如何优化此查询的执行时间

Question

我有以下查询：

SELECT "factures"."id" 
FROM   "factures" 
WHERE  ( "factures"."id" NOT IN (SELECT DISTINCT( "echeances"."facture_id" ) 
                                 FROM   "echeances" 
                                 WHERE  "echeances"."type_decheance" IN ( 2, 3, 4, 5, 8, 9 ) 
                                        AND "echeances"."facture_id" IS NOT NULL 
                                 LIMIT  100000)) <----- removing this limit makes the query take enormous time
ORDER  BY "factures"."id" DESC

这里是限制为100 000的解释分析：

Index Only Scan Backward using factures_id_pkey on factures  (cost=93516.76..211292.17 rows=530570 width=4) (actual time=1425.701..11466.759 rows=963698 loops=1)
  Filter: (NOT (hashed SubPlan 1))
  Rows Removed by Filter: 99997
  Heap Fetches: 1063695
  SubPlan 1
    ->  Limit  (cost=0.43..93266.34 rows=100000 width=4) (actual time=0.022..1229.925 rows=100000 loops=1)
          ->  Unique  (cost=0.43..264837.37 rows=283959 width=4) (actual time=0.022..1090.692 rows=100000 loops=1)
                ->  Index Scan using echeances__facture_id__idx on echeances  (cost=0.43..262883.29 rows=781631 width=4) (actual time=0.020..819.735 rows=100167 loops=1)
                      Index Cond: (facture_id IS NOT NULL)
"                      Filter: (type_decheance = ANY ('{2,3,4,5,8,9}'::integer[]))"
                      Rows Removed by Filter: 156995
Planning time: 0.249 ms
Execution time: 11960.423 ms

下面是无限解释：

->  Unique  (cost=0.43..264837.37 rows=283959 width=4)
Index Only Scan Backward using factures_id_pkey on factures  (cost=0.86..142233669403.15 rows=530570 width=4)
  Filter: (NOT (SubPlan 1))
  SubPlan 1
    ->  Materialize  (cost=0.43..267367.16 rows=283959 width=4)
                ->  Index Scan using echeances__facture_id__idx on echeances  (cost=0.43..262883.29 rows=781631 width=4)
                      Index Cond: (facture_id IS NOT NULL)
"                      Filter: (type_decheance = ANY ('{2,3,4,5,8,9}'::integer[]))"

这是架构

Table "factures"
id

Table "echeances"
id
facture_id (fk)
type_decheance (integer)

问题是 "factures" 和 "echeances" table 有大量的行，如果 :

limit在子查询中指定，比如limit 100000，查询时间快
limit没有在子查询中指定，耗费了很多时间，我等了15多分钟只好停止了。

目标是让这个查询运行在合理的时间内不受限制。

Answer 1

切换到NOT EXISTS：

SELECT f.id
FROM factures f
WHERE NOT EXISTS (SELECT 1
                  FROM echeances e
                  WHERE e.facture_id = f.id AND
                        e.type_decheance IN ( 2, 3, 4, 5, 8, 9 ) 
                 )
ORDER  BY f.id DESC;

请注意，我删除了所有双引号。不要引用您的标识符。它只会让查询更难编写和阅读。

那么您需要 echeances(facture_id, type_decheance) 上的索引。这应该非常快，因为可以通过简单的索引查找来检查每个制造 ID。

如何优化此查询的执行时间

How to optimize the execution time for this query

sql

postgresql

query-performance

psqlodbc