VM monitoring

Inspect VM activity and distinguish a resource problem from a missing measurement. Expanded guest monitoring is opt-in and available in DC-1.

Read the charts

Open a VM's Monitoring tab and select a time range. Hosted history retains seven days with 30-second source sampling. Wider ranges use coarser chart steps. Refreshing does not force the guest to produce a new sample.

• CPU: observed vCPU utilization normalized to capacity.
• Memory: guest capacity minus free memory, including cache; not application working set.
• Disk and network: observed traffic rates, not billing totals.
• Filesystem: guest-reported usage when supported; not the size of an attached disk.

Missing data is not zero

Available means a measurement is present; stale means its observation is too old; awaiting data means no fresh sample has arrived; unsupported means that source cannot supply it; error means the measurement could not be obtained. None of these labels alone establishes whether an application is healthy. Check observation times and the individual metric's explanation.

Optional guest metrics

On a new Ubuntu or Debian image, opt in while creating the VM. For an existing disk or another image, install from this page after the VM is running. Opening the page never restarts a VM.

1. Choose expanded monitoring on the VM. Read the device-readiness check first. If attachment requires a stop/start, explicitly schedule and approve that interruption.
2. Download the configuration for this VM. It contains installation identifiers, not an API key; copying it to another VM does not authorize collection.
3. Download the exact collector version offered for the environment. Verify its release signature and checksum before installing. Follow the installation guide included in that verified archive; do not substitute an arbitrary development artifact.
4. Run the displayed installation commands inside the guest. Metalhost does not execute them remotely. Start only the collector service and check its status.
5. Wait for Connected and a recent observation time. Counter-derived rates can take two minutes to appear.

The collector uses a private VM-to-host transport. It does not expose a guest HTTP port, shell, logs, files or command execution. Guest administrators can falsify their own counters, so these are not billing or provider-health evidence.

Services and GPUs

Service discovery is opt-in. Apply its configuration and, when required, the verified package's service-account profile; then select up to 32 discovered services. Discovery can be truncated and a missing service must not be treated as healthy. Apply changed configuration locally and restart only the collector.

GPU metrics require existing NVIDIA drivers and the package's explicit device-access profile. No driver is installed and no GPU settings are changed. Unsupported sensors remain absent. Do not run the collector as root or weaken its sandbox to force coverage.

Pause versus revoke

Pause stops accepting guest measurements while preserving the installation so you can resume it. Revoke permanently invalidates that installation. Neither action removes software from the guest or stops baseline monitoring. Uninstalling requires customer-run commands from the verified collector guide. Historical data remains subject to retention.

Use the same measurements elsewhere

Set up alerts and notifications, or connect Grafana and Prometheus. These are infrastructure measurements, not an external website-uptime probe. CLI v1.1.2 and SDK v1.1.2 cover the same APIs.