A minor thing, but a little annoying: in PostgresReplicationConnection.initPublication(), since rs (ResultSet) of "SELECT COUNT(1) FROM pg_publication WHERE..." query is not destroyed manually in a finally block, I think AWS thinks that it's still executing for a long time. I also see it in pg_stat_activity as an "active" query (although it's obvious that this query finished long time ago).

I think it's a simple and a safe fix, just call rs.close() in a finally block in this function.