CODEWITHSUNDEEP
sqlsql-serversubquerysql-viewcorrelated-subquery
Wes
Wes

Reputation: 301

Slow SQL view using several subqueries

There is probably a much better way to create these views. I have limited SQL experience so this is the way I designed it, I am hoping some of you SQL gurus can point me in a more efficient direction.

I essentially have 3 tables (sometimes 4) in my view, here is the essential structure:

Table USER
USER_ID | EMAIL | PASSWORD | CREATED_DATE

(Indexes: USER_ID)

Table USER_META
ID | USER_ID | NAME | VALUE

(Indexes: ID,USER_ID,NAME)

Table USER_SCORES
ID | USER_ID | GAME_ID | SCORE | CREATED_DATE

(Indexes: ID,USER_ID)

All the tables use the first ID column as an auto-increment primary key.

The second table "USER_META" is where I keep all the contact info and other misc. Primarily it is first_name,last_name, street,city, etc. - Depending on the user this could be 4 items or 140, which is why I use this table instead of having 150 columns in my USER table.

For reports, searching and editing I need about 20 values from USER_META, so I have views that look like this:

View V_USR_META
select USER_ID,EMAIL,
(select VALUE from USER_META 
   where NAME = 'FIRST_NAME' and USER_ID = u.USER_ID) as first_name,
(select VALUE from USER_META 
   where NAME = 'LAST_NAME'  and USER_ID = u.USER_ID) as last_name,
(select VALUE from USER_META 
   where NAME = 'CITY'  and USER_ID = u.USER_ID) as city,
(select VALUE from USER_META 
   where NAME = 'STATE'  and USER_ID = u.USER_ID) as state,
(select VALUE from USER_META 
   where NAME = 'ZIP'  and USER_ID = u.USER_ID) as zip,
    /* 10 more selects for different meta values here */
(select max(SCORE) from USER_SCORES 
   where USER_ID = u.USER_ID) as high_score,
(select top (1) CREATED_DATE from USER_SCORES 
   where USER_ID = u.USER_ID 
   order by id desc) as last_game
from USER u

This get's pretty slow, and there are actually many more sub queries, this is just to illustrate the query. I also have to query a few other tables to get misc. info about the user.

I use the view when searching for a user, searches use name or userid or email or score, etc. I also use it to populate the user information screen when I present all the data in one place.

So - Is there a better way to write the view?

Upvotes: 1

Views: 939

Answers (2)

adrianm
adrianm

Reputation: 14726

WITH Meta AS (
    SELECT USER_ID
          ,FIRST_NAME
          ,LAST_NAME
          ,CITY
          ,STATE
          ,ZIP
    FROM USER_META
    PIVOT (
        MAX(VALUE) FOR NAME IN (FIRST_NAME, LAST_NAME, CITY, STATE, ZIP)
    ) AS p
)
,MaxScores AS (
    SELECT USER_ID
          ,MAX(SCORE) AS Score
    FROM USER_SCORES
    GROUP BY USER_ID
)
,LastGames AS (
    SELECT USER_ID
          ,MAX(CREATED_DATE) AS GameDate
    FROM USER_SCORES
    GROUP BY USER_ID
)
SELECT USER.USER_ID
      ,USER.EMAIL
      ,Meta.FIRST_NAME
      ,Meta.LAST_NAME
      ,Meta.CITY
      ,Meta.STATE
      ,Meta.ZIP
      ,MaxScores.Score
      ,LastGames.GameDate
FROM USER
     INNER JOIN Meta
         ON USER.USER_ID = Meta.USER_ID
     LEFT JOIN MaxScores
         ON USER.USER_ID = MaxScores.USER_ID
     LEFT JOIN LastGames
         ON USER.USER_ID = LastGames.USER_ID

Upvotes: 0

sgeddes
sgeddes

Reputation: 62841

An alternative to all of those correlated subqueries would be to use max with case:

select u.USER_ID,
       u.EMAIL,
       max(case when um.name = 'FIRST_NAME' then um.value end) first_name,
       max(case when um.name = 'LAST_NAME' then um.value end) last_name
       ...
from USER u
       left join USER_META um 
                on u.user_id = um.user_id
group by u.user_id, u.email

Then you could add the user_scores results:

select u.USER_ID,
       u.EMAIL,
       max(case when um.name = 'FIRST_NAME' then um.value end) first_name,
       max(case when um.name = 'LAST_NAME' then um.value end) last_name
       ...,
       max(us.score) maxscore,
       max(us.created_date) maxcreateddate
from USER u
       left join USER_META um 
                on u.user_id = um.user_id
       left join USER_SCORES us
                on u.user_id = us.user_id
group by u.user_id, u.email

Upvotes: 1

Related Questions