Kubernetes distributions

Docs > Container Monitoring > Kubernetes > Kubernetes distributions

Overview

This section aims to document specifics and to provide good base configuration for all major Kubernetes distributions. These configurations can then be customized to add any Datadog feature.

AWS Elastic Kubernetes Service (EKS)
Azure Kubernetes Service (AKS)
Google Kubernetes Engine (GKE)
Red Hat OpenShift
Rancher
Oracle Container Engine for Kubernetes (OKE)
vSphere Tanzu Kubernetes Grid (TKG)

AWS Elastic Kubernetes Service (EKS)

No specific configuration is required.

If you are using AWS Bottlerocket OS on your nodes, add the following to enable container monitoring (containerd check):

In an EKS cluster, you can install the Operator using Helm or as an EKS add-on.

The configuration below is meant to work with either setup (Helm or EKS add-on) when the Agent is installed in the same namespace as the Datadog Operator.

kind: DatadogAgent
apiVersion: datadoghq.com/v2alpha1
metadata:
  name: datadog
spec:
  features:
    admissionController:
      enabled: false
    externalMetricsServer:
      enabled: false
      useDatadogMetrics: false
  global:
    credentials:
      apiKey: <DATADOG_API_KEY>
      appKey: <DATADOG_APP_KEY>
    criSocketPath: /run/dockershim.sock
  override:
    clusterAgent:
      image:
        name: gcr.io/datadoghq/cluster-agent:latest

Custom datadog-values.yaml:

datadog:
  apiKey: <DATADOG_API_KEY>
  appKey: <DATADOG_APP_KEY>
  criSocketPath: /run/dockershim.sock
  env:
    - name: DD_AUTOCONFIG_INCLUDE_FEATURES
      value: "containerd"

Azure Kubernetes Service (AKS)

AKS requires a specific configuration for the Kubelet integration due to how AKS has set up the SSL Certificates. Additionally, the optional Admission Controller feature requires a specific configuration to prevent an error when reconciling the webhook.

DatadogAgent Kubernetes Resource:

kind: DatadogAgent
apiVersion: datadoghq.com/v2alpha1
metadata:
  name: datadog
spec:
  features:
    admissionController:
      enabled: true
  global:
    credentials:
      apiKey: <DATADOG_API_KEY>
      appKey: <DATADOG_APP_KEY>
    kubelet:
      tlsVerify: false
  override:
    clusterAgent:
      containers:
        cluster-agent:
          env:
            - name: DD_ADMISSION_CONTROLLER_ADD_AKS_SELECTORS
              value: "true"

Custom datadog-values.yaml:

datadog:
  apiKey: <DATADOG_API_KEY>
  appKey: <DATADOG_APP_KEY>
  # Required as of Agent 7.35. See Kubelet Certificate note below.
  kubelet:
    tlsVerify: false

providers:
  aks:
    enabled: true

The providers.aks.enabled option sets the necessary environment variable DD_ADMISSION_CONTROLLER_ADD_AKS_SELECTORS="true" for you.

The kubelet.tlsVerify=false sets the environment variable DD_KUBELET_TLS_VERIFY=false for you to deactivate verification of the server certificate.

AKS Kubelet certificate

There is a known issue with the format of the AKS Kubelet certificate in older node image versions. As of Agent 7.35, it is required to use tlsVerify: false as the certificates did not contain a valid Subject Alternative Name (SAN).

If all the nodes within your AKS cluster are using a supported node image version, you can use Kubelet TLS Verification. Your version must be at or above the versions listed here for the 2022-10-30 release. You must also update your Kubelet configuration to use the node name for the address and map in the custom certificate path.

