XRDU Exporter Reference - SambaNova Documentation

The XRDU exporter publishes fan and temperature metrics from XRDU baseboard management controllers (BMCs). One exporter instance serves any number of XRDUs. Prometheus scrapes /probe?target=<address> once per BMC.

Property	Value
Port	9821 by default. Set `xrdu-exporter.port`.
Runs as	A Deployment, deployed by the `sambanova-hardware-monitoring` chart
Device protocol	The BMC’s HTTPS API, port 443, self-signed certificate
Authentication	Required. An XRDU BMC has no guest read.
Network	The cluster nodes need a route to the XRDU BMCs on the management network
Image version	0.1.1 or later. Version 0.1.0 rejects BMC certificates with a negative serial number.

Endpoints

Endpoint	Returns
`/probe?target=<address>`	The metrics for one BMC
`/metrics`	The exporter’s own process metrics. Contains no BMC data.
`/healthz`	`ok` while the exporter is running

/probe takes no optional parameters. A healthy BMC publishes about 97 series: 14 fan readings, 81 temperatures, xrdu_up, and xrdu_scrape_duration_seconds.

Metrics

Metric	Type	Labels	Description
`xrdu_fan_rpm`	gauge	`sensor`	Fan tachometer reading in revolutions per minute
`xrdu_temperature_celsius`	gauge	`sensor`	Temperature in degrees Celsius
`xrdu_up`	gauge	none	`1` when the BMC answered, `0` when it did not
`xrdu_scrape_duration_seconds`	gauge	none	Time taken to read the BMC

This is the complete supported set. A sensor the BMC reports as not working, absent, or implausible publishes no sample. No metric marks the missing reading, so the series simply has a gap.

Labels

Label	Values	Description
`xrdu`	The target’s `name`, or its address	Identifies the BMC. Added by Prometheus, not the exporter.
`sensor`	`fm0_fan0`, `inlet`, `RDU0_D0_CORETEMP`	The BMC’s own sensor name, unchanged
`rack`, `node`, `xrdu_index`	Whatever you declare	Declared in the target’s `labels`

The xrdu label means different things in different exporters. On xrdu_* series it holds a BMC hostname, such as <rack-id>-h1xr0. On sambanova_* series it holds the host’s own index, XRDU_0 to XRDU_7. A join on xrdu between the two matches nothing and reports no error.

Join XRDU thermals with host metrics

To join BMC metrics with sambanova_* series, declare the host’s XRDU index on each target alongside node:

xrdu-exporter:
  targets:
    - address: <xrdu-0-ip>
      name: <xrdu-0-hostname>
      labels:
        node: <node-name>
        xrdu_index: XRDU_0

The XRDU number in snctl rack show (Node/<n>/xrdu/<m>) matches the host’s index, XRDU_<m>. Join on both node and xrdu_index. The index is not unique on its own, because XRDU_2 exists on every node.

Behavior

Behavior	Detail
An unreachable BMC returns HTTP 200 with `xrdu_up 0`	A rejected credential and an unparseable certificate look identical from the metric alone. The exporter log distinguishes them. Alert on `xrdu_up == 0`.
The exporter reports readings, not verdicts	Nothing here tells you a sensor is out of range.
Scrape no faster than every 30 seconds	A BMC takes 2.3 to 2.7 seconds to answer. The exporter enforces a minimum scrape timeout of 5 seconds and refuses to start below it.
Extra replicas multiply BMC load	Replicas do not share a cache. Keep `replicaCount` at `1`.

Data the exporter does not publish

The exporter reads the BMC’s OpenBMC sensor tree over HTTPS and maps part of it onto the metrics above. The BMC returns more in the same response. If you need any of the following, it is not available through this exporter.

Not published	Detail
Voltage, current, power, and utilization sensors	Returned in the same response as fans and temperatures, but not published.
Threshold limits	The BMC declares warning and critical limits, high and low, per sensor.
Breach alarms	The BMC sets a flag when a reading enters a warning or critical band, including transiently.
Commanded fan duty	The tachometer reading in `xrdu_fan_rpm` is what describes the fan.
A sensor’s operating range	The BMC reports a minimum and maximum alongside each value.
Optical transceiver telemetry	QSFP case and die temperature, connection state, and firmware version.
BMC event logs and Redfish telemetry reports	Fault reports are a separate BMC interface. This exporter reads sensors only.
Power state and inventory	Chassis power state, serial numbers, and FRU data. Use SambaRack Manager for those.

