Looking at the system dashboards, it seems most visualisations do not prefilter on the data_stream.dataset value but instead query all of the metrics-* indices.

This is an issue as it is much less performant!