Optimizing nested query

Question

I need help optimizing this query below. I have table pt_votes with 30+k records which holds each vote(-1 or 1) for photo and I want to select all photos and their vote sum so I have query like this below but it takes about 9 seconds to execute. How I can optimize it?

 SELECT *, ifnull((SELECT SUM(vote) FROM pt_votes vo WHERE vo.pID = ph.pID),0) points,
                        (SELECT CONCAT(name, " ", surname) FROM pt_users us WHERE us.uID = ph.uID) name_surname
                    FROM pt_photos ph
                    WHERE 1

Answer 1

The biggest efficiency killer here is the correlated subqueries:

(SELECT CONCAT(name, " ", surname)
  FROM pt_users us
  WHERE us.uID = ph.uID) name_surname

... and:

ifnull((SELECT SUM(vote)
  FROM pt_votes vo
  WHERE vo.pID = ph.pID),0) points,

Each of these will run once for every row that makes it past the WHERE clause.

To eliminate the correlated subqueries you need to join to the pt_votes and pt_users tables. Also, because you're summing votes you'll need to GROUP BY, which means you really need to get rid of that SELECT * as was already recommended in the comments.

The query will look something like this. When you determine which pt_photos columns you need be sure to add them to the GROUP BY list:

SELECT
  pt_photos.pID,
  pt_photos.uID,
  pt_photos.this,
  pt_photos.that,
  CONCAT(pt_users.name, ' ', pt_users.surname) AS name_surname,
  IFNULL(SUM(pt_votes.vote), 0) AS points
FROM pt_photos
JOIN pt_users ON pt_photos.uID = pt_users.uID
LEFT JOIN pt_votes ON pt_photos.pID = pt_votes.pID
WHERE 1
GROUP BY
  pt_photos.pID,
  pt_photos.uID,
  pt_photos.this,
  pt_photos.that

And if your query really has a WHERE 1 clause you can drop it.

Answer 2

It's not tested in any way, but try and see if this helps

SELECT *, SUM(ifnull(vo.vote,0)) points, CONCAT(us.name, " ", us.surname) name_surname
FROM pt_photos ph
LEFT JOIN pt_votes vo ON vo.pID = ph.pID
JOIN pt_users us ON us.uID = ph.uID

Answer 3

Use JOINs, not correlated subqueries.

SELECT ph.*,
       IFNULL(SUM(vo.vote), 0) points,
       CONCAT(us.name, " ", us.surname) name_surname
FROM photos ph
LEFT JOIN pt_votes vo ON ph.pId = vo.pID
INNER JOIN pt_users us ON us.uID = ph.uID
GROUP BY ph.uID

As far as I know, there's no need for a WHERE 1 clause to select all rows. Just omit the WHERE clause.

Answer 4

Correlated subqueries make for light dimming queries on large sets, due to the number of times that subquery is executed.

Here's an alternative that should perform better:

SELECT ph.*
     , IFNULL(vs.points,0) AS points
     , CONCAT(us.name," ",us.surname) AS name_surname
  FROM pt_photos ph
  LEFT
  JOIN ( SELECT vo.pID
              , SUM(vo.vote) AS points
           FROM pt_votes vo
          GROUP BY vo.pID
       ) vs
    ON vs.pID = ph.pID
  LEFT
  JOIN pt_users us 
    ON us.uID = ph.uID
 WHERE 1

A suitable index (ideally a covering index) on pt_votes will improve performance of the GROUP BY... on pt_votes (pID, vote).

I'm going to assume that uID is the pirmary key of pt_users.

Answer 5

I would try rewriting the third select as an inner join, and possibly move the concat function to when you're outputting the results. This is a t-sql solution, but it should be similar enough:

select ph.*, 
concat(pu.name, " ", pu.surname) as name_surname,
isnull((select sum(vote) from pt_votes vo where vo.pid = ph.pid),0) as points
from pt_photos ph
inner join pt_users pu on ph.uid = pu.uid

You could also do a left join to the pt_votes table for the sum, but then you'd have to play with 'group by' clauses to get it all working, which might be worth it for the processing time.