in issue2551142, I "fixed" the problem by committing after every row in 
the csv import was created. This has performance implications.

Implementing this using subtransactions in posgresql:

  https://postgres-
py.readthedocs.io/en/latest/#postgres.Postgres.get_cursor

may allow us to solve the initial issue without having to
perform a per-row commit.

I am not sure how this interacts with the python DBI standard and how 
roundup interacts with the database, but it is worth a look.