主页 / 问题 / 636031
Accepted
Kamo Petrosyan
Asked:2020-03-06 19:05:40 +0000 UTC2020-03-06 19:05:40 +0000 UTC

SQL 查询优化或反规范化？

  • 772

问题在于对数据库的查询需要极长的时间（最多 400 ms）

    SELECT * FROM production_category AS cc
        INNER JOIN LATERAL (
            SELECT cc.id AS id, SUM(products) AS p_count FROM (
                SELECT cc.id AS parent_id, category_id, COUNT(id) AS products FROM production_product_categories WHERE category_id IN (
                    SELECT id FROM production_category WHERE lft <= cc.rght AND lft >= cc.lft AND tree_id = cc.tree_id)
                    AND product_id IN ( SELECT id FROM production_product WHERE manufacturer_id = 15 )
                    GROUP BY category_id
                ) AS sub_cc
            GROUP BY parent_id ) AS cp USING(id)
    WHERE cp.p_count > 0;

结构如下：

有一个制造商（manufacturer），他的产品（production_product）和任何产品都可以属于一个或多个类别（production_category）。也可能有类别中没有该制造商的产品。

任务是选择包含该制造商产品的类别。

所以问题出现了：要么我对查询太聪明了，要么应该对数据库进行非规范化并将制造商的类别信息存储在单独的表中（确切关系（制造商 ID，类别 ID））。

PostgreSQL (9.3.4)

     Nested Loop  (cost=207.57..129375.37 rows=42 width=223)
       ->  Seq Scan on production_category cc  (cost=0.00..25.20 rows=620 width=191)
       ->  Subquery Scan on cp  (cost=207.57..208.62 rows=1 width=36)
             Filter: (cc.id = cp.id)
             ->  HashAggregate  (cost=207.57..208.09 rows=42 width=12)
                   Filter: (sum((count(production_product_categories.id))) > 0::numeric)
                   ->  HashAggregate  (cost=206.41..206.83 rows=42 width=8)
                         ->  Nested Loop Semi Join  (cost=9.23..206.20 rows=42 width=8)
                               ->  Nested Loop  (cost=8.94..106.81 rows=45 width=12)
                                     ->  Bitmap Heap Scan on production_category  (cost=4.31..12.78 rows=1 width=4)
                                           Recheck Cond: ((lft <= cc.rght) AND (lft >= cc.lft))
                                           Filter: (tree_id = cc.tree_id)
                                           ->  Bitmap Index Scan on production_category_caf7cc51  (cost=0.00..4.31 rows=3 width=0)
                                                 Index Cond: ((lft <= cc.rght) AND (lft >= cc.lft))
                                     ->  Bitmap Heap Scan on production_product_categories  (cost=4.64..93.58 rows=45 width=12)
                                           Recheck Cond: (category_id = production_category.id)
                                           ->  Bitmap Index Scan on production_product_categories_b583a629  (cost=0.00..4.62 rows=45 width=0)
                                                 Index Cond: (category_id = production_category.id)
                               ->  Index Scan using production_product_pkey on production_product  (cost=0.29..2.20 rows=1 width=4)
                                     Index Cond: (id = production_product_categories.product_id)
                                     Filter: (manufacturer_id = 15)
sql

2 个回答

  1. 我认为它应该是这样的：

    SELECT c.*
FROM production_category c 
INNER JOIN production_product_categories pc ON pc.category_id = c.id
INNER JOIN production_product p ON p.id = pc.product_id
WHERE p.manufacturer_id = 15

    太糟糕了，现在没有什么可检查的。

    • 1
  2. Best Answer

    在我看来，以下请求的结果不错（大约 100 毫秒）：

    SELECT cat.*
FROM production_category cat
WHERE cat.tree_id IN (
    SELECT 
        pc.tree_id
    FROM
        production_category pc
        INNER JOIN public.production_product_categories ppc ON (pc.id = ppc.category_id)
        INNER JOIN public.production_product pp ON (ppc.product_id = pp.id)
    WHERE
        pp.manufacturer_id = 15
    GROUP BY
        pc.id
) AND cat.parent_id IS NULL
ORDER BY cat.name;

    我想最好在一个请求中完成。

    • 0

相关问题