postgres 11: remaining connection slots are reserved for non-replication superuser connections

I've run into the following problem. From time to time, my app receives remaining connection slots are reserved for non-replication superuser connections from postgres. It lasts for quite noticeable periods of time, from several seconds to couple of minutes. I've checked pg_stat_activity at this very period, and it showed about five application backend connections, several pgAdmin connections, and some postgres-specific processes like replication, autovaccuum, wal-writing, etc -- about 20 processes in total, nothing peculiar. Then I checked SHOW max_connections; -- it was still 200. I've no pooling, nothing except just postgres and a backend application. How come?