serviceruntime metric errors

[bytheway]

I think I'm running into a similar bug as #9, getting a 500 error for metrics from serviceruntime.googleapis.com:

4 error(s) occurred:

* collected metric stackdriver_consumed_api_serviceruntime_googleapis_com_api_request_count label:<name:"grpc_status_code" value:"0" > label:<name:"location" value:"us-central1" > label:<name:"method" value:"google.api.servicecontrol.v1.ServiceController.Check" > label:<name:"project_id" value:"______" > label:<name:"protocol" value:"" > label:<name:"response_code" value:"200" > label:<name:"response_code_class" value:"2xx" > label:<name:"service" value:"servicecontrol.googleapis.com" > label:<name:"unit" value:"1" > label:<name:"version" value:"v1" > gauge:<value:2 >  has label dimensions inconsistent with previously collected metrics in the same metric family...

Thanks for the great project!

[bytheway]

Investigating further, it looks like google includes the credential_id label on most metrics, but it is missing on others.

[bytheway]

(Sorry for the comment spam), This would also be helped by getting the go client fixed, as reported: prometheus/client_golang#355

[frodenas]

Interesting, according to the docs, the consumed_api resource type should return the credential_id label.

Inconsistent label dimensions have been a problem for proxy exporters like this. An easy workaround will be to always add that label when querying for serviceruntime metrics, but 1) I don't like applying specific logic for particular metrics 2) it will only solve the problem for this particular metric, and we cannot guarantee other metrics will face the same problem in the future. Adding a generic helper for missing labels seems the way to go, let's keep an eye on prometheus/client_golang#355.

[bytheway]

I think what is happening is that the consumed_api and the produced_api metrics are all mixed together in the same serviceruntime space. Here is what the produced apis are returning: https://cloud.google.com/monitoring/api/resources#tag_api