Question

I have a table that contains some user data:

user_id | guest_id | time_seen | action_performed | longitude | latitude
-------------------------------------------------------------------------  
 123     | NULL     | Jan 10    | search           | -127      | 35  
 152     | NULL     | Dec 10    | login            | -128      | 34   
 172     | NULL     | Dec 15    | search           | -125      | 35   
 123     | NULL     | Jan 10    | login            | -127      | 35   
 NULL    | GUEST1   | Jan 10    | search           | -127      | 35   
 NULL    | GUEST1   | Dec 10    | search           | -127      | 35   
 NULL    | GUEST2   | Jan 10    | browse           | -127      | 35   
 NULL    | GUEST3   | Dec 10    | browse           | -127      | 35   

I need to get a list of the unique user ids and guest ids for a given time span. The same row never has a valid user ID AND a valid guest ID at the same time. The query I’m using currently is:

SELECT * 
    FROM stats 
    WHERE time_seen >= "2011-12-1 00:00:00" 
      AND time_seen < "2012-1-1 00:00:00" 
    GROUP BY guest_id 
UNION 
SELECT * 
    FROM stats 
    WHERE time_seen >= "2011-12-1 00:00:00" 
      AND time_seen < "2012-1-1 00:00:00" 
    GROUP BY user_id;

So I’d expect to retrieve:

user_id | guest_id | time_seen | action_performed | longitude | latitude
-------------------------------------------------------------------------  
 152     | NULL     | Dec 10    | login            | -128      | 34   
 172     | NULL     | Dec 15    | search           | -125      | 35 
 NULL    | GUEST1   | Dec 10    | search           | -127      | 35
 NULL    | GUEST3   | Dec 10    | browse           | -127      | 35      

The real table contains about 11 million entries as of this time and is growing every day, so obviously I’m interested in making the query as efficient as possible. This query seems a little suboptimal (unless there’s internal optimization) since I perform the same query each time, just group it by different things afterwards.

Is there any way to improve my query, or is this the best I can do?

In response to questions below:

The dates do have a timestamp value, I just wanted to simplify for the purposes of the post.

There is no overlap between user_id values and guest_id values.

Answer 1

If there is no overlap between the user_id values and the guest_id values then you could GROUP BY the COALESCE of those two columns and not use the UNION