This feature is a limited release.
This page applies to the Kubernetes Integration version 2 only. Windows nodes are not supported by version 3.
To run our Kubernetes installation for Windows, follow the steps below to:
- Check the compatibility and requirements.
- Update the manifest file.
- Learn about the limitations of the integration.
When using containers in Windows, the container host version and the container image version must be the same. Our Kubernetes integration can run on Windows versions LTSC 2019 (1809), 20H2, and LTSC 2022.
To check your Windows version:
Open a command window.
Run the following command:bash$Reg Query "HKLM\SOFTWARE\Microsoft\Windows NT\CurrentVersion" /v$ReleaseIdcmd.exe
$kubectl exec -it busybox1-766bb4d6cc-rmsnj -- Reg Query$"HKLM\SOFTWARE\Microsoft\Windows NT\CurrentVersion" /v ReleaseId
You should see something like:
$HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows NT\CurrentVersion$ReleaseId REG_SZ 1809
A useful mapping between Release IDs and OS versions can be found here.
You can install the Kubernetes integration for Windows using Helm. See an example on how to install the integration in a cluster with nodes having different build versions of Windows (1809 and 2004):
- Add the New Relic Helm charts repo:
$helm repo add newrelic https://helm-charts.newrelic.com
- Create a namespace for newrelic:
$kubectl create namespace newrelic
- Install kube-state-metrics.
$helm repo add ksm https://kubernetes.github.io/kube-state-metrics$helm install ksm ksm/kube-state-metrics --version 2.13.2
The command above allows installing kube-state-metrics, which is a mandatory dependency of the integration, on a linux node. If you cluster does not have any linux node, the command above might not work. New Relic cannot offer support on how to install kube-state-metrics on non-linux nodes.
- Create a
values.ymlfile with the follow data to be used by Helm:
global:licenseKey: _YOUR_NEW_RELIC_LICENSE_KEY_cluster: _K8S_CLUSTER_NAME_enableLinux: true # Set to true if your cluster also has linux nodesenableWindows: truewindowsOsList:- version: 2019 # Human-readable version identifierimageTag: 2-windows-1809-alpha # Tag to be used for nodes running the windows version abovebuildNumber: 10.0.17763 # Build number for your nodes running the version above. Used as a selector.- version: 20h2imageTag: 2-windows-20H2-alphabuildNumber: 10.0.19042- version: 2022imageTag: 2-windows-ltsc2022-alphabuildNumber: 10.0.20348
- Install the integration with:
$helm upgrade --install newrelic newrelic/newrelic-infrastructure \>--namespace newrelic --create-namespace \>--version 3.2.11 \>-f values-newrelic.yaml
- Check that pods are being deployed and reach a stable state:
$kubectl -n newrelic get pods -w
The Helm chart will create one DeamonSet per each version of Windows that is in the list and use NodeSelector to deploy the corresponding Pod per Node.
The following limitations apply to the Kubernetes integration for Windows:
- The Windows agent only sends the Kubernetes samples (
ProcessSampleare not generated.
- Some Kubernetes metrics are missing because the Windows kubelet doesn’t have them:
fsInodes: not sent
fsInodesFree: not sent
fsInodesUsed: not sent
memoryMajorPageFaultsPerSecond: always returns zero as a value
memoryPageFaults: always returns zero as a value
memoryRssBytes: always returns zero as a value
runtimeInodes: not sent
runtimeInodesFree: not sent
runtimeInodesUsed: not sent
net.errorsPerSecond: not sent
net.rxBytesPerSecond: not sent
net.txBytesPerSecond: not sent
containerID: not sent
containerImageID: not sent
memoryUsedBytes: in the UI, this is displayed in the pod card that appears when you click on a pod, and will show no data. We will soon fix this by updating our charts to use
fsUsedBytes: zero, so
There are a couple of issues with the Windows version of Kubelet that can prevent the integration from fetching data:
- Issue 90554: This issue makes the Kubelet return 500 errors when the integration makes a request to the
/stats/summaryendpoint. It will be included in the Kubernetes 1.19 release and has been backported to the releases 1.16.11, 1.17.7, and 1.18.4. There is no solution on the integration side for this problem, we advise you to update to one of the patch versions as soon as possible. You can see if you're being affected by this problem by enabling verbose logs and looking for messages of the type:
error querying Kubelet. Get "https://<KUBELET_IP>/stats/summary": error calling kubelet endpoint. Got status code: 500
- Issue 87730: This issue makes the Kubelet metrics very slow when running minimal load. It makes the integration fail with a timeout error. A patch for this issue has been added for Kubernetes 1.18 and backported to 1.15.12, 1.16.9, and 1.17.5. We advise you to update to one of the patch versions as soon as possible. To mitigate this issue you can increase the integration timeout with the
TIMEOUTconfig option. You can see if you're being affected by this problem by enabling verbose logs and looking for messages of the type:
error querying Kubelet. Get "https://<KUBELET_IP>/stats/summary": context deadline exceeded (Client.Timeout exceeded while awaiting headers)