DatadogAgent Kubernetes Resource:

kind: DatadogAgent
apiVersion: datadoghq.com/v2alpha1
metadata:
  name: datadog
spec:
  features:
    admissionController:
      enabled: true
  global:
    credentials:
      apiKey: <DATADOG_API_KEY>
      appKey: <DATADOG_APP_KEY>
    kubelet:
      host:
        fieldRef:
          fieldPath: spec.nodeName
      hostCAPath: /etc/kubernetes/certs/kubeletserver.crt
  override:
    clusterAgent:
      containers:
        cluster-agent:
          env:
            - name: DD_ADMISSION_CONTROLLER_ADD_AKS_SELECTORS
              value: "true"

Custom datadog-values.yaml:

datadog:
  apiKey: <DATADOG_API_KEY>
  appKey: <DATADOG_APP_KEY>
  # Requires supported node image version
  kubelet:
    host:
      valueFrom:
        fieldRef:
          fieldPath: spec.nodeName
    hostCAPath: /etc/kubernetes/certs/kubeletserver.crt

providers:
  aks:
    enabled: true

Using spec.nodeName keeps TLS verification. In some clusters, DNS resolution for spec.nodeName inside Pods may not work in AKS. This has been reported on all AKS Windows nodes, as well as Linux nodes when the cluster is set up in a Virtual Network using custom DNS. In this case, use the first AKS configuration provided: remove any settings for the Kubelet host path (which defaults to status.hostIP) and use tlsVerify: false. This setting is required. Do NOT set the Kubelet host path and tlsVerify: false in the same configuration.

Google Kubernetes Engine (GKE)

GKE can be configured in two different mode of operation:

Standard: You manage the cluster’s underlying infrastructure, giving you node configuration flexibility.
Autopilot: GKE provisions and manages the cluster’s underlying infrastructure, including nodes and node pools, giving you an optimized cluster with a hands-off experience.

Depending on the operation mode of your cluster, the Datadog Agent needs to be configured differently.

Standard

Since Agent 7.26, no specific configuration is required for GKE (whether you run Docker or containerd).

Note: When using COS (Container Optimized OS), the eBPF-based OOM Kill and TCP Queue Length checks are supported starting from the version 3.0.1 of the Helm chart. To enable these checks, configure the following setting:

datadog.systemProbe.enableDefaultKernelHeadersPaths to false.

Autopilot

GKE Autopilot requires some configuration, shown below.

Datadog recommends that you specify resource limits for the Agent container. Autopilot sets a relatively low default limit (50m CPU, 100Mi memory) that may lead the Agent container to quickly OOMKill depending on your environment. If applicable, also specify resource limits for the Trace Agent and Process Agent containers. Additionally, you may wish to create a priority class for the Agent to ensure it is scheduled.

Note: Network Performance Monitoring is not supported for GKE Autopilot.

Custom datadog-values.yaml:

datadog:
  apiKey: <DATADOG_API_KEY>
  appKey: <DATADOG_APP_KEY>
  clusterName: <CLUSTER_NAME>

  # The site of the Datadog intake to send Agent data to (example: `us3.datadoghq.com`)
  # Default value is `datadoghq.com' (the US1 site)
  # Documentation: https://docs.datadoghq.com/getting_started/site/
  site: <DATADOG_SITE>

agents:
  containers:
    agent:
      # resources for the Agent container
      resources:
        requests:
          cpu: 200m
          memory: 256Mi

    traceAgent:
      # resources for the Trace Agent container
      resources:
        requests:
          cpu: 100m
          memory: 200Mi

    processAgent:
      # resources for the Process Agent container
      resources:
        requests:
          cpu: 100m
          memory: 200Mi

  priorityClassCreate: true

providers:
  gke:
    autopilot: true

Spot pods and compute classes

Using Spot Pods in GKE Autopilot clusters introduces taints to the corresponding Spot GKE nodes. When using Spot Pods, additional configuration is required to provide the Agent DaemonSet with a matching toleration.

agents:
  #(...)
  # agents.tolerations -- Allow the DaemonSet to schedule on tainted nodes (requires Kubernetes >= 1.6)
  tolerations:
  - effect: NoSchedule
    key: cloud.google.com/gke-spot
    operator: Equal
    value: "true"

