Question

在以下情况下，Postgres查询的执行速度非常慢。谁能帮助我指出问题的原因或这种现象的原因？

我有3个不同的平台，每个平台都使用两个结构相同的表：

订单（每个平台的所有订单信息）
order_product（每个订单购买的所有产品=每个平台的客户篮子）

还有一个全局产品表，其中包含所有平台上的产品信息：

产品

CREATE TABLE产品（ product_id字符变化（64）， product_name字符变化（512））;

我还具有将所有平台的所有order和order_product表与UNION ALL相结合的视图。

CREATE VIEW order_product_all_platforms AS
SELECT "platform", "order_id", "product_id" 
FROM "public"."order_product_1"
UNION ALL 
SELECT "platform", "order_id", "product_id" 
FROM "public"."order_product_2"
UNION ALL 
SELECT "platform", "order_id", "product_id" 
FROM "public"."order_product_3" ;

..

CREATE VIEW order_all_platforms AS
SELECT "platform", "order_id", "purchase_date" 
FROM "public"."order_1"
UNION ALL 
SELECT "platform", "order_id", "purchase_date" 
FROM "public"."order_2"
UNION ALL 
SELECT "platform", "order_id", "purchase_date" 
FROM "public"."order_3"

当我想在查询中合并所有这三个表时，就会出现问题：

 SELECT "t1.platform", "t1.order_id", t2.product_id, t3.product_name
    FROM order_all_platforms t1 
      INNER JOIN order_product_all_platforms t2 ON t1.platform = t2.platform and t1.order_id = t2.order_id
      INNER JOIN product t3 ON t2.product_id = t3.product_id
 WHERE purchase_date >= '2018-01-01'

我认为查询计划中有问题的部分看起来像这样。尽管存在product_id的索引，但Postgres仍对product_id执行顺序扫描：

->  Hash Join  (cost=23068.58..1963856.51 rows=37164213 width=67)"
        Hash Cond: ((t2.product_id)::text = (t3.product_id)::text)"
        ->  Append  (cost=0.00..1104593.13 rows=37164213 width=29)"
              ->  Seq Scan on order_product_table1 (cost=0.00..1066147.91 rows=35935991 width=29)"
              ->  Seq Scan on order_product_table2 (cost=0.00..35008.37 rows=1126337 width=32)"
              ->  Seq Scan on order_product_table3 (cost=0.00..3436.85 rows=101885 width=30)"
        ->  Hash  (cost=16209.37..16209.37 rows=548737 width=47)"
              ->  Seq Scan on product t3 (cost=0.00..16209.37 rows=548737 width=47)"

我该怎么做才能加快速度？

这些是索引。第一个索引也存在于订单表上，第二个索引也存在于产品表上：

CREATE INDEX  order_product_table1_index1
  ON  order_product_table1
  USING btree
  (platform_name COLLATE pg_catalog."default", order_id);  

CREATE INDEX mat_ order_product_table1_index2
  ON  order_product_table1
  USING btree
  (product_id COLLATE pg_catalog."default");

Postgres表现不佳UNION ALL + INNER JOIN

0 个答案: