MySQL中使用联接的查询缓慢

Question

I have a query that runs too slow but I can't figure out the reason behind it, or a solution to it. The query took around 22 sec to run at first, but then I tried to remove a foreign key in the table Book that is a FK to ParentID in the same table. After this removal the query took around 1 sec . This is of course good but there should be room for even more improvements, either restructure the query itself or by using different indexes.

Can anyone advise? Why does the foreign key to Book.ParentID cause such an increase in query time? What is the best approach here, try to rewrite the query or look into the indexes?

Except for dropping the foreign key to Book.ParentID , I've tried using use index on the selected table and some of the joins but with no real luck.

Here's a fiddle: http://sqlfiddle.com/#!9/7f7bbb/9

However, the explain plan in the fiddle differ from the one I get in production. This is the explain plan from production (with the foreign key to Book.ParentID):

'1','SIMPLE','Bk',NULL,'index_merge','PRIMARY,fk_CategoryParent_ID_idx','PRIMARY,fk_CategoryParent_ID_idx','8,9',NULL,'2','100.00','Using union(PRIMARY,fk_CategoryParent_ID_idx); Using where; Using temporary; Using filesort'
'1','SIMPLE','CGB2',NULL,'ref','uq_ChapterGroup_Category,book_chaptergroup_unique_row_constraint,fk_Category_ChapterGroup_ID,ix_chapterGroupBook_chapterGroupID','book_chaptergroup_unique_row_constraint','9','Hypo.Bk.ID','8','100.00','Using where'
'1','SIMPLE','CCG2',NULL,'ref','uq_ChapterID_ChapterGroupItemInChapterGroup,uq_ExamID_ChapterGroupItemInChapterGroup,fk_ChapterGroup_Chapter_ID,ix_chapterChapterGroup_chapterID','fk_ChapterGroup_Chapter_ID','8','Hypo.CGB2.ChapterGroupID','2','100.00','Using where'
'1','SIMPLE','ATM',NULL,'ref','PRIMARY,fk_Chapter_AnswerTextMarker_ID,fk_QuestionGroup_AnswerTextMarker_ID','fk_Chapter_AnswerTextMarker_ID','8','Hypo.CCG2.ChapterID','5','100.00','Using where'
'1','SIMPLE','QG',NULL,'eq_ref','PRIMARY','PRIMARY','8','Hypo.ATM.QuestionGroupID','1','100.00','Using index'
'1','SIMPLE','AQS',NULL,'ref','PRIMARY,fk_AQSession_AnswerTextMarker_ID,ix_answerQuestionSession_questionSessionID','fk_AQSession_AnswerTextMarker_ID','8','Hypo.ATM.ID','313','100.00',NULL
'1','SIMPLE','QS',NULL,'eq_ref','PRIMARY,fk_QSession_User_ID,ix_QuestionSession_userId_type_sessionDate','PRIMARY','8','Hypo.AQS.QuestionSessionID','1','5.00','Using where'

Row count in my environment:

AnswerQuestionSession 3992125
QuestionSession 367334
AnswerTextMarker 9696
QuestionGroup 18793
ChapterGroupItemInChapterGroup 42360
ChapterGroupBook 2297
Book 378

Answer 1

ChapterGroupBook looks like a many-to-many mapping table. If so, why are all the columns nullable? Make them (and any others elsewhere) NOT NULL where appropriate.

Because of the above, that table has no PRIMARY KEY . This is bad for InnoDB. Also see the my many:many tips .

OR is a performance killer. Is there any way you could avoid it in (Bk.ID=551 OR Bk.ParentID=551) ? If necessary, us a UNION in a derived table:

FROM ( ( SELECT id FROM Book WHERE id = 551 )
       UNION ALL
       ( SELECT id FROM Book WHERE parent_id = 551 ) )
JOIN ...

("Index merge" was used, but is likely to be slower.)

Please reformat the EXPLAIN so that I can see the Rows column clearly.