Similarly when using GKE Autopilot Compute classes to run workloads that have specific hardware requirements, take note of the taints that GKE Autopilot is applying to these specific nodes and add matching tolerations to the Agent DaemonSet. You can match the tolerations on your corresponding pods. For example for the Scale-Out compute class use a toleration like:

agents:
  #(...)
  # agents.tolerations -- Allow the DaemonSet to schedule on tainted nodes (requires Kubernetes >= 1.6)
  tolerations:
  - effect: NoSchedule
    key: cloud.google.com/compute-class
    operator: Equal
    value: Scale-Out

Red Hat OpenShift

OpenShift comes with hardened security by default with SELinux and SecurityContextConstraints (SCC). As a result, it requires some specific configurations:

Elevated SCC access for the Node Agent and Cluster Agent
Kubelet API certificates may not always be signed by cluster CA
Tolerations are required to schedule the Node Agent on master and infra nodes
Cluster name should be set as it cannot be retrieved automatically from cloud provider
(Optional) Set hostNetwork: true in the Node Agent to allow the Agent to make requests to cloud provider metadata services (IMDS)

This core configuration supports OpenShift 3.11 and OpenShift 4, but it works best with OpenShift 4.

Additionally log collection and APM have slightly different requirements as well.

The use of Unix Domain Socket (UDS) for APM and DogStatsD can work in OpenShift. However, Datadog does not recommend this, as it requires additional privileged permissions and SCC access to both your Datadog Agent pod and your application pod. Without these, your application pod can fail to deploy. Datadog recommends disabling the UDS option to avoid this, allowing the Admission Controller to inject the appropriate TCP/IP setting or Service setting for APM connectivity.

When using the Datadog Operator in OpenShift, Datadog recommends that you use the Operator Lifecycle Manager to deploy the Datadog Operator from OperatorHub in your OpenShift Cluster web console. Refer to the Operator install steps. The configuration below works with that setup, which creates the ClusterRole and ClusterRoleBinding based access to the SCC for the specified ServiceAccount datadog-agent-scc. This DatadogAgent configuration should be deployed in the same namespace as the Datadog Operator.

kind: DatadogAgent
apiVersion: datadoghq.com/v2alpha1
metadata:
  name: datadog
  namespace: openshift-operators # set as the same namespace where the Datadog Operator was deployed
spec:
  features:
    logCollection:
      enabled: true
      containerCollectAll: true
    apm:
      enabled: true
      hostPortConfig:
        enabled: true
      unixDomainSocketConfig:
        enabled: false
    dogstatsd:
      unixDomainSocketConfig:
        enabled: false
  global:
    credentials:
      apiKey: <DATADOG_API_KEY>
      appKey: <DATADOG_APP_KEY>
    clusterName: <CLUSTER_NAME>
    kubelet:
      tlsVerify: false
  override:
    clusterAgent:
      serviceAccountName: datadog-agent-scc
    nodeAgent:
      serviceAccountName: datadog-agent-scc
      hostNetwork: true
      securityContext:
        runAsUser: 0
        seLinuxOptions:
          level: s0
          role: system_r
          type: spc_t
          user: system_u
      tolerations:
        - key: node-role.kubernetes.io/master
          operator: Exists
          effect: NoSchedule
        - key: node-role.kubernetes.io/infra
          operator: Exists
          effect: NoSchedule

Note: The nodeAgent.securityContext.seLinuxOptions override is necessary for log collection when deploying with the Operator. If log collection is not enabled, you can omit this override.

The configuration below creates custom SCCs for the Agent and Cluster Agent Service Accounts.

