[feature request] Add Support for Exposing userLabels from GCP Resources as Prometheus Labels

[elenaknofs]

We would like to request support for exposing userLabels from Google Cloud resources (e.g., Compute Engine, Cloud SQL) as Prometheus metric labels in the Google Stackdriver Prometheus Exporter.

Currently, while the exporter successfully maps Stackdriver metrics and their associated resource labels to Prometheus, custom userLabels defined on resources in Google Cloud are not included. These labels are essential for enabling dynamic filtering, categorization, and fine-grained alerting in Prometheus.

Adding support for userLabels would improve the usability of the exporter by aligning it with real-world use cases where custom metadata is used to manage metrics across complex environments.

Use Case

In environments with a large number of Google Cloud resources, userLabels are commonly used to define metadata such as:

• Maintenance states (monitoring=true/false)
• Environment information (env=prod, env=stage)
• Resource ownership (team=devops, team=dba)
  Including userLabels in the exported metrics would allow users to:

1. Dynamically filter metrics based on userLabels in Prometheus queries.
2. Set up targeted alerts for resources with specific labels (e.g., suppress alerts for resources with monitoring=false).
3. Enhance observability by grouping metrics using custom tags.

Implementation Details

• Extend the exporter’s logic to fetch userLabels from the resource.labels field for each monitored resource.
• Map the userLabels to Prometheus metric labels during metric export.
• Allow users to optionally enable or disable userLabels in the exported metrics via a command-line flag (e.g., --include-user-labels).

More info about GCP resource labeling can be found here

[emrahbecer]

This would be a great feature. It is common to use custom labels on GCP.

[Daniel-Vaz]

As far as I understand to have the metadata block populated with user_labels you need to force a reduction that “groups by” one or more metadata user_label keys.
In the API the metadata is only returned on a time series when you aggregate the data in a way that explicitly groups by the desired metadata labels.

So I don't see how just creating a generic flag like --include-user-labels will work.

Users will need to explicitly mention the User Labels they desire to look for, and pass the remaining needed aggregation parameters like AggregationPerSeriesAligner \ AggregationCrossSeriesReducer and AggregationAlignmentPeriod to the timeSeriesListCall.

To support aggregation, I'm doing some experiments, but still without any major success. I'm following the initial idea mentioned by @SuperQ in here, where the extra aggregation parameters are passed trough each call of the --monitoring.metrics-prefixes flag.