A Prometheus/OpenMetrics exporter for uninterruptable power supplies (UPSes) using Network UPS Tools (NUT).
Set up NUT in server mode and make sure the TCP port (3493 by default) is accessible (without authentication).
If you want to test that it's working, run telnet <nut-server> 3493
and then VER
, LIST UPS
and LIST VAR <ups>
.
Example docker-compose.yml
:
version: "3.7"
services:
nut-exporter:
# Stable v1
image: hon95/prometheus-nut-exporter:1
environment:
- TZ=Europe/Oslo
- HTTP_PATH=/metrics
# Defaults
#- RUST_LOG=info
#- HTTP_PORT=9995
#- HTTP_PATH=/nut
#- LOG_REQUESTS_CONSOLE=false
#- PRINT_METRICS_AND_EXIT=false
ports:
- "9995:9995/tcp"
Example prometheus.yml
:
global:
scrape_interval: 15s
scrape_timeout: 10s
scrape_configs:
- job_name: "nut"
static_configs:
# Insert NUT server address here
- targets: ["nut-server:3493"]
metrics_path: /nut
relabel_configs:
- source_labels: [__address__]
target_label: __param_target
- source_labels: [__param_target]
target_label: instance
- target_label: __address__
# Insert NUT exporter address here
replacement: nut-exporter:9995
In the above example, nut-exporter:9995
is the address and port of the NUT exporter while nut-server:3493
is the address and port of the NUT server to query through the exporter.
Example container resources requests and limits. This was done by scraping one NUT server with two UPSes. Resource usage was observed across 7 days period. This can be lowered even more but should be sufficient as a starting point.
resources:
limits:
cpu: "10m"
memory: "16Mi"
requests:
cpu: "1m"
memory: "8Mi"
Use e.g. 1
for stable v1.y.z releases and latest
for bleeding/dev/unstable releases.
RUST_LOG
(defaults to info
): The log level used by the console/STDOUT. Set to debug
so show HTTP requests and trace
to show extensive debugging info.HTTP_ADDRESS
(defaults to ::
): The HTTP server will listen on this IP. Set to 127.0.0.1
or ::1
to only allow local access.HTTP_PORT
(defaults to 9995
): The HTTP server port.HTTP_PATH
(defaults to nut
): The HTTP server metrics path. You may want to set it to /metrics
on new setups to avoid extra Prometheus configuration (not changed here due to compatibility).PRINT_METRICS_AND_EXIT
(defaults to false
): Print a Markdown-formatted table consisting of all metrics and then immediately exit. Used mainly for generating documentation.See metrics.
GNU General Public License version 3 (GPLv3).