OTel host receiver and process metrics

The OTel host receiver collects host metrics in OTLP format, replacing node_exporter. The process receiver adds per-process metrics. Together they cover the standard host-monitoring needs in OTel form.

Host receiver

The hostmetrics receiver collects:

• CPU usage (per core and aggregate).
• Memory (total, available, used, cached).
• Disk I/O (per device, per mount).
• Network I/O (per interface).
• Filesystem usage.
• Load average.
• Process count and state.
• Paging.

receivers:
  hostmetrics:
    collection_interval: 30s
    scrapers:
      cpu:
        metrics:
          system.cpu.utilization:
            enabled: true
      memory:
      disk:
      load:
      processes:
      paging:
      filesystem:
        include_mount_points:
          mount_points: [/, /var]
          match_type: strict
      network:
        include:
          interfaces: [eth0]
          match_type: strict

Three details in that config are easy to get wrong.

No type: field. In collector configuration the map key is the component type. Writing type: hostmetrics inside the hostmetrics: block adds an unknown key, and the collector refuses to start with has invalid keys: type.

Each scraper appears exactly once. network: is listed once, with its filter. Listing it twice is a hard failure - see below.

Filters are nested, not bare lists. The filesystem scraper takes include_mount_points.mount_points plus a match_type; the network scraper takes include.interfaces plus a match_type. A bare mount_points: or interfaces: list at the scraper level is not a valid key.

The receiver runs as part of the OTel agent (or gateway). Output is in OTLP format and can be exported to Prometheus or any OTel-compatible backend.

Process scraper

Per-process metrics are not a separate receiver. They come from the process scraper inside the same hostmetrics receiver, so they go in the same scrapers: map:

receivers:
  hostmetrics:
    collection_interval: 30s
    scrapers:
      cpu:
      memory:
      process:
        metrics:
          process.cpu.utilization:
            enabled: true
          process.memory.usage:
            enabled: true
          process.disk.io:
            enabled: true

Note the difference between two similarly named scrapers. processes: reports system-wide process counts by state. process: reports metrics per individual process.

Per-process metrics are valuable for identifying which process is the resource consumer. They replace ad-hoc tools like pidstat for monitoring purposes.

Export to Prometheus

exporters:
  prometheus:
    endpoint: 0.0.0.0:8889
    resource_to_telemetry_conversion:
      enabled: true

service:
  pipelines:
    metrics:
      receivers: [hostmetrics]
      exporters: [prometheus]

The pipeline lists hostmetrics only. There is no process receiver to list - the process scraper is already inside hostmetrics. Naming a component in a pipeline that is not defined in the corresponding section is another start-up failure.

The OTel Collector exposes Prometheus metrics on port 8889. Prometheus scrapes this like any other target.

The resource_to_telemetry_conversion adds resource attributes (host, service) as Prometheus labels.

Replace node_exporter

For a fleet already on OTel, the host receiver replaces node_exporter:

Migration:

1. Validate the config on one host: otelcol validate --config /etc/otelcol/config.yaml.
2. Deploy the OTel agent with the host receiver on that one host and confirm the service is active (running), not restarting.
3. Configure Prometheus to scrape the OTel endpoint.
4. Update dashboards to use the OTel metric names.
5. Run both node_exporter and the OTel agent in parallel until the new dashboards and alerts have fired at least once on real data.
6. Only then remove node_exporter.

Metric name changes:

• node_cpu_seconds_total → system.cpu.time (or system.cpu.utilization).
• node_memory_MemAvailable_bytes → system.memory.available.
• node_filesystem_size_bytes → system.filesystem.usage.

The semantic is the same; the name is OTel-idiomatic.

Knowledge check