Overview

Get metrics from Kubernetes service in real-time to:

• Visualize and monitor Kubernetes states.
• Be notified about Kubernetes failovers and events.

The Kubernetes State Metrics Core check leverages kube-state-metrics version 2+ and includes major performance and tagging improvements compared to the legacy kubernetes_state check.

As opposed to the legacy check, with the Kubernetes State Metrics Core check, you no longer need to deploy kube-state-metrics in your cluster.

Kubernetes State Metrics Core provides a better alternative to the legacy kubernetes_state check as it offers more granular metrics and tags. See the Major Changes and Data Collected for more details.

Setup

Installation

The Kubernetes State Metrics Core check is included in the Datadog Cluster Agent image, so you don’t need to install anything else on your Kubernetes servers.

Requirements

• Datadog Cluster Agent v1.12+

Configuration

datadog:
  # (...)
  kubeStateMetricsCore:
    enabled: true

kind: DatadogAgent
apiVersion: datadoghq.com/v2alpha1
metadata:
  name: datadog
spec:
  global:
    credentials:
      apiKey: <DATADOG_API_KEY>
  features:
    kubeStateMetricsCore:
      enabled: true

Migration from kubernetes_state to kubernetes_state_core

Tags removal

In the original kubernetes_state check, several tags have been flagged as deprecated and replaced by new tags. To determine your migration path, check which tags are submitted with your metrics.

In the kubernetes_state_core check, only the non-deprecated tags are submitted. Before migrating from kubernetes_state to kubernetes_state_core, verify that only official tags are used in monitors and dashboards.

Here is the mapping between deprecated tags and the official tags that have replaced them:

Backward incompatibility changes

The Kubernetes State Metrics Core check is not backward compatible, be sure to read the changes carefully before migrating from the legacy kubernetes_state check.

kubernetes_state.node.by_condition

A new metric with node name granularity. The legacy metric kubernetes_state.nodes.by_condition is deprecated in favor of this one. Note: This metric is backported into the legacy check, where both metrics (it and the legacy metric it replaces) are available.

kubernetes_state.persistentvolume.by_phase

A new metric with persistentvolume name granularity. It replaces kubernetes_state.persistentvolumes.by_phase.

kubernetes_state.pod.status_phase

The metric is tagged with pod level tags, like pod_name.

kubernetes_state.node.count

The metric is not tagged with host anymore. It aggregates the nodes count by kernel_version os_image container_runtime_version kubelet_version.

kubernetes_state.container.waiting and kubernetes_state.container.status_report.count.waiting

These metrics no longer emit a 0 value if no pods are waiting. They only report non-zero values.

kube_job

In kubernetes_state, the kube_job tag value is the CronJob name if the Job had CronJob as an owner, otherwise it is the Job name. In kubernetes_state_core, the kube_job tag value is always the Job name, and a new kube_cronjob tag key is added with the CronJob name as the tag value. When migrating to kubernetes_state_core, it’s recommended to use the new tag or kube_job:foo*, where foo is the CronJob name, for query filters.

kubernetes_state.job.succeeded

In kubernetes_state, the kubernetes.job.succeeded was count type. In kubernetes_state_core it is gauge type.

Node-level tag assignment

Host or node-level tags no longer appear on cluster-centric metrics. Only metrics relative to an actual node in the cluster, like kubernetes_state.node.by_condition or kubernetes_state.container.restarts, continue to inherit their respective host or node level tags.

To add tags globally, use the DD_TAGS environment variable, or use the respective Helm or Operator configurations. Instance-only level tags can be specified by mounting a custom kubernetes_state_core.yaml into the Cluster Agent.

datadog:
  kubeStateMetricsCore:
    enabled: true
  tags: 
    - "<TAG_KEY>:<TAG_VALUE>"

kind: DatadogAgent
apiVersion: datadoghq.com/v2alpha1
metadata:
  name: datadog
spec:
  global:
    credentials:
      apiKey: <DATADOG_API_KEY>
    tags:
      - "<TAG_KEY>:<TAG_VALUE>"
  features:
    kubeStateMetricsCore:
      enabled: true

Metrics like kubernetes_state.container.memory_limit.total or kubernetes_state.node.count are aggregate counts of groups within a cluster, and host or node-level tags are not added.

Legacy check

datadog:
  # (...)
  kubeStateMetricsEnabled: false

Important Note: The Kubernetes State Metrics Core check is an alternative to the legacy kubernetes_state check. Datadog recommends not enabling both checks simultaneously to guarantee consistent metrics.

Data Collected

Metrics

Note: You can configure Datadog Standard labels on your Kubernetes objects to get the env service version tags.

Events

The Kubernetes State Metrics Core check does not include any events.

Default labels as tags

Default recommended Kubernetes and Helm labels

Default recommended Kubernetes node labels

Datadog labels (Unified Service Tagging)

Service Checks

kubernetes_state.cronjob.complete

Whether the last job of the cronjob is failed or not. Tags:kube_cronjob kube_namespace (env service version from standard labels).

kubernetes_state.cronjob.on_schedule_check

Alert if the cronjob’s next schedule is in the past. Tags:kube_cronjob kube_namespace (env service version from standard labels).

kubernetes_state.job.complete

Whether the job is failed or not. Tags:kube_job or kube_cronjob kube_namespace (env service version from standard labels).

kubernetes_state.node.ready

Whether the node is ready. Tags:node condition status.

kubernetes_state.node.out_of_disk

Whether the node is out of disk. Tags:node condition status.

kubernetes_state.node.disk_pressure

Whether the node is under disk pressure. Tags:node condition status.

kubernetes_state.node.network_unavailable

Whether the node network is unavailable. Tags:node condition status.

kubernetes_state.node.memory_pressure

Whether the node network is under memory pressure. Tags:node condition status.

Validation

Run the Cluster Agent’s status subcommand inside your Cluster Agent container and look for kubernetes_state_core under the Checks section.

Troubleshooting

Timeout errors

By default, the Kubernetes State Metrics Core check waits 10 seconds for a response from the Kubernetes API server. For large clusters, the request may time out, resulting in missing metrics.

You can avoid this by setting the environment variable DD_KUBERNETES_APISERVER_CLIENT_TIMEOUT to a higher value than the default 10 seconds.

apiVersion: datadoghq.com/v2alpha1
kind: DatadogAgent
metadata:
  name: datadog
spec:
  override:
    clusterAgent:
      env:
        - name: DD_KUBERNETES_APISERVER_CLIENT_TIMEOUT
          value: <value_greater_than_10>

kubectl apply -n $DD_NAMESPACE -f datadog-agent.yaml

clusterAgent:
  env:
    - name: DD_KUBERNETES_APISERVER_CLIENT_TIMEOUT
      value: <value_greater_than_10>

helm upgrade -f datadog-values.yaml <RELEASE_NAME> datadog/datadog

Need help? Contact Datadog support.

Further Reading

• Our Journey Taking Kubernetes State Metrics to the Next Level