Prometheus IPMI Exporter

This is an IPMI exporter for Prometheus.

It supports both the regular /metrics endpoint, exposing metrics from the host that the exporter is running on, as well as an /ipmi endpoint that supports IPMI over RMCP - one exporter running on one host can be used to monitor a large number of IPMI interfaces by passing the target parameter to a scrape.

The exporter relies on tools from the FreeIPMI suite for the actual IPMI implementation.

Installation

You need a Go development environment. Then, run the following to get the source code and build and install the binary:

go get github.com/soundcloud/ipmi_exporter

Running

A minimal invocation looks like this:

./ipmi_exporter

Supported parameters include:

web.listen-address: the address/port to listen on (default: ":9290")
config.file: path to the configuration file (default: none)
freeipmi.path: path to the FreeIPMI executables (default: rely on $PATH)

For syntax and a complete list of available parameters, run:

./ipmi_exporter -h

Make sure you have the following tools from the FreeIPMI suite installed:

ipmimonitoring/ipmi-sensors
ipmi-dcmi
bmc-info

Configuration

Simply scraping the standard /metrics endpoint will make the exporter emit local IPMI metrics. No special configuration is required.

For remote metrics, the general configuration pattern is similar to that of the blackbox exporter, i.e. Prometheus scrapes a small number (possibly one) of IPMI exporters with a target and module URL parameter to tell the exporter which IPMI device it should use to retrieve the IPMI metrics. We offer this approach as IPMI devices often provide useful information even while the supervised host is turned off. If you are running the exporter on a separate host anyway, it makes more sense to have only a few of them, each probing many (possibly thousands of) IPMI devices, rather than one exporter per IPMI device.

IPMI exporter

The exporter can read a configuration file by setting config.file (see above). To collect local metrics, you might not even need one. For remote metrics, it must contain at least user names and passwords for IPMI access to all targets to be scraped. You can additionally specify the IPMI driver type and privilege level to use (see man 5 freeipmi.conf for more details and possible values).

The config file supports the notion of "modules", so that different configurations can be re-used for groups of targets. See the section below on how to set the module parameter in Prometheus. The special module "default" is used in case the scrape does not request a specific module.

The configuration file also supports a blacklist of sensors, useful in case of OEM-specific sensors that FreeIPMI cannot deal with properly or otherwise misbehaving sensors. This applies to both local and remote metrics.

There are two commented example configuration files, see ipmi_local.yml for scraping local host metrics and ipmi_remote.yml for scraping remote IPMI interfaces.

Prometheus

Local metrics

Collecting local IPMI metrics is fairly straightforward. Simply configure your server to scrape the default metrics endpoint on the hosts running the exporter.

- job_name: ipmi
  scrape_interval: 1m
  scrape_timeout: 30s
  metrics_path: /metrics
  scheme: http
  static_configs:
  - targets:
    - 10.1.2.23:9290
    - 10.1.2.24:9290
    - 10.1.2.25:9290

Remote metrics

To add your IPMI targets to Prometheus, you can use any of the supported service discovery mechanism of your choice. The following example uses the file-based SD and should be easy to adjust to other scenarios.

Create a YAML file that contains a list of targets, e.g.:

---
- targets:
  - 10.1.2.23
  - 10.1.2.24
  - 10.1.2.25
  - 10.1.2.26
  - 10.1.2.27
  - 10.1.2.28
  - 10.1.2.29
  - 10.1.2.30
  labels:
    job: ipmi_exporter

This file needs to be stored on the Prometheus server host. Assuming that this file is called /srv/ipmi_exporter/targets.yml, and the IPMI exporter is running on a host that has the DNS name ipmi-exporter.internal.example.com, add the following to your Prometheus config:

- job_name: ipmi
  params:
    module: default
  scrape_interval: 1m
  scrape_timeout: 30s
  metrics_path: /ipmi
  scheme: http
  file_sd_configs:
  - files:
    - /srv/ipmi_exporter/targets.yml
    refresh_interval: 5m
  relabel_configs:
  - source_labels: [__address__]
    separator: ;
    regex: (.*)
    target_label: __param_target
    replacement: ${1}
    action: replace
  - source_labels: [__param_target]
    separator: ;
    regex: (.*)
    target_label: instance
    replacement: ${1}
    action: replace
  - separator: ;
    regex: .*
    target_label: __address__
    replacement: ipmi-exporter.internal.example.com:9290
    action: replace

This assumes that all hosts use the default module. If you are using modules in the config file, like in the provided ipmi_remote.yml example config, you will need to specify on job for each module, using the respective group of targets.

