使用 COUNT 进行子查询的慢 MYSQL 查询

Question

Right I have no idea why but this query takes well over 6 seconds to execute, index's are all setup correctly and if I run each query separately it works great with less than 0.5 seconds to execute.

Here is the query

SELECT c.supplier_id, supplier_name, address1, address2, address3, address4, suppliertype, postcode, contact_name,
(SELECT COUNT(*)
    FROM supplier_questions q1
    WHERE c.supplier_id = q1.supplier_id AND q1.incomplete = '0') AS questions, 
IF (active=1,'Yes', IF (active=2, 'NCR Only','Inactive')) AS rated, 
(SELECT COUNT(*)
    FROM supplier_questions q2
    WHERE c.supplier_id = q2.supplier_id AND q2.reviewed = '1') AS reviewed, 
questapproved, 
ss.supplier_no AS supplier_no
FROM suppliers c
INNER JOIN supplier_site ss ON c.supplier_id = ss.supplier_id
WHERE c.supplier_id != '0' AND ss.site_id = '2'
GROUP BY c.supplier_id
ORDER BY c.supplier_name ASC
LIMIT 0, 20

Results of the Explain query is as follows

id  select_type table   type    possible_keys   key key_len ref rows    Extra
1   PRIMARY ss  ref site_id,supplier_id site_id 4   const   1287    Using where; Using temporary; Using filesort
1   PRIMARY c   eq_ref  PRIMARY PRIMARY 4   ss.supplier_id  1   
3   DEPENDENT SUBQUERY  q2  ref supplier_id,reviewed    reviewed    4   const   263 Using where
2   DEPENDENT SUBQUERY  q1  ref supplier_id,incomplete  incomplete  4   const   254 Using where

The reason the count queries are in there is because I need to know the number of rows from those tables, this can't be done in another query as the results also need to be sorted by those values:(

Answer 1

As a stab in the dark, does this run faster? (I havent got a mysql to verify the syntax on, so forgive any slight mistakes, but you might get the idea)

SELECT c.supplier_id, supplier_name, address1, address2, address3, address4, suppliertype, postcode, contact_name, questions, reviewed 
IF (active=1,'Yes', IF (active=2, 'NCR Only','Inactive')) AS rated,  
questapproved,  ss.supplier_no AS supplier_no 
FROM suppliers c 
INNER JOIN supplier_site ss ON c.supplier_id = ss.supplier_id 
inner join 
(SELECT supplier_id, sum(if(incomplete='0',1,0)) as questions,  sum(if(incomplete='1',1,0)) as reviewed FROM supplier_questions q1 group by supplier_id) as tmp
on c.supplier_id = tmp.supplier_id
WHERE c.supplier_id != '0' AND ss.site_id = '2' 
GROUP BY c.supplier_id 
ORDER BY c.supplier_name ASC LIMIT 0, 20 

Answer 2

FROM suppliers c
INNER JOIN supplier_site ss ON c.supplier_id = ss.supplier_id
WHERE c.supplier_id != '0' AND ss.site_id = '2'
GROUP BY c.supplier_id 
ORDER BY c.supplier_name ASC

Since autogenerated primary keys are never equal to 0 (unless big db design mistake) you can drop the c.supplier_id.= '0' clause.

ss.site_id = '2' should be in the JOIN condition for readability.

It looks like this should match only one row in table supplier_site per supplier (if this is your usual 1-N thing-addresses relation, ie you're selecting the second address of each supplier, maybe '2' corresponds to 'billing address' or something) so the GROUP BY c.supplier_id is useless. If the GROUP BY actually does something, then the query is wrong, since the "address" columns, which presumably come from supplier_site table, would come from a random row.

So here's the simplified FROM (the WHERE is gone):

FROM suppliers c
INNER JOIN supplier_site ss ON 
    (c.supplier_id = ss.supplier_id AND ss.site_id = '2')
ORDER BY c.supplier_name ASC

I suppose you got an index on c.supplier_name so this part of the query should be very fast.

