I have some experience with mass, remote replication from MySQL, MSSQL and PostgreSQL source databases into one PostgreSQL target database. According to advices above, I would prefer to tune the mass-insertion logic rather than threading or forking. If you can design upserts specifically for every case, following ideas could help you:
- try to recognize what should be inserted and what updated (IDs, columns storing last change time, ...) generally, not row-by-row
- use previously advised ways of insert or update
- use stored procedure upsert_my_table() on the target database
- try to insert more than one row in one dml command
- try to take advantage of existing indexes (on both source and target db) - sometimes upserting from properly sorted source table to (properly sorted) target table makes sense
