Skip to content
FuseScopeFuseScopeFuseScope

Hosts

The Hosts tab of Infrastructure lists your machines with their CPU, memory, disk and load. That data comes from the OpenTelemetry Collector’s hostmetrics receiver, running on each machine, plus the resourcedetection processor, which names the machine.

Use this page when the Hosts tab says No host metrics yet, an instrumentation check reports missing host metrics or host.name, or you add a new server or VM.

On Kubernetes you don’t need this page: the k8s-infra chart already runs the receiver on every node. See hostmetrics receiver.

What the Hosts tab needs

Section titled “What the Hosts tab needs”
Data Produced by Used for
system.cpu.time hostmetrics cpu scraper CPU usage and Iowait columns
system.memory.usage memory scraper Memory usage column
system.cpu.load_average.15m load scraper Load avg (15min) column
system.filesystem.usage filesystem scraper Disk usage column
host.name resource attribute resourcedetection processor One row per host; without it the list is empty
os.type resource attribute resourcedetection (system detector) OS Type filter

These four metrics and host.name are what the instrumentation checks on the Hosts tab look for.

The host’s details panel draws more charts from the same receiver:

Charts Metrics Scraper
System load average system.cpu.load_average.1m, .5m, .15m load
Disk IO, operations, operation time, queue size system.disk.io, system.disk.operations, system.disk.operation_time, system.disk.pending_operations disk
Network usage, packets, errors, drops, connections system.network.io, system.network.packets, system.network.errors, system.network.dropped, system.network.connections network

Process metrics are optional. FuseScope’s processes check looks for process.cpu.time and process.memory.usage with the process.pid resource attribute, from the process scraper. The host panel has no processes view yet; chart them in Metrics.

The config

Section titled “The config”

Add this to the collector on the machine. It extends the base config from Collector on a VM or Collector on Docker: keep your otlp receiver, exporter and other pipelines.

config.yaml
receivers:
  hostmetrics:
    collection_interval: 30s
    scrapers:
      cpu: {}
      memory: {}
      load: {}
      filesystem: {}
      disk: {}
      network: {}


processors:
  resourcedetection:
    detectors: [env, system]
    system:
      hostname_sources: [os]
  memory_limiter:
    check_interval: 1s
    limit_percentage: 80
    spike_limit_percentage: 25
  batch: {}


exporters:
  otlphttp:
    endpoint: http://fusescope-collector.example.com:4318


service:
  pipelines:
    metrics/host:
      receivers: [hostmetrics]
      processors: [memory_limiter, resourcedetection, batch]
      exporters: [otlphttp]
  • hostname_sources: [os] sets host.name to the machine’s hostname. The default, [dns, os], may use a fully qualified DNS name instead.
  • env reads OTEL_RESOURCE_ATTRIBUTES from the collector’s environment and runs first, so a value there wins, for example OTEL_RESOURCE_ATTRIBUTES=host.name=web-1,deployment.environment.name=production.
  • To also tag app traces and logs with host.name, add resourcedetection to those pipelines too. Traces in the host panel are matched by host.name.

On a cloud VM

Section titled “On a cloud VM”

Cloud detectors add the instance’s region, zone and ID, and on most clouds a host.name of their own. Add the one for your cloud:

processors:
  resourcedetection:
    detectors: [env, ec2, system]   # or gcp, azure
    system:
      hostname_sources: [os]

When several detectors set the same attribute, the first one in the list wins. Put system before the cloud detector if you want the OS hostname as host.name.

Optional metrics

Section titled “Optional metrics”

system.cpu.utilization and system.memory.utilization (0 to 1) are off by default. The Hosts tab doesn’t need them; turn them on if your own dashboards or alerts use them:

receivers:
  hostmetrics:
    scrapers:
      cpu:
        metrics:
          system.cpu.utilization:
            enabled: true
      memory:
        metrics:
          system.memory.utilization:
            enabled: true

Process metrics

Section titled “Process metrics”

The process scraper sends one series per process, which adds many series on a busy machine. Without root it can’t read other users’ processes, and logs an error for each one. Mute those errors:

receivers:
  hostmetrics:
    scrapers:
      process:
        mute_process_name_error: true
        mute_process_exe_error: true
        mute_process_io_error: true
        mute_process_user_error: true
        mute_process_cgroup_error: true

mute_process_all_errors: true mutes them all at once. Use include or exclude with names to limit which processes are scraped.

Run it

Section titled “Run it”

Install otelcol-contrib as in Collector on a VM, add the config above to /etc/otelcol-contrib/config.yaml, then:

Terminal window
sudo systemctl restart otelcol-contrib
sudo journalctl -u otelcol-contrib -f

The receiver reads /proc and /sys directly; no root_path is needed.

Check that it works

Section titled “Check that it works”
  1. Look for errors in the collector logs, such as failed to read from a scraper.
  2. Open Infrastructure › Hosts. The host appears with Active status within a few collection intervals. Click Recheck on the instrumentation checks.
  3. If the host is missing but metrics arrive, search for system.cpu.time in Metrics and check that it has host.name.
Section titled “Related”