mysql 子查询的强制索引

Question

in this query we only talk about the ability to set the index JOIN for the subquery. Don't concern about the performance of the query

SELECT cate.id, cate.name as category_name, b.id book_id, b.name book_name
FROM categories as cate
LEFT JOIN (select * from books where books.name='some pets 1') as b
ON b.category_id = cate.id;

I would like to reduce rows of table books by using sub-query to create temporary table 'b', then try to force join index for 'b' but it didn't work (" force index (category_id) "). My draft code:

SELECT cate.id, cate.name as category_name, b.id book_id, b.name book_name
FROM categories as cate
LEFT JOIN (select * from books force index (idx_books_name) where books.name='some pets 1') b force index (category_id)
ON b.category_id = cate.id;

It through the syntax error at "force index". Is the idea possible, you think?

Answer 1

Your "subquery" b is a derived table. A derived table is (basically) a completely new table. This is what you intended, you wanted to create a small temporary table b . But this new table has no access to the original indexes, as it is a new table. This is why you get the error message.

Imagine how MySQL would need to use the category index. That index itself didn't change, so it will still contain entries for all rows of the original table. So you would use this index to find records with that category_id in the whole table (as that is what this index does), and then somehow need to check if that entry is in your derived table. Since a derived table can consist of virtually anything, even without any reference to a base table, these is no generic way to do this "somehow".

In your case, this could be done by reevaluating the condition where books.name='some pets 1' , but this would mean you lose any advantage you might get from using the idx_books_name index and your derived table in the first place, so it would be the same as

LEFT JOIN books b ON b.category_id = cate.id AND b.name='some pets 1');

which would, with an an index books(category_id, name) , be using indexes as you intended.

Using the derived table the way you did is basically already an optimization that wants to prevent MySQL from doing the join in the usual way (although MySQL would be allowed to merge your condition back into the outer query). You'll do this if you have special circumstances, eg if there is only a very small percentage of rows that match this condition or if you don't have a useful index on category_id .

MySQL might decide to generate a new index for your derived table on category_id . You (currently) cannot enforce generating those indexes by optimizer hints, and it is worth noting that indexes on small tables may not be as useful as you might think. Indexes scale logarithmically, so while the effect on large tables is huge, for small tables the overhead of generating such an index might be larger than the benefit.