Documentation / DC Telemetry Bridge

DC Telemetry Bridge

OPC UA device drivers that bring IT telemetry into Ignition tags, so facilities SCADA can see what the servers see.

Version 0.1.1 for Ignition 8.3, 0.1.0 for 8.1 · early release · module id org.cognidata.dctelemetry

Overview

The module adds two device types to the gateway's OPC UA server. Each device polls one HTTP endpoint and exposes what it finds as read-only tags.

Device typeReads fromTypical tags
Redfish (Server BMC)A server's baseboard management controller over /redfish/v1 — iDRAC, iLO, XCC, OpenBMC, Supermicro.Inlet and CPU temperatures, fan speeds, PSU input watts, chassis power, health.
Prometheus / OpenMetricsAn exporter's metrics endpoint (NVIDIA DCGM-exporter, node_exporter, PDU and UPS exporters), or PromQL against a Prometheus server.Per-GPU power and temperature, or pre-aggregated values such as GPU watts per rack.

There are no addresses to type. The browse tree is discovered from the first poll, and you drag the tags you want into a tag provider like any other OPC device. Nothing is ever written back to a BMC or to Prometheus.

Requirements & downloads

  • An Ignition Gateway with the OPC UA module installed and running. It ships with a standard install.
  • Network access from the gateway host to each BMC, exporter or Prometheus server over HTTP or HTTPS.
  • The build that matches your platform line. The OPC UA device API differs between 8.1 and 8.3, so the two files are not interchangeable.

Ignition 8.1

v0.1.0 · .modl · Requires 8.1.33 or newer in the 8.1 line

Download for Ignition 8.1

Ignition 8.3

v0.1.1 · .modl · Requires 8.3.0 or newer

Tested on a live Ignition 8.3.10 gateway: install, both device types, an embedded password, tag values, and outage recovery.

Download for Ignition 8.3

The module is free. There is no trial timer and no licence to activate. The two builds carry different version numbers because 0.1.1 fixed a settings-form problem that only exists on 8.3; the drivers themselves are the same.

Install the module

  1. Open the gateway web interface and go to the Modules page — Config → System → Modules on 8.1, Platform → System → Modules on 8.3.
  2. Choose Install or Upgrade Module and select the .modl file for your platform.
  3. Accept the module certificate when prompted. It is issued by Cognidata and is self-signed, and the gateway says so. You are asked once; later upgrades install without the prompt.
  4. On 8.3, restart the gateway. The module is listed as Pending Restart after an install, an upgrade or an uninstall, and its device types do not appear until the gateway has restarted.
  5. Confirm DC Telemetry Bridge appears in the module list as active.

Then create a device: Config → OPC UA → Device Connections → Create new Device on 8.1, or Connections → Devices → Connections → Create Device Connection on 8.3. Both new drivers appear in the list.

Redfish (Server BMC) device

One device per BMC. Each poll starts at the Redfish service root and follows the links the BMC returns, so it works across vendors that name their resources differently (System.Embedded.1, 1, Self).

Settings

SettingDefaultWhat it does
Connection
BMC URLhttps://Scheme and host of the BMC, for example https://10.20.0.15. Do not add a path — /redfish/v1 is appended for you.
Username—Redfish account. A read-only role is enough; the module never writes to the BMC.
Password—Sent with HTTP Basic auth on every request. On 8.3 this is a secret field: choose None, Embedded (stored encrypted in the device's configuration) or Referenced (from a secret provider).
Polling
Poll Rate (ms)15000Delay between the end of one poll and the start of the next. BMCs are slow; 10000–30000 ms is typical.
Include SystemsonAlso read /Systems: power state, health, model, serial number and BIOS version.
Advanced
Verify TLS CertificateonTurn off for BMCs with self-signed certificates. Turning it off also skips the hostname check, for this device only.
Request Timeout (ms)10000Applies to each HTTP request. One poll makes several requests.

Tags produced

Every chassis the BMC reports gets a folder, with one sub-folder per sensor. A property the BMC does not report is simply absent.

[bmc01]
├─ Chassis/<chassis id>/
│   ├─ Health, State, PowerState
│   ├─ Temperatures/<sensor>/  ReadingCelsius, UpperThresholdCritical,
│   │                          UpperThresholdFatal, Health, State
│   ├─ Fans/<fan>/             Reading, ReadingUnits, Health, State
│   ├─ PowerControl/<name>/    PowerConsumedWatts, PowerCapacityWatts,
│   │                          AverageConsumedWatts, MinConsumedWatts,
│   │                          MaxConsumedWatts, PowerLimitWatts
│   ├─ PowerSupplies/<psu>/    PowerInputWatts, LastPowerOutputWatts,
│   │                          LineInputVoltage, PowerCapacityWatts,
│   │                          Health, State
│   └─ Voltages/<rail>/        ReadingVolts, Health, State
├─ Systems/<system id>/        PowerState, Health, State, Model,
│                              SerialNumber, BiosVersion
└─ _Diagnostics/               see Diagnostics tags

A real path looks like [bmc01]Chassis/System_Embedded_1/Temperatures/System Board Inlet Temp/ReadingCelsius. When two sensors in the same list share a name, the second gets its Redfish MemberId appended so both stay addressable.

