It's geospatial data from OpenStreetMap stored in a schema optimized for PostGIS extension (produced by osm2pgsql).
BTW: Having said (to Martijn) that using Postgres is probably more efficient, than programming an in-memory database in a decent language: OpenStreetMap has a very, very large Node table which is heavily used by other tables (like ways) - and becomes rather slow in Postgres. Since it's of fixed length I'm looking at file_fixed_length_record_fdw extension [1][2] (which is in-memory) to get the best of both worlds.
> How can Postgres be used and configured as an In-Memory Database? >
we've put the data directory on our buildserver directly on a ramdisk (e.g. /dev/shm) to improve build times.
Obviously you then don't care too much about durability here, so one can switch off all related settings (as has already been pointed out). The only thing to do on a server reboot would be to re-create a fresh data directory on the ramdisk.
So if you're able to start from scratch relatively cheap (i.e. on a server reboot), don't care about durability/crash safety at all and your database fits into ram that solution is easy to handle.
I've also tried having only a separate tablespace on ramdisk but abandoned the idea because postgres seemed too surprised to see the tablespace empty after a reboot (all tables gone).
Overall the above solution works and improves our build times but I think there are better ways to have in-memory/application caches than using a postgres.