When writing custom metrics you have resource labels and metric labels.

Resource labels are labels that are required by the resource. So if I choose the generic_task resource type and exclude the task_id resource label I get an error.

Metric labels seem to be arbitrary additional labels?

I can aggregate metrics by both resource and metric labels.

What is the point of having both? Is it merely to enforce consistency? So I can say generic_task and know for sure all metrics of that type have specific labels? And then if I still need more labels I use metric labels?

Is it a best practice to choose the best resource type and avoid metric labels were possible (for consistency)?

It seems to me like I should just always use the global type because it requires the least amount of labels which means I’m not spending money on dimensions I don’t need.

Is there a cost difference or other operational difference between resource and metric labels?

I’m having trouble finding clear best practices in the google docs.