To read a sensor straight from the device, bypassing the exporter:

read -rs -p "XRDU BMC password: " XRDU_PASSWORD; echo

curl -sk -u root:"$XRDU_PASSWORD" \
  "https://<bmc-address>/xyz/openbmc_project/sensors/enumerate" | jq

unset XRDU_PASSWORD

Both -k and -u are required. The BMC presents a self-signed certificate and requires authentication.

Configuration

Helm values

Key	Default	Description
`xrdu-exporter.enabled`	`true`	Deploy the exporter
`xrdu-exporter.probe.enabled`	`true`	Create the `Probe` that scrapes the BMCs. Without it, nothing scrapes them.
`xrdu-exporter.probe.interval`	`30s`	Scrape interval. Do not shorten it.
`xrdu-exporter.probe.scrapeTimeout`	`15s`	Prometheus scrape timeout. Must be longer than `xrdu-exporter.scrapeTimeout`.
`xrdu-exporter.probe.labels`	`{}`	Must match your Prometheus `probeSelector`
`xrdu-exporter.targets`	`[]`	Required when enabled. See Targets.
`xrdu-exporter.auth.existingSecret`	`""`	Secret with `XRDU_USERNAME` and `XRDU_PASSWORD`. One credential setting is required.
`xrdu-exporter.auth.username` / `.password`	`""`	Used only when `auth.existingSecret` is empty. Set `auth.password` (and optionally `auth.username`) and the chart creates `xrdu-exporter-credentials`. The username defaults to `root`. Prefer `auth.existingSecret`: a password set here is stored in the Helm release.
`xrdu-exporter.auth.usernameKey` / `.passwordKey`	`XRDU_USERNAME` / `XRDU_PASSWORD`	Key names inside the Secret
`xrdu-exporter.cacheTTL`	`15s`	How long one BMC’s read is reused
`xrdu-exporter.scrapeTimeout`	`10s`	Exporter-side timeout. Minimum `5s`.
`xrdu-exporter.logLevel`	`info`	`debug`, `info`, `warn`, or `error`
`xrdu-exporter.serviceMonitor.enabled`	`false`	Scrape the exporter’s own `/metrics`
`xrdu-exporter.serviceMonitor.interval`	`60s`	Applies to the exporter’s own endpoint only. `probe.interval` governs how often BMCs are read.
`xrdu-exporter.serviceMonitor.scrapeTimeout`	`10s`	Must not exceed the interval
`xrdu-exporter.replicaCount`	`1`	Number of exporter pods
`xrdu-exporter.port`	`9821`	Port the exporter listens on
`xrdu-exporter.resources`	50m / 64Mi, limit 256Mi	Pod requests and limits
`xrdu-exporter.nodeSelector` / `.tolerations` / `.affinity`	`{}` / `[]` / `{}`	Pin the pod to nodes that can reach the management network
`xrdu-exporter.podSecurityContext` / `.securityContext`	non-root, read-only rootfs, drop ALL	Hardened by default

Targets

xrdu-exporter:
  enabled: true
  probe:
    enabled: true
  auth:
    existingSecret: xrdu-creds
  targets:
    - address: <xrdu-0-ip>
      name: <xrdu-0-hostname>
      labels:
        rack: "<rack-id>"
    - <xrdu-1-ip>                   # no labels: excluded from per-rack panels and host joins
    - address: <xrdu-2-ip>

Field	Description
`address`	Required. Goes into `?target=` as written.
`name`	Replaces the address as the `xrdu` label
`labels`	Each key becomes a label on every series from this BMC

The username defaults to root, so you usually set only the password. Get it with snctl secret get --device-type xrdu --show-value.

Get started

Models

Features

Build

Resources

XRDU exporter

Endpoints

Metrics

Labels

Join XRDU thermals with host metrics

Behavior

Data the exporter does not publish

Configuration

Helm values

Targets

Next steps

​Endpoints

​Metrics

​Labels

​Join XRDU thermals with host metrics

​Behavior

​Data the exporter does not publish

​Configuration

​Helm values

​Targets

​Next steps

Endpoints

Metrics

Labels

Join XRDU thermals with host metrics

Behavior

Data the exporter does not publish

Configuration

Helm values

Targets

Next steps