Now try this query:

SELECT a.*,
    questapproved, 
    ss.supplier_no AS supplier_no,
    IF (active=1,'Yes', IF (active=2, 'NCR Only','Inactive')) AS rated, 
    sum( q.incomplete = '0') AS questions,
    sum( q.reviewed = '1' ) AS reviewed
FROM
(
    SELECT c.supplier_id, supplier_name, address1, address2, address3, address4, suppliertype, postcode, contact_name
    FROM suppliers c
    INNER JOIN supplier_site ss ON 
        (c.supplier_id = ss.supplier_id AND ss.site_id = '2')
    ORDER BY c.supplier_name ASC
    LIMIT 0, 20
) a
LEFT JOIN supplier_questions q ON (q.supplier_id = c.supplier_id)
GROUP BY c.supplier_id
ORDER BY c.supplier_name;

Answer 3

If you remove the sub-selects you end up with something like this:

SELECT c.supplier_id, supplier_name, address1, address2, address3, address4, suppliertype, postcode, contact_name,
COUNT(IF (q1.incomplete = '0', '0', null)) AS questions, 
IF (active=1,'Yes', IF (active=2, 'NCR Only','Inactive')) AS rated, 
COUNT(IF (q1.reviewed = '1', '1', null)) AS reviewed,
questapproved, 
ss.supplier_no AS supplier_no
FROM suppliers c
INNER JOIN supplier_site ss ON c.supplier_id = ss.supplier_id
LEFT OUTER JOIN supplier_questions q1 ON c.supplier_id = q1.supplier_id
WHERE c.supplier_id != '0' AND ss.site_id = '2'
GROUP BY c.supplier_id
ORDER BY c.supplier_name ASC
LIMIT 0, 20

I don't have a MySQL database available so there may be errors in my SQL. The idea is to remove the subqueries and replace them with an outer join and use IF to only count relevant rows.

Answer 4

I would first attempt restructure by pre-querying the aggregates by supplier the count of questions and reviewed ONCE. Then, join to the rest of the details. By using the STRAIGHT_JOIN keyword, it should process in the order displayed. This will pre-aggregate first and use THAT as the basis to join back to suppliers and then supplier sites. No outer group by needed since its based on a supplier ID anyhow. However, the join to supplier_sites (your ss.supplier_no) would imply a supplier has more than one location. Does that mean the address and active status columns are originating from that table?

Should the join of questions be associated with a specific supplier and it's corresponding site location or not?

Additionally, since the prequery has the WHERE clause on supplier_id,= '0', it's not needed down stream since that will be the basis of a normal join to the other tables. thus eliminating them out of the result set.

SELECT STRAIGHT_JOIN
      PreAggregate.supplier_id, 
      PreAggregate.supplier_name, 
      address1, 
      address2, 
      address3, 
      address4, 
      suppliertype, 
      postcode, 
      contact_name,
      PreAggregate.Questions,
      IF (active=1,'Yes', IF (active=2, 'NCR Only','Inactive')) AS rated, 
      PreAggregate.Reviewed,
      questapproved, 
      ss.supplier_no AS supplier_no
   FROM 
      (select 
             s1.Supplier_ID,
             s1.Supplier_Name,
             SUM( IF( q1.Incomplete = '0', 1, 0 )) Questions,
             SUM( IF( q1.Reviewed = '1', 1, 0 )) Reviewed
          from 
             suppliers s1
                join supplier_questions q1
                   ON s1.supplier_id = q1.supplier_id
          where
             s1.supplier_id != '0'
          group by
             s1.Supplier_ID 
          ORDER BY 
             s1.supplier_name ASC ) PreAggregate

       JOIN suppliers c
          ON PreAggregate.Supplier_ID = c.Supplier_ID

       JOIN supplier_site ss 
          ON PreAggregate.Supplier_ID = ss.supplier_id
          AND ss.Site_ID = '2'
  LIMIT 0, 20