为什么我的查询在 MySQL 5.7 上需要超过 30 分钟，但在 MySQL 8 上需要几秒钟？

Question

I have the following query. On MySQL 5.7 it takes over 30 mins minutes to run but on MySQL 8 it takes less than 10 seconds. There are around 20k records in students_t, 500 in classes_t and 1100 in groups_t. student_enrolled_classes_t contains 100288 records and 160156 records.There are no foreign keys in place.

SELECT DISTINCT
       s.forename
     , s.surname
     , count(c.id) as 'classes'
     , group_concat(DISTINCT g.name) 'groups'
  FROM students_t s
     , student_enrolled_classes_t c
     , student_enrolled_groups_t eg
     , groups_t g
 WHERE s.id = c.student_id
   AND s.id = eg.student_id
   AND eg.group_id = g.id
GROUP BY s.id, c.id
ORDER BY s.forename ASC;

The table structures are (generated with Hibernate):

CREATE TABLE classes_t 
(id BIGINT auto_increment PRIMARY KEY);

CREATE TABLE groups_t 
(id BIGINT auto_increment PRIMARY KEY
,name VARCHAR(255) NULL);

CREATE TABLE student_enrolled_classes_t
(id BIGINT auto_increment PRIMARY KEY
,class_id BIGINT NULL
,student_id BIGINT NULL);

CREATE TABLE student_enrolled_groups_t
(id BIGINT NOT NULL PRIMARY KEY
,group_id BIGINT NULL
,student_id BIGINT NULL);

CREATE TABLE students_t
(id BIGINT auto_increment PRIMARY KEY
,forename VARCHAR(255) NULL
,surname  VARCHAR(255) NULL);

The question is: Why does it take 30 mins on MySQL 5.7 but only 10 seconds on MySQL 8?

Thank you

Answer 1

You could write the query as follows. It uses newer join syntax and takes into a account situations where student is not part of a class/group.

SELECT 
       s.forename
     , s.surname
     , count(distinct c.id) as 'classes'
     , group_concat(distinct g.name) 'groups'
  FROM students_t s
    LEFT JOIN student_enrolled_classes_t c ON c.student_id=s.id
    LEFT JOIN student_enrolled_groups_t eg ON eg.student_id=s.id
    LEFT JOIN groups_t g ON eg.group_id = g.id
GROUP BY s.id
ORDER BY s.forename ASC;

Answer 2

We need to see EXPLAIN SELECT... to analyze why one runs so slowly. While you are at it, also get EXPLAIN FORMAT=JSON SELECT... .

This reformulation and indexes may speed up the query on both servers, possibly making them both fast. Note that it gets rid of the GROUP BY and DISTINCT , both of which slow down the query.

SELECT  s.forename, s.surname, 
        ( SELECT  COUNT(*)
            FROM  student_enrolled_classes_t
            WHERE  student_id = s.id 
        ) AS 'classes', 
        ( SELECT  GROUP_CONCAT(g.name)
                    FROM  student_enrolled_groups_t AS eg
                    JOIN  groups_t AS g  ON eg.group_id = g.id
                    WHERE  eg.student_id = s.id 
        ) AS 'groups'
    FROM  student_t AS s
    ORDER BY  s.forename ASC;

It seems that student_enrolled_groups_t is a many-to-many mapping table. In that case, get rid of id and have these two indexes:

PRIMARY KEY(student_id, group_id)
INDEX(group_id, student_id)

When writing JOINs , please use the newer syntax involving ON -- for saying how the tables are 'related'. Filtering is done in WHERE .