Understanding how to add custom metrics autoscalers

Add a custom metrics autoscaler to a workload
Add a custom metrics autoscaler to a job
Additional resources

To add a custom metrics autoscaler, create a ScaledObject custom resource for a deployment, stateful set, or custom resource. Create a ScaledJob custom resource for a job.

You can create only one scaled object for each workload that you want to scale. Also, you cannot use a scaled object and the horizontal pod autoscaler (HPA) on the same workload.

Add a custom metrics autoscaler to a workload

You can create a custom metrics autoscaler for a workload that is created by a Deployment, StatefulSet, or custom resource object.

Prerequisites

The Custom Metrics Autoscaler Operator must be installed.

If you use a custom metrics autoscaler for scaling based on CPU or memory:

Your cluster administrator must have properly configured cluster metrics. You can use the oc describe PodMetrics <pod-name> command to determine if metrics are configured. If metrics are configured, the output appears similar to the following, with CPU and Memory displayed under Usage.

$ oc describe PodMetrics openshift-kube-scheduler-ip-10-0-135-131.ec2.internal

Example output

Name:         openshift-kube-scheduler-ip-10-0-135-131.ec2.internal
Namespace:    openshift-kube-scheduler
Labels:       <none>
Annotations:  <none>
API Version:  metrics.k8s.io/v1beta1
Containers:
  Name:  wait-for-host-port
  Usage:
    Memory:  0
  Name:      scheduler
  Usage:
    Cpu:     8m
    Memory:  45440Ki
Kind:        PodMetrics
Metadata:
  Creation Timestamp:  2019-05-23T18:47:56Z
  Self Link:           /apis/metrics.k8s.io/v1beta1/namespaces/openshift-kube-scheduler/pods/openshift-kube-scheduler-ip-10-0-135-131.ec2.internal
Timestamp:             2019-05-23T18:47:56Z
Window:                1m0s
Events:                <none>

The pods associated with the object you want to scale must include specified memory and CPU limits. For example:

Example pod spec

apiVersion: v1
kind: Pod
# ...
spec:
  containers:
  - name: app
    image: images.my-company.example/app:v4
    resources:
      limits:
        memory: "128Mi"
        cpu: "500m"
# ...

Procedure

Create a YAML file similar to the following. Only the name <2>, object name <4>, and object kind <5> are required:
Example scaled object
```
apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  annotations:
    autoscaling.keda.sh/paused-replicas: "0"
  name: scaledobject
  namespace: my-namespace
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    name: example-deployment
    kind: Deployment
    envSourceContainerName: .spec.template.spec.containers[0]
  cooldownPeriod:  200
  maxReplicaCount: 100
  minReplicaCount: 0
  metricsServer:
    auditConfig:
      logFormat: "json"
      logOutputVolumeClaim: "persistentVolumeClaimName"
      policy:
        rules:
        - level: Metadata
        omitStages: "RequestReceived"
        omitManagedFields: false
      lifetime:
        maxAge: "2"
        maxBackup: "1"
        maxSize: "50"
  fallback:
    failureThreshold: 3
    replicas: 6
    behavior: static
  pollingInterval: 30
  advanced:
    restoreToOriginalReplicaCount: false
    horizontalPodAutoscalerConfig:
      name: keda-hpa-scale-down
      behavior:
        scaleDown:
          stabilizationWindowSeconds: 300
          policies:
          - type: Percent
            value: 100
            periodSeconds: 15
  triggers:
  - type: prometheus
    metadata:
      serverAddress: https://thanos-querier.openshift-monitoring.svc.cluster.local:9092
      namespace: kedatest
      metricName: http_requests_total
      threshold: '5'
      query: sum(rate(http_requests_total{job="test-app"}[1m]))
      authModes: basic
    authenticationRef:
      name: prom-triggerauthentication
      kind: TriggerAuthentication
```
where:
metadata.annotations.autoscaling.keda.sh/paused-replicas

Specifies that the Custom Metrics Autoscaler Operator is to scale the replicas to the specified value and stop autoscaling, as described in the "Pausing the custom metrics autoscaler for a workload" section. This field is optional.

metadata.name

Specifies a name for this custom metrics autoscaler.

spec.scaleTargetRef.apiVersion

Specifies the API version of the target resource. The default is apps/v1. This field is optional.

spec.scaleTargetRef.name

Specifies the name of the object that you want to scale.

spec.scaleTargetRef.kind

Specifies the kind as Deployment, StatefulSet or CustomResource.

spec.scaleTargetRef.envSourceContainerName

Specifies the name of the container in the target resource, from which the custom metrics autoscaler gets environment variables holding secrets and so forth. The default is .spec.template.spec.containers[0]. This field is optional.

