Support for expanded metric scopes in GCP Stackdriver scaler

[marcos07-uy]

I'm working with KEDA in a GCP environment and ran into a limitation with the current GCP Stackdriver scaler regarding federated metrics and expanded metric scopes.

We want to configure a ScaledObject that reads a metric from Project A, but that metric is federated — it's originally collected in Project B, and made visible in Project A through GCP's Monitoring Workspace metric scoping.

Although the metric is visible in Project A (and queryable through the Monitoring API or GCP console), the KEDA GCP Stackdriver scaler currently fails to read it. This is because the scaler constructs the query assuming that the metric is local to the projectId defined in the metadata, and it hardcodes that project into the filter (e.g., via resource.labels.project_id = "project-a"). code

keda/pkg/scalers/gcp/gcp_stackdriver_client.go

Line 250 in e25ab90

filter += ` AND resource.labels.project_id="` + pid + `"`

This causes the scaler to return no results, since the actual source project of the metric is different.

Questions:

• Is there already a supported way to handle this use case in KEDA?
• Would the maintainers be open to a PR introducing scopingProjectId as an optional enhancement for this scaler?

Thanks!
--Marcos

[JorTurFer]

Hello
We have just deprecated the GCP scalers based on MQL (More info here)

We propose to use prometheus scaler and I think that using the prometheus scaler you can achieve it easier. Prometheus scaler users the monitoring API from GCP and it's NOT prometheus for GKE (that's another service), you don't need to enable any extra service or so to use prometheus scaler as replacement