Nicolas Grilly schrieb am 30.08.2016 um 13:12:
> We rely on clustered indexes to preserve data locality for each
> tenant. Primary keys start with the tenant ID. This way, rows
> belonging to the same tenant are stored next to each other. Because
> all requests hit only one tenant, this is a great performance
> improvement.
>
What about partitioning by tenant? With a local index on each partition.
Partitioning is currently a bit limited in Postgres (e.g. you can't have incoming foreign keys) but this would fit your requirements pretty much as I understand them.