Metrics

The agent collects the following metrics on every collection interval (default 10s). All metrics follow the OpenTelemetry semantic conventions and are exported via OTLP.

Metric reference

Metric name	Type	Unit	Description
system.cpu.usage	Gauge	%	Overall CPU usage percentage across all cores
system.memory.used	Gauge	bytes	RAM currently in use
system.memory.total	Gauge	bytes	Total installed RAM
system.disk.used	Gauge	bytes	Disk space used on the root partition
system.disk.total	Gauge	bytes	Total disk capacity of the root partition
system.network.bytes_in	Counter	bytes	Total bytes received across all interfaces
system.network.bytes_out	Counter	bytes	Total bytes sent across all interfaces

OTLP payload example

Each batch sent to the collector looks like the following (simplified JSON representation):

OTLP payload (simplified)

{
  "resource_metrics": [{
    "resource": {
      "attributes": {
        "host.name": "my-server",
        "os.type": "linux"
      }
    },
    "scope_metrics": [{
      "metrics": [
        {
          "name": "system.cpu.usage",
          "gauge": { "data_points": [{ "as_double": 12.4, "time_unix_nano": 1700000000000000000 }] }
        },
        {
          "name": "system.memory.used",
          "gauge": { "data_points": [{ "as_int": 3328000000, "time_unix_nano": 1700000000000000000 }] }
        }
      ]
    }]
  }]
}

Collection details

CPU — read from /proc/stat (Linux) or GetSystemTimes (Windows). Averaged across all logical cores.
Memory — read from /proc/meminfo (Linux) or GlobalMemoryStatusEx (Windows).
Disk — read from statvfs("/") (Linux) or the system drive (Windows). Only the root/system partition is measured in v0.1.
Network — cumulative counters from /proc/net/dev (Linux) or GetIfTable (Windows). All interfaces are summed.

Timestamps

Timestamps are generated at the moment of collection, not at the moment of transmission. This ensures accurate time-series data even when the offline buffer replays delayed batches.