Custom datadog-values.yaml:

datadog:
  apiKey: <DATADOG_API_KEY>
  appKey: <DATADOG_APP_KEY>
  clusterName: <CLUSTER_NAME>
  kubelet:
    tlsVerify: false
  apm:
    portEnabled: true
    socketEnabled: false
agents:
  podSecurity:
    securityContextConstraints:
      create: true
  useHostNetwork: true
  tolerations:
    - effect: NoSchedule
      key: node-role.kubernetes.io/master
      operator: Exists
    - effect: NoSchedule
      key: node-role.kubernetes.io/infra
      operator: Exists
clusterAgent:
  podSecurity:
    securityContextConstraints:
      create: true

Rancher

Rancher installations are similar to vanilla Kubernetes installations, requiring only some minor configuration:

Tolerations are required to schedule the Node Agent on controlplane and etcd nodes.
The cluster name should be set as it cannot be retrieved automatically from the cloud provider.

DatadogAgent Kubernetes Resource:

kind: DatadogAgent
apiVersion: datadoghq.com/v2alpha1
metadata:
  name: datadog
spec:
  features:
    logCollection:
      enabled: false
    liveProcessCollection:
      enabled: false
    liveContainerCollection:
      enabled: true
    apm:
      enabled: false
    cspm:
      enabled: false
    cws:
      enabled: false
    npm:
      enabled: false
    admissionController:
      enabled: false
    externalMetricsServer:
      enabled: false
      useDatadogMetrics: false
  global:
    credentials:
      apiKey: <DATADOG_API_KEY>
      appKey: <DATADOG_APP_KEY>
    clusterName: <CLUSTER_NAME>
    kubelet:
      tlsVerify: false
  override:
    clusterAgent:
      image:
        name: gcr.io/datadoghq/cluster-agent:latest
    nodeAgent:
      image:
        name: gcr.io/datadoghq/agent:latest
      tolerations:
        - key: node-role.kubernetes.io/controlplane
          operator: Exists
          effect: NoSchedule
        - key: node-role.kubernetes.io/etcd
          operator: Exists
          effect: NoExecute

Custom datadog-values.yaml:

datadog:
  apiKey: <DATADOG_API_KEY>
  appKey: <DATADOG_APP_KEY>
  clusterName: <CLUSTER_NAME>
  kubelet:
    tlsVerify: false
agents:
  tolerations:
    - effect: NoSchedule
      key: node-role.kubernetes.io/controlplane
      operator: Exists
    - effect: NoExecute
      key: node-role.kubernetes.io/etcd
      operator: Exists

Oracle Container Engine for Kubernetes (OKE)

No specific configuration is required.

vSphere Tanzu Kubernetes Grid (TKG)

TKG requires some small configuration changes, shown below. For example, setting a toleration is required for the controller to schedule the Node Agent on the master nodes.

DatadogAgent Kubernetes Resource:

kind: DatadogAgent
apiVersion: datadoghq.com/v2alpha1
metadata:
  name: datadog
spec:
  features:
    eventCollection:
      collectKubernetesEvents: true
    kubeStateMetricsCore:
      enabled: true
  global:
    credentials:
      apiSecret:
        secretName: datadog-secret
        keyName: api-key
      appSecret:
        secretName: datadog-secret
        keyName: app-key
    kubelet:
      tlsVerify: false
  override:
    nodeAgent:
      tolerations:
        - key: node-role.kubernetes.io/master
          effect: NoSchedule

Custom datadog-values.yaml:

datadog:
  apiKey: <DATADOG_API_KEY>
  appKey: <DATADOG_APP_KEY>
  kubelet:
    # Set tlsVerify to false since the Kubelet certificates are self-signed
    tlsVerify: false
  # Disable the `kube-state-metrics` dependency chart installation.
  kubeStateMetricsEnabled: false
  # Enable the new `kubernetes_state_core` check.
  kubeStateMetricsCore:
    enabled: true
# Add a toleration so that the agent can be scheduled on the control plane nodes.
agents:
  tolerations:
    - key: node-role.kubernetes.io/master
      effect: NoSchedule