While running this in production I noticed a number of ghost processes with all their clients dead before they released the connection, so let's track them to log it and remove them from clients