Kubernetes Cluster Autoscaling

이 페이지는 아직 한국어로 제공되지 않습니다. 번역 작업 중입니다.
현재 번역 프로젝트에 대한 질문이나 피드백이 있으신 경우 언제든지 연락주시기 바랍니다.
Join the Preview!

Datadog Kubernetes Cluster Autoscaling is in Preview. Use this form to submit your request today.

Request Access
During the Preview period, Datadog recommends non-production use only of Datadog Kubernetes Autoscaling.

Requirements

  • Datadog Agent v7.75+

  • Datadog Operator v1.23.0+ or Datadog Helm chart v3.155.1+

  • Remote Configuration must be enabled both at the organization level and on the Agents in your target cluster. See Enabling Remote Configuration for setup instructions.

  • Datadog user permissions:

    • Org Management
    • API Keys Write
    • Autoscaling Manage

    By default, all Datadog admin users have Autoscaling Manage, and all Datadog users have read access to view recommendations and reporting.

    To grant Autoscaling Manage to an additional standard or custom role user, see Datadog Role Permissions for instructions.

Setup

  1. Activate Cluster Autoscaling.

      Set features.autoscaling.cluster.enabled to true in your DatadogAgent spec.

      spec:
  features:
    autoscaling:
      cluster:
        enabled: true

      Set datadog.autoscaling.cluster.enabled to true in your datadog-values.yaml file.

      datadog:
  autoscaling:
    cluster:
      enabled: true

      Then run helm upgrade.

    • If you do not have an existing installation of Karpenter, install the Datadog kubectl plugin. This plugin assists in installing or connecting to your existing Karpenter configuration.

      kubectl krew install datadog

    • Run the kubectl plugin and pass in AWS credentials for the AWS project that the cluster is running:

      AWS_PROFILE=<foo> AWS_REGION=<bar> kubectl datadog autoscaling cluster install

      This command creates:

      • Two CloudFormation stacks named dd-karpenter-$CLUSTER_NAME
      • A Helm release named karpenter in the Kubernetes namespace dd-karpenter
      • Karpenter’s EC2NodeClass and NodePool objects

      For more detailed command options, run kubectl datadog autoscaling cluster install --help.

    Autoscale a cluster

    In Datadog, navigate to Kubernetes Autoscaling. You can review all your clusters that are available for autoscaling on the Autoscaling Settings page. Toggle the Enabled column for your cluster of choice.

    The Autoscaling Settings page showing the Enable Clusters for Autoscaling section. Displays a filterable table of Kubernetes clusters with columns for cluster name, nodes, CPU capacity, memory capacity, CPU usage percentage with visual bars, memory usage percentage with visual bars, autoscaled workloads count, cluster scaling availability, workload scaling availability, and an enabled toggle switch. Toggle switches on the right show which clusters have autoscaling enabled.

    Use the Cluster Scaling page and select a cluster to review its recommendations. Identify a cluster you want to begin optimizing. During the Preview period, Datadog recommends selecting a non-production cluster.

    The Cluster Scaling detail page for cluster dev-eks-shopist-a-us-east-1. The Cost Summary section displays four metrics: cluster cost of $140.7k with negative $3.1k month-over-month change, idle cost of $93.66k with negative $2.29k change, CPU usage/request efficiency at 27.24%, and memory usage/request efficiency at 35.15%. The Scaling Events section shows two time-series graphs tracking CPU and memory metrics over the past month, with fluctuating usage patterns, alongside a timeline of recent scaling events including pod and DatadogPodAutoscaler activities. The Scaling Recommendations section displays a table of workloads with idle CPU and memory, showing 8.7% autoscaling enabled.

    Click Start Autoscaling to apply recommendations.