spec.cooldownPeriod

Specifies the period in seconds to wait after the last trigger is reported before scaling the deployment back to 0 if the minReplicaCount is set to 0. The default is 300. This field is optional.

spec.maxReplicaCount

Specifies the maximum number of replicas when scaling up. The default is 100. This field is optional.

spec.minReplicaCount

Specifies the minimum number of replicas when scaling down. This field is optional.

spec.metricsServer

Specifies the parameters for audit logs. as described in the "Configuring audit logging" section. This field is optional.

spec.fallback

Specifies the number of replicas to fall back to if a scaler fails to get metrics from the source for the number of times defined by the failureThreshold parameter. For more information on fallback behavior, see the KEDA documentation. This field is optional.

spec.fallback.behavior
Specifies the replica count to be used if a fallback occurs. Enter one of the following options or omit the parameter. This field is optional.
Enter static to use the number of replicas specified by the fallback.replicas parameter. This is the default.

Enter currentReplicas to maintain the current number of replicas.

Enter currentReplicasIfHigher to maintain the current number of replicas, if that number is higher than the fallback.replicas parameter. If the current number of replicas is lower than the fallback.replicas parameter, use the fallback.replicas value.

Enter currentReplicasIfLower to maintain the current number of replicas, if that number is lower than the fallback.replicas parameter. If the current number of replicas is higher than the fallback.replicas parameter, use the fallback.replicas value.
spec.pollingInterval

Specifies the interval in seconds to check each trigger on. The default is 30. This field is optional.

spec.advanced.restoreToOriginalReplicaCount

Specifies whether to scale back the target resource to the original replica count after the scaled object is deleted. The default is false, which keeps the replica count as it is when the scaled object is deleted. This field is optional.

spec.advanced.horizontalPodAutoscalerConfig.name

Specifies a name for the horizontal pod autoscaler. The default is keda-hpa-{scaled-object-name}. This field is optional.

spec.advanced.horizontalPodAutoscalerConfig.behavior

Specifies a scaling policy to use to control the rate to scale pods up or down, as described in the "Scaling policies" section. This field is optional.

spec.triggers[].type

Specifies the trigger to use as the basis for scaling, as described in the "Understanding the custom metrics autoscaler triggers" section. This example uses OKD monitoring.

spec.triggers[].authenticationRef
Specifies a trigger authentication or a cluster trigger authentication. For more information, see "Understanding the custom metrics autoscaler trigger authentication". This field is optional.
Enter TriggerAuthentication to use a trigger authentication. This is the default.

Enter ClusterTriggerAuthentication to use a cluster trigger authentication.
Create the custom metrics autoscaler by running the following command:
```
$ oc create -f <filename>.yaml
```

Verification

View the command output to verify that the custom metrics autoscaler was created:
```
$ oc get scaledobject <scaled_object_name>
```
Example output
```
NAME            SCALETARGETKIND      SCALETARGETNAME        MIN   MAX   TRIGGERS     AUTHENTICATION               READY   ACTIVE   FALLBACK   AGE
scaledobject    apps/v1.Deployment   example-deployment     0     50    prometheus   prom-triggerauthentication   True    True     True       17s
```
Note the following fields in the output:
- TRIGGERS: Indicates the trigger, or scaler, that is being used.
- AUTHENTICATION: Indicates the name of any trigger authentication being used.
- READY: Indicates whether the scaled object is ready to start scaling:
  
  If True, the scaled object is ready.
  
  If False, the scaled object is not ready because of a problem in one or more of the objects you created.
- ACTIVE: Indicates whether scaling is taking place:
  
  If True, scaling is taking place.
  
  If False, scaling is not taking place because there are no metrics or there is a problem in one or more of the objects you created.
- FALLBACK: Indicates whether the custom metrics autoscaler is able to get metrics from the source
  
  If False, the custom metrics autoscaler is getting metrics.
  
  If True, the custom metrics autoscaler is getting metrics because there are no metrics or there is a problem in one or more of the objects you created.

Add a custom metrics autoscaler to a job

You can create a custom metrics autoscaler for any Job object.

Scaling by using a scaled job is a Technology Preview feature only. Technology Preview features are not supported with Red Hat production service level agreements (SLAs) and might not be functionally complete. Red Hat does not recommend using them in production. These features provide early access to upcoming product features, enabling customers to test functionality and provide feedback during the development process.

For more information about the support scope of Red Hat Technology Preview features, see Technology Preview Features Support Scope.

Prerequisites

The Custom Metrics Autoscaler Operator must be installed.

