We use the same setup, though we use PGBouncer so after switching primary we just force reconnect all clients from PGBouncer instead.

The clients will have to retry on-going transactions, but that's a basic fault tolerant requirement anyway.