如何更快地下订单？

Question

I have a table with several columns to order by. Those columns are float values with their corresponding indexes on them. The table has around 5 million rows .

The table structure id something like that:

MyTable(id,a,b,c,d,e,f,g,h)
MyWords(id,word)
TableWords(wordID,MyTableID)

I need to run this query in a faster way which is basically a 100 top. I would like to have some suggestions on how to do that better:

select sql_no_cache a.* 
from MyWords w
left join TableWords wl on w.id = wl.wordID
left join MyTable a on wl.MyTableID = a.id
where w.word = 'WordToSearch'
order by a.a desc
limit 100

The indexes:

ALTER TABLE `a` ADD INDEX `a` (`a`);
ALTER TABLE `a` ADD INDEX `b` (`b`);

Using explain:

1   SIMPLE  w   const   uniqueWord,fulltext_word    uniqueWord  63  const   1   Using temporary; Using filesort
1   SIMPLE  wl  ref wordID  wordID  4   const   5597    
1   SIMPLE  a   eq_ref  PRIMARY,id  id  4   test.wl.a   1   

Describe TableWords:

MyTableID   int(11)   NO    MUL
wordID      int(11)   NO    MUL

Describe MyWords:

id      int(11)     NO  PRI     auto_increment
word    varchar(20) YES UNI     

Describe MyTable:

id  int(11) NO  PRI auto_increment
a   float   YES MUL 
b   float   YES MUL 

I can order by my table and it is fast, but in the moment I make a left join to that table it doesnt use the indexes anymore. I put more data in my question.

Thanks

Answer 1

You need to provide a lot more information if you are looking for specific optimizations. However in a general sense here is some things you can look into. I'll provide a short list with a link to the full details.

1. Use Indexes
2. Modify/Update filesort algorithm
3. Make sure columns use only the smallest amount of space required
4. Have lots of space available in the temporary directory

Optimize Order By: https://dev.mysql.com/doc/refman/5.7/en/order-by-optimization.html

Optimize Limit: https://dev.mysql.com/doc/refman/5.7/en/limit-optimization.html

Answer 2

The only way to speed it up is to have word and a in the same table and have INDEX(word, a) .

By doing that, both the filtering ( WHERE word = '...' ) and the ordering can be done in an index. That allows LIMIT 100 to simply peel off the first hundred rows -- instead of collecting all the filtered rows, sorting them, and only then being able to peel off 100.

There is another requirement: 'WordToSearch' must map to only one a.id . It is unclear from your description whether the joins are 1:1 or 1:many or many:many.