A very common requirement of background task systems is to govern concurrent access based on aspects of the data, like "only run one task per unique customer_id at the same time" or "only run n tasks per database name at a time". This should be separate from the key of a task and should be expressed in more granular dependencies/parameter annotations.