If one chassis fails to read, the rest of the poll still completes and the failure is reported as a warning rather than faulting the device.

Prometheus / OpenMetrics device

One device per endpoint. Pick the mode by where the numbers live: scrape an exporter directly when you want one machine's raw series, or query a Prometheus server when you want values aggregated across a fleet.

Settings

SettingDefaultWhat it does
Source
ModeEXPORTEREXPORTERscrapes one exporter's metrics endpoint directly. QUERY runs the PromQL you list against a Prometheus server.
URLhttp://localhost:9400/metricsEXPORTER: the full metrics URL, for example http://gpu-node01:9400/metrics. QUERY: the Prometheus base URL, for example http://prometheus:9090.
PromQL Queries—QUERY mode only. One query per line as name = promql. See QUERY mode.
Series Selection
Include Metrics (regex)blank (keep all)EXPORTER mode. Only metric names that fully match are kept, for example DCGM_FI_DEV_(GPU_TEMP|POWER_USAGE|MEMORY_TEMP).
Exclude Metrics (regex)blank (drop none)EXPORTER mode. Metric names that fully match are dropped, for example go_.*|process_.*.
Path Labelsblank (all labels)Comma-separated label keys that become folders, in the order given, for example Hostname,gpu. Blank uses every label sorted by key, which is always unique but verbose.
Max Series5000Upper bound on values per poll. Protects the gateway from a high-cardinality exporter; anything beyond it is dropped and reported as a warning.
Poll Rate (ms)5000Delay between scrapes. Match the exporter's own collection interval — polling faster than the exporter refreshes only repeats values.
Authentication & Advanced
Username / Password—HTTP Basic auth, for endpoints behind a proxy that requires it.
Bearer Token—Sent as Authorization: Bearer …. Takes precedence over Basic auth when both are set.
Verify TLS CertificateonTurn off for self-signed endpoints. Also skips the hostname check, for this device only.
Request Timeout (ms)5000Applies to each HTTP request.

EXPORTER mode

The device fetches the metrics URL and turns each sample into a tag. The path is the metric name, then one folder per path label written as key_value. A metric with no labels is itself the value tag.

# Exporter output
DCGM_FI_DEV_POWER_USAGE{gpu="0",UUID="GPU-5a1f…",Hostname="gpu-node01"} 287.4
DCGM_FI_DEV_POWER_USAGE{gpu="1",UUID="GPU-9c2e…",Hostname="gpu-node01"} 301.2

# Path Labels = gpu
[gpu01]DCGM_FI_DEV_POWER_USAGE/gpu_0
[gpu01]DCGM_FI_DEV_POWER_USAGE/gpu_1

# Path Labels = Hostname,gpu
[gpu01]DCGM_FI_DEV_POWER_USAGE/Hostname_gpu-node01/gpu_0

Choose path labels that together identify a series. If two series collapse to the same path, only one of them can be the tag. Leaving the field blank avoids that at the cost of long paths.

Exporters publish far more than you need — a node_exporter emits hundreds of series. Set Include Metrics before the first poll: tags, once created, are not removed.

QUERY mode

The device runs each query against the server's instant-query API on every poll. Write one query per line as name = promql. The name becomes the top-level folder; only the first = separates the two, so the PromQL may contain its own. Blank lines and lines starting with # are ignored.

# GPU power per rack
rack_gpu_watts = sum by (rack) (DCGM_FI_DEV_POWER_USAGE)

# Hottest GPU in the room (no labels left, so this is a single tag)
hottest_gpu_celsius = max(DCGM_FI_DEV_GPU_TEMP)
[prom]rack_gpu_watts/rack_R01
[prom]rack_gpu_watts/rack_R02
[prom]hottest_gpu_celsius

Queries must return an instant vector or a scalar. A query that fails is reported as a warning and the others still run. The include and exclude filters do not apply in this mode — the query already decides what comes back.

Using the tags

The devices appear under the gateway's own OPC UA server. In the Designer, open the OPC Browser, expand Ignition OPC UA Server → Devices, and drag folders or individual values into a tag provider.

