mirror of
https://github.com/ilya-zlobintsev/LACT.git
synced 2026-08-17 16:34:54 -05:00
Ilya Zlobintsev
3d50c4dd44
* feat: opentelemetry metrics exporter * chore: run first export after the period has passed * feat: include gpu name in attributes * chore: fix typo * chore: update docs * fix: disable coverage action
1.4 KiB
1.4 KiB
OpenTelemetry exporter
The LACT daemon supports exporting GPU metrics using the standard OpenTelemetry metrics API. In particular, it uses the OTLP/HTTP protocol with a JSON payload.
To use this functionality, you need an OpenTelemetry-compatible metrics collector such as Prometheus with OTLP support enabled.
After this, you need to set the collector URL in /etc/lact/config.yaml. For example:
daemon:
metrics:
# This example is a local Prometheus URL. Other collectors might simply use `/v1/metrics` as the path.
collector_address: 'http://localhost:9090/api/v1/otlp/v1/metrics'
# Metrics collection interval in seconds. Not necessary to set, defaults to 30 seconds.
interval: 30
Restart the daemon after changing the config.
The following metrics are currently supported :
- lact_gpu_usage
- lact_gpu_power_usage
- lact_gpu_power_usage_cap
- lact_gpu_temperature
- lact_gpu_clockspeed
- lact_gpu_voltage
- lact_gpu_fan_speed
- lact_gpu_fan_speed_max
- lact_gpu_fan_percent
- lact_gpu_vram_used
- lact_gpu_vram_total
Note: this is the source metric name, collectors might alter it for queries. For example, Prometheus adds a unit suffix, e.g. lact_gpu_power_usage -> lact_gpu_power_usage_watts
