_files:
    - /etc/config/rules.yaml
    - /etc/config/alerts.yaml
    scrape_configs:
    - job_name: prometheus-io-endpoints
      kubernetes_sd_configs:
      - role: endpoints
      relabel_configs:
      - action: keep
        regex: true
        source_labels:
        - __meta_kubernetes_service_annotation_prometheus_io_scrape
      - action: replace
        regex: (.+)
        source_labels:
        - __meta_kubernetes_service_annotation_prometheus_io_path
        target_label: __metrics_path__
      - action: replace
        regex: (https?)
        source_labels:
        - __meta_kubernetes_service_annotation_prometheus_io_scheme
        target_label: __scheme__
      - action: replace
        regex: ([^:]+)(?::\d+)?;(\d+)
        replacement: $1:$2
        source_labels:
        - __address__
        - __meta_kubernetes_service_annotation_prometheus_io_port
        target_label: __address__
      - action: replace
        source_labels:
        - __meta_kubernetes_namespace
        target_label: namespace
      - action: replace
        source_labels:
        - __meta_kubernetes_pod_name
        target_label: pod
      - action: keep
        regex: (.+)
        source_labels:
        - __meta_kubernetes_endpoint_port_name
    - job_name: prometheus-io-services
      kubernetes_sd_configs:
      - role: service
      metrics_path: /probe
      params:
        module:
        - http_2xx
      relabel_configs:
      - action: replace
        source_labels:
        - __address__
        target_label: __param_target
      - action: replace
        replacement: blackbox
        target_label: __address__
      - action: keep
        regex: true
        source_labels:
        - __meta_kubernetes_service_annotation_prometheus_io_probe
      - action: replace
        source_labels:
        - __meta_kubernetes_namespace
        target_label: namespace
      - action: replace
        source_labels:
        - __meta_kubernetes_pod_name
        target_label: pod
    - job_name: prometheus-io-pods
      kubernetes_sd_configs:
      - role: pod
      relabel_configs:
      - action: keep
        regex: true
        source_labels:
        - __meta_kubernetes_pod_annotation_prometheus_io_scrape
      - action: replace
        regex: (.+)
        source_labels:
        - __meta_kubernetes_pod_annotation_prometheus_io_path
        target_label: __metrics_path__
      - action: replace
        regex: ([^:]+)(?::\d+)?;(\d+)
        replacement: $1:$2
        source_labels:
        - __address__
        - __meta_kubernetes_pod_annotation_prometheus_io_port
        target_label: __address__
      - action: replace
        source_labels:
        - __meta_kubernetes_namespace
        target_label: namespace
      - action: replace
        source_labels:
        - __meta_kubernetes_pod_name
        target_label: pod
kind: ConfigMap
metadata:
  name: stackdriver-prometheus-app
  namespace: kube-system
---

