With the current implementation, CloudKitty stores all the dataframes on the same "bucket" called stream in Loki's vocabulary. This stream is identified by {service="cloudkitty"}.

This means that every query that retrieves dataframes from Loki has to go through all the dataframes for the period. Since CloudKitty always adds the project_id from which the user is retrieving dataframes, it makes a lot of sense to create a different stream per project_id. That will have many times less dataframes to search through and will make the storage pretty escalable.

The new streams will be identified by {service="cloudkitty", project_id="abcdef12345"}.