Procedure

Create a YAML file similar to the following:
```
kind: ScaledJob
apiVersion: keda.sh/v1alpha1
metadata:
  name: scaledjob
  namespace: my-namespace
spec:
  failedJobsHistoryLimit: 5
  jobTargetRef:
    activeDeadlineSeconds: 600
    backoffLimit: 6
    parallelism: 1
    completions: 1
    template:
      metadata:
        name: pi
      spec:
        containers:
        - name: pi
          image: perl
          command: ["perl",  "-Mbignum=bpi", "-wle", "print bpi(2000)"]
  maxReplicaCount: 100
  pollingInterval: 30
  successfulJobsHistoryLimit: 5
  failedJobsHistoryLimit: 5
  envSourceContainerName:
  rolloutStrategy: gradual
  scalingStrategy:
    strategy: "custom"
    customScalingQueueLengthDeduction: 1
    customScalingRunningJobPercentage: "0.5"
    pendingPodConditions:
      - "Ready"
      - "PodScheduled"
      - "AnyOtherCustomPodCondition"
    multipleScalersCalculation : "max"
  triggers:
  - type: prometheus
    metadata:
      serverAddress: https://thanos-querier.openshift-monitoring.svc.cluster.local:9092
      namespace: kedatest
      metricName: http_requests_total
      threshold: '5'
      query: sum(rate(http_requests_total{job="test-app"}[1m]))
      authModes: "bearer"
    authenticationRef:
      name: prom-cluster-triggerauthentication
```
where:
spec.jobTargetRef.activeDeadlineSeconds

Specifies the maximum duration the job can run.

spec.jobTargetRef.backoffLimit

Specifies the number of retries for a job. The default is 6.

spec.jobTargetRef.parallelism
Specifies how many pod replicas a job should run in parallel; defaults to 1. This field is optional.
For non-parallel jobs, leave unset. When unset, the default is 1.
spec.jobTargetRef.completions
Specifies how many successful pod completions are needed to mark a job completed. This field is optional.
For non-parallel jobs, leave unset. When unset, the default is 1.

For parallel jobs with a fixed completion count, specify the number of completions.

For parallel jobs with a work queue, leave unset. When unset the default is the value of the parallelism parameter.
spec.jobTargetRef.template

Specifies the template for the pod the controller creates.

spec.maxReplicaCount

Specifies the maximum number of replicas when scaling up. The default is 100. This field is optional.

spec.pollingInterval

Specifies the interval in seconds to check each trigger on. The default is 30. This field is optional.

spec.successfulJobsHistoryLimit

Specifies the number of successful finished jobs should be kept. The default is 100. This field is optional.

spec.failedJobsHistoryLimit

Specifies how many failed jobs should be kept. The default is 100. This field is optional.

spec.envSourceContainerName

Specifies the name of the container in the target resource, from which the custom autoscaler gets environment variables holding secrets and so forth. The default is .spec.template.spec.containers[0]. This field is optional.

spec.rolloutStrategy
Specifies whether existing jobs are terminated whenever a scaled job is being updated. This field is optional.
default: The autoscaler terminates an existing job if its associated scaled job is updated. The autoscaler recreates the job with the latest specs.

gradual: The autoscaler does not terminate an existing job if its associated scaled job is updated. The autoscaler creates new jobs with the latest specs.
spec.scalingStrategy

Specifies a scaling strategy: default, custom, or accurate. The default is default. This field is optional.

spec.triggers[].type

Specifies the trigger to use as the basis for scaling. For more information, see "Understanding custom metrics autoscaler triggers".

spec.triggers[].authenticationRef
Specifies a trigger authentication or a cluster trigger authentication. For more information, see "Understanding custom metrics autoscaler trigger authentications". This field is optional.
Enter TriggerAuthentication to use a trigger authentication. This is the default.

Enter ClusterTriggerAuthentication to use a cluster trigger authentication.
Create the custom metrics autoscaler by running the following command:
```
$ oc create -f <filename>.yaml
```

Verification

View the command output to verify that the custom metrics autoscaler was created:
```
$ oc get scaledjob <scaled_job_name>
```
Example output
```
NAME        MAX   TRIGGERS     AUTHENTICATION              READY   ACTIVE    AGE
scaledjob   100   prometheus   prom-triggerauthentication  True    True      8s
```
Note the following fields in the output:
- TRIGGERS: Indicates the trigger, or scaler, that is being used.
- AUTHENTICATION: Indicates the name of any trigger authentication being used.
- READY: Indicates whether the scaled object is ready to start scaling:
  
  If True, the scaled object is ready.
  
  If False, the scaled object is not ready because of a problem in one or more of the objects you created.
- ACTIVE: Indicates whether scaling is taking place:
  
  If True, scaling is taking place.
  
  If False, scaling is not taking place because there are no metrics or there is a problem in one or more of the objects you created.

Add a custom metrics autoscaler to a workload

Add a custom metrics autoscaler to a job

Additional resources