redfish_exporter

A Prometheus multi-target exporter to fetch metrics from Redfish-capable hardware. Forked from https://github.com/FlxPeters/redfish_exporter, see Why a fork for rationale.

Configuration

Configuration of the redfish_exporter is done via YAML file.

See CONFIGURATION.md for details.

Building

To build the redfish_exporter executable:

make build

At the current time, this fork does not publish its own container images. We hope to in the near future.

Running

Running directly on Linux

The exporter can run directly on Linux as a binary:

redfish_exporter -config.file=redfish_exporter.yaml

Run redfish_exporter -h for more options.

Running in a container

If using an image, an invocation like the following should also work:

# For Podman users
podman run -v $(pwd)/redfish_exporter.yaml:/config.yml -p 9610:9610  some-image-reference:<some-image-tag> -config.file="/config.yml"
# For Docker users
docker run -v $(pwd)/redfish_exporter.yaml:/config.yml -p 9610:9610  some-image-reference:<some-image-tag> -config.file="/config.yml"

Scraping

Clients - like Prometheus, your browser, or cURL - collect Redfish metrics from a system via the /redfish endpoint and at minimum, a target parameter:

curl -s http://<redfish_exporter host>:9610/redfish?target=<ip>

If using modules, clients would specify one or more module parameters:

# Single module
curl -s "http://localhost:9610/redfish?target=<ip>&module=telemetry"
# Multiple modules, where 'rf_version' is possibly a custom JSON collector
curl -s "http://localhost:9610/redfish?target=<ip>&module=telemetry&module=rf_version"

Reloading Configuration

PUT /-/reload
POST /-/reload

The /-/reload endpoint triggers a reload of the redfish_exporter configuration. 500 will be returned when the reload fails.

Alternatively, a configuration reload can be triggered by sending SIGHUP to the redfish_exporter process as well.

Prometheus Configuration

Prometheus scrapes targets using something like the following scrape job configuration:

  - job_name: 'redfish-exporter'

    # metrics_path defaults to '/metrics'
    metrics_path: /redfish
    
    # params below will apply to all targets in the scrape job. Useful in homogeneous hardware deployments (like job-by-hardware-type)
    params:
    # (optional) when using modules
      module: [telemetry, gpu]
      # (optional) when using group config
      group: [foo]
    # scheme defaults to 'http'.
    scheme: http
    
    static_configs:
    - targets:
       - <ip> # List of IPs to collect Redfish metrics from
    relabel_configs:
      - source_labels: [__address__]
        target_label: __param_target
      - source_labels: [__param_target]
        target_label: instance
      - target_label: __address__
        replacement: localhost:9610  # The address of the redfish-exporter
      # NOTE: If certain params (like module or group) may be variable per host in the job, often depending on service discovery, do not use 'params' above but instead add relabels like so:
      - target_label: __param_group
        replacement: "foo" # likely a regex and using service discovery data
      - target_label: __param_module
        replacement: "foobar"

Testing and Development

Mock Server

We provide a mock Redfish server for testing the exporter without requiring actual hardware:

# Test with specific captured data
make mock-test TESTDATA=my_system/capture.txt

This starts both a mock Redfish server (with HTTP and HTTPS support) and the exporter configured to scrape it.

Capturing Real Hardware Data

To capture data from real hardware for testing:

# Capture Redfish responses from a BMC
make capture HOST=10.0.0.100 USER=admin PASS=password OUTPUT=my_system

# Then test with the captured data
make mock-test TESTDATA=my_system/capture.txt

See tools/mock-server/README.md and tools/capture/README.md for detailed documentation.

Supported Devices (tested)

Prior to the fork (should also still work post-fork):

• Enginetech EG520R-G20 (Supermicro Firmware Revision 1.76.39)
• Enginetech EG920A-G20 (Huawei iBMC 6.22)
• Lenovo ThinkSystem SR850 (BMC 2.1/2.42)
• Lenovo ThinkSystem SR650 (BMC 2.50)
• Dell PowerEdge R440, R640, R650, R6515, C6420
• GIGABYTE G292-Z20, G292-Z40, G482-Z54
• GIGABYTE R263-Z32 (AMI MegaRAC SP-X)

Since the fork:

• Nvidia B200
• Nvidia GB200/GB300 NVL72
• (really, most Nvidia server architectures/OEMs)

Why a Fork?

We decided to fork the existing exporter for several reasons:

• Specialized needs: In the GPU Cloud/AI/Superintelligence space, hardware platforms evolve so quickly and in specialized ways, requiring tailored behaviors.
• Updated dependencies: The upstream repository had more than a few outdated dependencies. We want to stay (more) up to date.
• Tests: We want to encourage testable behavior as much as possible, especially when working with bleeding-edge hardware.
• Deviation in behavior: We envision the redfish_exporter operating more similary to other Prometheus ecosystem exporters, like blackbox_exporter.

Acknowledgements

• https://github.com/stmcginnis/gofish, whose work made this exporter and its predecessors possible
• https://github.com/jenningsloy318/redfish_exporter, where our fork was forked from
• https://github.com/FlxPeters/redfish_exporter (which this particular project forks)