When I was dealing with time series, the most common query was an 'as of' join. To the best of my knowledge, no common relational database supports this efficiently (which isn't surprising, as it is somewhat incompatible with the relational model in general)
That's a good counterpoint. For most purposes a `last ... over partition by ...` query will do the job, but you are right that it will be not nearly as efficient as the native operation in a timeseries database.
With that in mind, I don't come to a different conclusion though. Just about every database I've worked with has had some timeseries data in it, but separating data out into multiple systems/models is usually something I would avoid until there is no other option.