In a more extreme case, for example if you are using different passwords on every host, a good approach is to generate an exporter config file that uses the target name as module names, which would allow you to have single job that uses label replace to set the module. Leave out the params in the job definition and instead add a relabel rule like this one:

  - source_labels: [__address__]
    separator: ;
    regex: (.*)
    target_label: __param_module
    replacement: ${1}
    action: replace

For more information, e.g. how to use mechanisms other than a file to discover the list of hosts to scrape, please refer to the Prometheus documentation.

Exported data

Scrape meta data

These metrics provide data about the scrape itself:

ipmi_up{collector="<NAME>"} is 1 if the data for this collector could successfully be retrieved from the remote host, 0 otherwise. The following collectors are available and can be enabled or disabled in the config:
- ipmi: collects IPMI sensor data. If it fails, sensor metrics (see below) will not be available
- dcmi: collects DCMI data, currently only power consumption. If it fails, power consumption metrics (see below) will not be available
- bmc: collects BMC details. If if fails, BMC info metrics (see below) will not be available
ipmi_scrape_duration_seconds is the amount of time it took to retrieve the data

BMC info

This metric is only provided if the bmc collector is enabled.

For some basic information, there is a constant metric ipmi_bmc_info with value 1 and labels providing the firmware revision and manufacturer as returned from the BMC. Example:

ipmi_bmc_info{firmware_revision="2.52",manufacturer_id="Dell Inc. (674)"} 1

Power consumption

This metric is only provided if the dcmi collector is enabled.

The metric ipmi_dcmi_power_consumption_current_watts can be used to monitor the live power consumption of the machine in Watts. If in doubt, this metric should be used over any of the sensor data (see below), even if their name might suggest that they measure the same thing. This metric has no labels.

Sensors

These metric are only provided if the ipmi collector is enabled.

IPMI sensors in general have one or two distinct pieces of information that are of interest: a value and/or a state. The exporter always exports both, even if the value is NaN or the state non-sensical. This is so one can still always find the metrics to avoid ending up in a situation where one is looking for e.g. the value of a sensor that is in a critical state, but can't find it and assume this to be a problem.

The state of a sensor can be one of nominal, warning, critical, or N/A, reflected by the metric values 0, 1, 2, and NaN respectively. Think of this as a kind of severity.

For sensors with known semantics (i.e. units), corresponding specific metrics are exported. For everything else, generic metrics are exported.

Temperature sensors

Temperature sensors measure a temperature in degrees Celsius and their state usually reflects the temperature going above the vendor-recommended value. For each temperature sensor, two metrics are exported (state and value), using the sensor ID and the sensor name as labels. Example:

ipmi_temperature_celsius{id="18",name="Inlet Temp"} 24
ipmi_temperature_state{id="18",name="Inlet Temp"} 0

Fan speed sensors

Fan speed sensors measure fan speed in rotations per minute (RPM) and their state usually reflects the speed being to low, indicating the fan might be broken. For each fan speed sensor, two metrics are exported (state and value), using the sensor ID and the sensor name as labels. Example:

ipmi_fan_speed_rpm{id="12",name="Fan1A"} 4560
ipmi_fan_speed_state{id="12",name="Fan1A"} 0

Voltage sensors

Voltage sensors measure a voltage in Volts. For each voltage sensor, two metrics are exported (state and value), using the sensor ID and the sensor name as labels. Example:

ipmi_voltage_state{id="2416",name="12V"} 0
ipmi_voltage_volts{id="2416",name="12V"} 12

Current sensors

Current sensors measure a current in Amperes. For each current sensor, two metrics are exported (state and value), using the sensor ID and the sensor name as labels. Example:

ipmi_current_state{id="83",name="Current 1"} 0
ipmi_current_amperes{id="83",name="Current 1"} 0

Power sensors

Power sensors measure power in Watts. For each power sensor, two metrics are exported (state and value), using the sensor ID and the sensor name as labels. Example:

ipmi_power_state{id="90",name="Pwr Consumption"} 0
ipmi_power_watts{id="90",name="Pwr Consumption"} 70

Note that based on our observations, this may or may not be a reading reflecting the actual live power consumption. We recommend using the more explicit power consumption metrics for this.

Generic sensors

For all sensors that can not be classified, two generic metrics are exported, the state and the value. However, to provide a little more context, the sensor type is added as label (in addition to name and ID). Example:

ipmi_sensor_state{id="139",name="Power Cable",type="Cable/Interconnect"} 0
ipmi_sensor_value{id="139",name="Power Cable",type="Cable/Interconnect"} NaN

pandada8/ipmi_exporter