Windows Containers Issues

This page describes known and open issues for Containerized Windows Applications Monitoring.

Common issues

Containerized Windows Applications Monitoring requires Datadog Agent 7.19+.

The supported OS versions are:

• Windows Server 2019 (LTSC / 1809)
• Windows Server 2019 1909
• Windows Server 2019 2004 / 20H1
• Windows Server 2019 20H2 (Agent 7.33+)

Hyper-V isolation mode is not supported.

Host metrics for disk, IO, and network are disabled. They are not supported by Windows Server, hence the Agent Checks are disabled by default.

Docker issues

Live processes do not appear in containers (except for the Datadog Agent).

Kubernetes issues

Live processes do not appear in containers (except for the Datadog Agent).

Mixed clusters (Linux + Windows)

The recommended way of deploying the Datadog Agent on a mixed cluster is to perform two installations of the Helm chart with different targetSystems.

The Datadog Agent uses a nodeSelector to automatically select Linux or Windows nodes based on targetSystem.

However it’s not the case for Kube State Metrics (which is installed by default), leading to situations where Kube State Metrics cannot be scheduled on Windows nodes.

Three options are available to avoid this issue:

• Taint your Windows nodes. On Windows, the Agent always allows the node.kubernetes.io/os=windows:NoSchedule taint.
• Set Kube State Metrics node selector through Datatog Helm chart values.yaml:

kube-state-metrics:
  nodeSelector:
    beta.kubernetes.io/os: linux // Kubernetes < 1.14
    kubernetes.io/os: linux // Kubernetes >= 1.14

• Deploy Kube State Metrics yourself separately by setting datadog.kubeStateMetricsEnabled to false.

Note: When using two Datadog installations (one with targetSystem: linux, one with targetSystem: windows), make sure the second one has datadog.kubeStateMetricsEnabled set to false to avoid deploying two instances of Kube State Metrics.

Mixed clusters with the Datadog Cluster Agent

With Cluster Agent v1.18+, a configuration with mixed clusters is supported by the Datadog Cluster Agent.

Use the following values.yaml file to configure communication between Agents deployed on Windows nodes and the Cluster Agent.

targetSystem: windows
existingClusterAgent:
  join: true
  serviceName: "<EXISTING_DCA_SERVICE_NAME>" # from the first Datadog Helm chart
  tokenSecretName: "<EXISTING_DCA_SECRET_NAME>" # from the first Datadog Helm chart

# Disable datadogMetrics deployment since it should have been already deployed with the first chart.
datadog-crds:
  crds:
    datadogMetrics: false
# Disable kube-state-metrics deployment
datadog:
  kubeStateMetricsEnabled: false

Limited configuration options for Windows deployments

Some configuration options are not available on Windows. The following is a list of unsupported options:

HostPort for APM or DogStatsD

HostPort is partially supported on Kubernetes, depending on the underlying OS version and CNI plugin. Requirements to have HostPort working are the following:

• Windows Server version must be >= 1909
• CNI plugin must support portMappings capability

Currently, at least two CNI plugins support this capability:

• Official win-bridge plugin (version >= 0.8.6) - used by GKE
• Azure CNI Plugin - used by AKS

If your setup does not meet these requirements, APM and DogStatsD will only work when pod-to-pod networking is configured between the Tracer and the Agent.

Kubelet check

Depending on your Kubernetes version, some Kubelet metrics might not be available (or Kubelet check might timeout). For optimal experience, please use any of the following:

• Kubelet >= 1.16.13 (1.16.11 on GKE)
• Kubelet >= 1.17.9 (1.17.6 on GKE)
• Kubelet >= 1.18.6
• Kubelet >= 1.19

With Agent version >= 7.19.2

Please note that not all kubernetes.* are available on Windows, you can find the list of available ones below:

• kubernetes.cpu.usage.total
• kubernetes.containers.restarts
• kubernetes.containers.running
• kubernetes.cpu.capacity
• kubernetes.ephemeral_storage.usage
• kubernetes.kubelet.container.log_filesystem.used_bytes
• kubernetes.kubelet.network_plugin.latency.count
• kubernetes.kubelet.network_plugin.latency.quantile
• kubernetes.kubelet.network_plugin.latency.sum
• kubernetes.kubelet.runtime.errors
• kubernetes.kubelet.runtime.operations
• kubernetes.memory.capacity
• kubernetes.pods.running
• kubernetes.rest.client.latency.count
• kubernetes.rest.client.latency.sum
• kubernetes.rest.client.requests
• kubernetes.network.tx_bytes
• kubernetes.network.rx_bytes
• kubernetes.cpu.usage.total
• kubernetes.memory.working_set
• kubernetes.filesystem.usage
• kubernetes.filesystem.usage_pct