If you only got a two-digit number of rows per PPI it should be safe to to switch to
PRIMARY INDEX (customer_id)
PARTITION BY (
RANGE_N(dt BETWEEN '2000-04-01'
AND '2016-12-31' -- go as far into the future as needed to avoid ADD partition
EACH INTERVAL '1' MONTH)
If the source data is known to be unique a MULTISET table will speed up the Insert/Select.
Doing it this way will still allow me to filter quickly, say on country, even without a secondary index? How does that work?
It's not unique, as a single customer may have multiple rows.