ntp_exporter

ntp_exporter

This is a Prometheus exporter that, when running on a node, checks the drift of that node's clock against a given NTP server.

Installation

Compile make && make install or docker build. The binary can also be installed with go get:

go get github.com/sapcc/ntp_exporter

We also publish pre-built images on Docker Hub as sapcc/ntp-exporter:

docker pull sapcc/ntp-exporter:v1.1.3

Usage

Common command-line options:

-ntp.source string
   source of information about ntp server (cli / http). (default "cli")
-version
   Print version information.
-web.listen-address string
   Address on which to expose metrics and web interface. (default ":9559")
-web.telemetry-path string
   Path under which to expose metrics. (default "/metrics")

Mode 1: Fixed NTP server

By default, or when the option -ntp.source cli is specified, the NTP server and connection options is defined by command-line options:

-ntp.measurement-duration duration
   Duration of measurements in case of high (>10ms) drift. (default 30s)
-ntp.protocol-version int
   NTP protocol version to use. (default 4)
-ntp.server string
   NTP server to use (required).

Mode 2: Variable NTP server

When the option -ntp.source http is specified, the NTP server and connection options are obtained from the query parameters on each GET /metrics HTTP request:

• server: NTP server to use
• protocol: NTP protocol version (2, 3 or 4)
• duration: duration of measurements in case of high drift

For example:

$ curl 'http://localhost:9559/metrics?server=ntp.example.com&protocol=4&duration=10s'

Frequently asked questions (FAQ)

Is there a metric for checking that the exporter is working?

Several people have suggested adding a metric like ntp_up that's always 1, so that people can alert on absent(ntp_up) or something like that. This is not necessary. Prometheus already generates such a metric by itself during scraping. A suitable alert expression could look like

up{job="ntp_exporter",instance="example.com:9559"} != 1 or absent(up{job="ntp_exporter",instance="example.com:9559"})

but the concrete labels will vary depending on your setup and scrape configuration.