To create an OPC tag by hand, use the path as the OPC item path:

OPC Server     Ignition OPC UA Server
OPC Item Path  ns=1;s=[bmc01]Chassis/System_Embedded_1/Temperatures/System Board Inlet Temp/ReadingCelsius

From there they are ordinary tags: history, alarms, UDTs and bindings all work. A tag group faster than the device's poll rate does not make the values fresher — the device's own poll rate sets how often they change.

Diagnostics tags

Every device has a _Diagnostics folder. These tags stay good-quality even when the endpoint is unreachable, which makes them the right place to hang a communications alarm.

TagTypeMeaning
ConnectedBooleanTrue when the last poll succeeded. The natural tag to alarm on.
LastErrorStringWhy the last poll failed; empty when it succeeded.
LastPollDurationMsDoubleHow long the last poll took. Watch this on slow BMCs.
SeriesCountDoubleNumber of values returned by the last poll.
WarningCountDoubleNon-fatal problems in the last poll, such as a skipped chassis or dropped series.
LastWarningStringThe first of those warnings; empty when there were none.

The device status shown on the gateway's device list carries the same information: Connected, Connected, N warning(s), or Faulted: … with the error.

Quality, types & names

SituationValueQuality
Value returned by the pollUpdatedGood
Poll succeeded, this value missing from itLast value keptUncertain_LastUsableValue
Poll failedLast value keptBad_CommunicationError
  • Tags are never removed. A sensor that drops out keeps its tag and last value, flagged as above, until the device is restarted — editing and saving the device restarts it and rebuilds the tree from the next poll.
  • Data types are fixed when a tag is first seen — Double for numbers, Boolean, or String — and later values are converted to that type, so a tag never changes type underneath a binding.
  • Names are made tag-safe. Letters, digits, spaces and _ - ( ) are kept; every other character becomes an underscore. That is why System.Embedded.1 shows up as System_Embedded_1.
  • Polls never overlap. The poll rate is a delay after the previous poll finishes, so a slow endpoint slows its own device down rather than stacking requests. The minimum is 250 ms.
  • The gateway log stays quiet. A device logs when its error state changes, not on every failed poll. Loggers are named DcTelemetry.… followed by the device name.

Troubleshooting

Start with the device status and _Diagnostics/LastError; between them they say why a poll failed. For more detail, filter the gateway logs for DcTelemetry.

The module will not install on 8.1.
The 8.1 build requires Ignition 8.1.33 and the OPC UA module. Check the gateway version, and make sure you picked the 8.1 file — the 8.3 file will not load on 8.1, or the reverse.
On 8.3 the module is listed as Pending Restart and the drivers are missing from the device list.
Restart the gateway. On 8.3 the module is only loaded, upgraded or removed at startup.
The device is created but has no tags to browse.
The tree is built from the first successful poll. Wait one poll interval and refresh the browser. If the status reads Faulted, the reason is in the status text and in _Diagnostics/LastError.
Status is Faulted: HttpStatusException: HTTP 401 …
The endpoint rejected the credentials. Check the username and password (or bearer token), and that the account is allowed to use Redfish.
Status mentions an SSL or certificate error.
The endpoint has a certificate the gateway does not trust — the normal state of a BMC out of the box. Turn off Verify TLS Certificate for that device, or install a trusted certificate on the endpoint.
Status mentions a timeout.
The gateway cannot reach the endpoint, or it answers too slowly. Confirm the URL and firewall path from the gateway host itself, then raise Request Timeout for slow BMCs.
LastWarning reads N series dropped by the max-series limit
The exporter returned more series than Max Series allows. Narrow the scrape with Include Metrics, or raise the limit if the gateway can carry it.
LastWarning reads query 'x': unsupported resultType 'matrix'
The query returns a range vector. Use an expression that produces an instant vector or a scalar — wrap it in a function such as rate(…[5m]) or avg_over_time(…[5m]).
Some tags show Uncertain_LastUsableValue.
The poll succeeded but those values were missing from it — a sensor the BMC stopped reporting, or a series that disappeared from the exporter. The last value is kept.
Tag paths are long and full of labels.
Path Labels is blank, so every label becomes a folder. List only the labels that identify a series, for example gpu or Hostname,gpu.

Limitations

  • Read-only. The module cannot power-cycle a server, set a power cap, or change anything on the endpoint.
  • Redfish uses HTTP Basic auth on each request; session-token login is not implemented yet.
  • Redfish reads the Thermal and Power resources. The newer ThermalSubsystem, PowerSubsystem and Sensors model is not read yet, so a BMC that has dropped the older resources will show few tags.
  • QUERY mode supports instant vectors and scalars, not range queries.
  • This is an early release. Both builds have been run on live gateways, but against simulated Redfish and Prometheus endpoints — not yet against physical BMCs or a real DCGM-exporter. Try it on a test gateway first.

← All documentation·See the live demo