# The main section of custom metrics adapter.
kind: ServiceAccount
apiVersion: v1
metadata:
  name: custom-metrics-apiserver
  namespace: kube-system
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: custom-metrics:system:auth-delegator
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
  name: system:auth-delegator
subjects:
- kind: ServiceAccount
  name: custom-metrics-apiserver
  namespace: kube-system
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: custom-metrics-server-resources
rules:
- apiGroups:
  - custom.metrics.k8s.io
  resources: ["*"]
  verbs: ["*"]
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: custom-metrics-resource-reader
rules:
- apiGroups:
  - ""
  resources:
  - nodes
  - namespaces
  - pods
  - services
  verbs:
  - get
  - watch
  - list
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: custom-metrics-resource-reader
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
  name: custom-metrics-resource-reader
subjects:
- kind: ServiceAccount
  name: custom-metrics-apiserver
  namespace: kube-system
---
apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
  name: custom-metrics-auth-reader
  namespace: kube-system
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: Role
  name: extension-apiserver-authentication-reader
subjects:
- kind: ServiceAccount
  name: custom-metrics-apiserver
  namespace: kube-system
---
apiVersion: v1
kind: ConfigMap
metadata:
  name: adapter-config
  namespace: kube-system
data:
  config.yaml: |
    rules:
    default: false
      # fliter all metrics
    - seriesQuery: '{pod=~".+"}'
      seriesFilters: []
      resources:
        # resource name is mapped as it is. ex. namespace -> namespace
        template: <<.Resource>>
      name:
        matches: ^(.*)$
        as: ""
      # Aggregate metric on resource level
      metricsQuery: sum(<<.Series>>{<<.LabelMatchers>>}) by (<<.GroupBy>>)
---
apiVersion: apps/v1
kind: Deployment
metadata:
  labels:
    app: custom-metrics-apiserver
  name: custom-metrics-apiserver
  namespace: kube-system
spec:
  replicas: 1
  selector:
    matchLabels:
      app: custom-metrics-apiserver
  template:
    metadata:
      labels:
        app: custom-metrics-apiserver
      name: custom-metrics-apiserver
    spec:
      serviceAccountName: custom-metrics-apiserver
      containers:
      - name: custom-metrics-apiserver
        resources:
          requests:
            cpu: 15m
            memory: 20Mi
          limits:
            cpu: 100m
            memory: 150Mi
        image: registry.k8s.io/prometheus-adapter/prometheus-adapter:v0.11.0
        args:
        - /adapter
        - --cert-dir=/var/run/serving-cert
        - --secure-port=6443
        - --prometheus-url=http://stackdriver-prometheus-app.kube-system.svc:9090/
        - --metrics-relist-interval=1m
        - --config=/etc/adapter/config.yaml
        ports:
        - containerPort: 6443
        volumeMounts:
        - name: serving-cert
          mountPath: /var/run/serving-cert
        - mountPath: /etc/adapter/
          name: config
          readOnly: true
      nodeSelector:
        kubernetes.io/os: linux
      volumes:
      - name: serving-cert
        emptyDir:
          medium: Memory
      - name: config
        configMap:
          name: adapter-config
---
apiVersion: v1
kind: Service
metadata:
  name: custom-metrics-apiserver
  namespace: kube-system
spec:
  ports:
  - port: 443
    targetPort: 6443
  selector:
    app: custom-metrics-apiserver
---
apiVersion: apiregistration.k8s.io/v1
kind: APIService
metadata:
  name: v1beta1.custom.metrics.k8s.io
spec:
  service:
    name: custom-metrics-apiserver
    namespace: kube-system
  group: custom.metrics.k8s.io
  version: v1beta1
  insecureSkipTLSVerify: true
  groupPriorityMinimum: 100
  versionPriority: 100
---
apiVersion: apiregistration.k8s.io/v1
kind: APIService
metadata:
  name: v1beta2.custom.metrics.k8s.io
spec:
  service:
    name: custom-metrics-apiserver
    namespace: kube-system
  group: custom.metrics.k8s.io
  version: v1beta2
  insecureSkipTLSVerify: true
  groupPriorityMinimum: 100
  versionPriority: 100
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: hpa-controller-custom-metrics
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
  name: custom-metrics-server-resources
subjects:
- kind: ServiceAccount
  name: horizontal-pod-autoscaler
  namespace: kube-system

Create the Deployment and the Service:

kubectl --kubeconfig USER_CLUSTER_KUBECONFIG apply -f custom-metrics-adapter.yaml

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-dummy-exporter:v0.2.0
            name: prometheus-example-exporter
            command:
            - ./prometheus-dummy-exporter
            args:
            - --metric-name=example_monitoring_up
            - --metric-value=1
            - --port=9090
            resources:
              requests:
                cpu: 100m
    
  2. 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/v2
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-manager 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:        autoscaling.alpha.kubernetes.io/conditions:
                        [{"type":"AbleToScale","status":"True","lastTransitionTime":"2023-08-23T22:07:24Z","reason":"ReadyForNewScale","message":"recommended size...
                      autoscaling.alpha.kubernetes.io/current-metrics: [{"type":"Pods","pods":{"metricName":"example_monitoring_up","currentAverageValue":"1"}}]
                      autoscaling.alpha.kubernetes.io/metrics: [{"type":"Pods","pods":{"metricName":"example_monitoring_up","targetAverageValue":"20"}}]
  CreationTimestamp:  Wed, 23 Aug 2023 22:07:09 +0000
  Reference:          Deployment/example-monitoring
  Min replicas:       1
  Max replicas:       5
  Deployment pods:    1 current / 1 desired
  

Costs

Using custom metrics for HPA does not incur any additional Cloud Monitoring charges. The Pods for enabling custom metrics consume additional CPU and memory based on the amount of metrics it scrapes.