Kubernetes Host Metrics

Overview

Kubernetes Host Metrics is a thin K8s-aware wrapper around the Host Metrics source. It collects node-level OS metrics (CPU, memory, disk, network, load, …) by reading the node's /proc and /sys filesystems rather than the container's. The receiver is intended to run as a DaemonSet so every node in the cluster is sampled.

The wrapper applies two K8s-specific defaults on top of Host Metrics:

1. root_path: /hostfs — the receiver looks for /proc, /sys, etc. under /hostfs instead of /. The pipeline's deployment packaging mounts the node's root filesystem at /hostfs automatically (hostPath: / → /hostfs, read-only) so this just works.
2. scrapers UI shape — accepts a flat array (["cpu", "memory", "load"]) and an optional_metrics array, and translates them into the nested {cpu: true, cpu_optional: {enable: true}} map the underlying Host Metrics receiver expects. This keeps the form schema simple while preserving access to the upstream optional-metric toggles.

For the full scraper / metric list, see Host Metrics.

Supported types: Metrics

Deployment mode: DaemonSet (one pod per node). The packaging layer forces Kind: DaemonSet regardless of any other setting, since per-node host sampling only makes sense as a DaemonSet.

Basic Configuration

Parameter	Type	Default	Required	Description
collection_interval	string	10s	No	How often the receiver samples the host.
scrapers	string[]	—	Yes	List of scrapers to enable. Pick from: cpu, memory, disk, filesystem, load, network, paging, processes, process, system.
optional_metrics	string[]	[]	No	Subset of scrapers to additionally enable each scraper's optional (off-by-default) metrics for. See the Host Metrics doc for the full list.
root_path	string	/hostfs	No	Override only when you mount the node root at a non-default path. The deployment packaging defaults this for you — leave unset in normal use.

K8s Deployment Options

The following fields control how the resulting OpenTelemetryCollector CR / DaemonSet pod is provisioned. They are read by extractDeployMeta and applied at packaging time, not passed to the underlying receiver.

Parameter	Type	Default	Description
kind	string	DaemonSet	Workload kind. Forced to DaemonSet for this source — overrides are ignored.
priorityClass	bool	false	When true, applies priorityClassName to the pod.
priorityClassName	string	—	The PriorityClass to assign (typically system-node-critical for node-agent workloads).
apply_tolerations	bool	false	When true, the pod tolerates all taints — required to schedule on tainted nodes (control-plane, GPU, etc.).

Example Configuration

{
  "collection_interval": "30s",

  "scrapers": ["cpu", "memory", "disk", "filesystem", "load", "network"],

  // Enable optional metrics for cpu and memory only
  "optional_metrics": ["cpu", "memory"],

  // K8s deploy options
  "priorityClass": true,
  "priorityClassName": "system-node-critical",
  "apply_tolerations": true,
}

Notes

• Volume mount is automatic. When this source is used in a pipeline, the cloud-side packaging adds hostPath: / mounted read-only at /hostfs on the DaemonSet pod. You don't need to declare it.
• process scraper requires elevated privileges. Listing other processes' details means the collector pod must run with sufficient host visibility — typically hostPID: true and securityContext.runAsUser: 0. The default unprivileged collector image will silently see only its own process.
• DaemonSet merge. When a pipeline contains two or more of k8s_logs, k8s_kubeletstats, and k8s_hostmetrics, all of them are co-located in a single per-node DaemonSet (<pipeline>-node-agent) instead of separate ones. Default on — set ADL_K8S_MERGE_NODE_AGENT=false on the cloud side to opt out. See the pipeline architecture doc.