Add chip label to DRM metrics

[Deezzir]

Issue

Add chip label to the node_drm_card_info metric.

Solution

Extend the DRM collector to have the chip label from the ClassDRMCardAMDGPUStats.
The change was also proposed in the procfs repo, and the PR was merged.

Context

The DRM metrics are impossible to relate to the hwmon metrics, which export other helpful information about AMD GPUs. The extension will allow us to relate metrics from both hwmon and drm and, in turn, create a better GPU Dashboard for AMD with proper filtering and labelling.

Example metrics:

# DRM
node_drm_card_info{card="card0",chip="0000:9d:00_0_0000:9e:00_0",memory_vendor="unknown",power_performance_level="auto",unique_id="",vendor="amd"} 1

# HWMON
node_hwmon_chip_names{chip="0000:9d:00_0_0000:9e:00_0",chip_name="amdgpu"} 1
node_hwmon_fan_enable{chip="0000:9d:00_0_0000:9e:00_0",sensor="fan1"} 0
node_hwmon_fan_max_rpm{chip="0000:9d:00_0_0000:9e:00_0",sensor="fan1"} 6900
node_hwmon_fan_min_rpm{chip="0000:9d:00_0_0000:9e:00_0",sensor="fan1"} 1800
node_hwmon_fan_rpm{chip="0000:9d:00_0_0000:9e:00_0",sensor="fan1"} 2035
node_hwmon_fan_target_rpm{chip="0000:9d:00_0_0000:9e:00_0",sensor="fan1"} 2035
node_hwmon_freq_freq_mhz{chip="0000:9d:00_0_0000:9e:00_0",sensor="mclk"} 1500
node_hwmon_freq_freq_mhz{chip="0000:9d:00_0_0000:9e:00_0",sensor="sclk"} 214
node_hwmon_in_volts{chip="0000:9d:00_0_0000:9e:00_0",sensor="in0"} 0.85
node_hwmon_power_average_watt{chip="0000:9d:00_0_0000:9e:00_0",sensor="power1"} 6.075
node_hwmon_power_cap_default_watt{chip="0000:9d:00_0_0000:9e:00_0",sensor="power1"} 35
node_hwmon_power_cap_max_watt{chip="0000:9d:00_0_0000:9e:00_0",sensor="power1"} 35
node_hwmon_power_cap_min_watt{chip="0000:9d:00_0_0000:9e:00_0",sensor="power1"} 0
node_hwmon_power_cap_watt{chip="0000:9d:00_0_0000:9e:00_0",sensor="power1"} 35
node_hwmon_pwm{chip="0000:9d:00_0_0000:9e:00_0",sensor="pwm1"} 49
node_hwmon_pwm_enable{chip="0000:9d:00_0_0000:9e:00_0",sensor="pwm1"} 2
node_hwmon_pwm_max{chip="0000:9d:00_0_0000:9e:00_0",sensor="pwm1"} 255
node_hwmon_pwm_min{chip="0000:9d:00_0_0000:9e:00_0",sensor="pwm1"} 0
node_hwmon_sensor_label{chip="0000:9d:00_0_0000:9e:00_0",label="edge",sensor="temp1"} 1
node_hwmon_sensor_label{chip="0000:9d:00_0_0000:9e:00_0",label="mclk",sensor="freq2"} 1
node_hwmon_sensor_label{chip="0000:9d:00_0_0000:9e:00_0",label="sclk",sensor="freq1"} 1
node_hwmon_sensor_label{chip="0000:9d:00_0_0000:9e:00_0",label="slowPPT",sensor="power1"} 1
node_hwmon_sensor_label{chip="0000:9d:00_0_0000:9e:00_0",label="vddgfx",sensor="in0"} 1
node_hwmon_temp_celsius{chip="0000:9d:00_0_0000:9e:00_0",sensor="temp1"} 42
node_hwmon_temp_crit_celsius{chip="0000:9d:00_0_0000:9e:00_0",sensor="temp1"} 97
node_hwmon_temp_crit_hyst_celsius{chip="0000:9d:00_0_0000:9e:00_0",sensor="temp1"} -273.15000000000003

[Deezzir]

@SuperQ and @discordianfish PTAL

[SuperQ]

This needs a rebase.

[Deezzir]

@SuperQ done

[Deezzir]

@SuperQ PTAL

[rtreffer]

I think this can be simplified by using just one metric and exporting an empty string if missing (which is a common pattern).

Other than that: having common join keys would be helpful.

[SuperQ]

I agree with @rtreffer, let's keep this simple and put chip on the existing info metric.

[Deezzir]

@SuperQ, @rtreffer I've simplified the logic, thanks

[Deezzir]

@SuperQ Can you reapprove plz, had to rebase

[rtreffer]

@SuperQ I think there is one gap/risk in the cleanup/naming diverging between hwmon and drm. Aligning that would probably be significant work and I am not 100% convinced it will work easily or if it is relevant in the context of GPUs.

I think this is cool overall.

[rtreffer]

This should relate to https://github.com/prometheus/node_exporter/blob/master/collector/hwmon_linux.go#L517

However both parts come from different sources and will diverge.
This is annoying after #3646

I am not sure this is something that could actually harm us, but calling it out as an issue overall. Making both implementations align is probably not worth the effort.

[Deezzir]

@SuperQ Can you approve the workflows plz?