optimization from counting group by time in kafka stream

Here is my simple requirement:Counting log in my website every day(daily). if just group by date in login time,then millions of logins will routes to a single partition,there is no parallel computing.So is anybody handle this similar problems in actual?