在EXISTS条件上创建索引

Question

我的表结构：

table_a(id, company_id, approval_status, is_locked)
table_b(tba_id, status)

我的查询：

SELECT COUNT(id) filter (WHERE approval_status = 2 
AND is_locked = true AND EXISTS 
(SELECT 1 from table_b WHERE table_b.tba_id = table_a.id 
AND table_b.status = 2) 
FROM table_a
GROUP BY company_id

我目前有以下索引，但性能仍然很慢：

CREATE INDEX multiple_filter_index ON table_a (approval_status, is_locked)

可以通过添加更好的索引来改善此查询的性能吗？

这是查询计划：

HashAggregate  (cost=463013.07..463013.10 rows=2 width=11) (actual time=47632.476..47632.476 rows=2 loops=1)
  Group Key: table_a.company_id
  ->  Seq Scan on table_a  (cost=0.00..3064.62 rows=100062 width=11) (actual time=0.003..23.326 rows=100062 loops=1)
  SubPlan 1
    ->  Seq Scan on table_b  (cost=0.00..477.27 rows=104 width=0) (actual time=1.430..1.430 rows=0 loops=33144)
          Filter: ((tba_id = table_a.id) AND (status = 2))
          Rows Removed by Filter: 17411
  SubPlan 2
    ->  Seq Scan on table_b table_b_1  (cost=0.00..433.73 rows=5820 width=4) (never executed)
          Filter: (status = 2)
Planning time: 0.902 ms
Execution time: 47632.565 ms

Answer 1

您当前的执行计划显示Postgres没有使用您定义的索引。相反，它只是对每个表执行两次顺序扫描，如果这些表很大，则不会特别有效。

首先，AFAIK您的查询将执行与此相同：

SELECT COUNT(id)
FROM table_a
WHERE
    approval_status = 2 AND
    is_locked = true AND
    EXISTS (SELECT 1 from table_b WHERE table_b.tba_id = table_a.id AND table_b.status = 2)
GROUP BY company_id;

也就是说，Postgres过滤器实际上只是行为与该逻辑在正式的WHERE子句中的行为相同。

我建议在两个表中的每一个上创建一个索引：

CREATE INDEX table_a_idx ON table_a (approval_status, is_locked, company_id);
CREATE INDEX table_b_idx ON table_b (status, tba_id);

table_a_idx指数的原因是我们希望使用approval_status和is_locked滤波器消除尽可能多的记录。我还在此索引中包含了company_id，以涵盖GROUP BY列，希望避免在遍历索引后执行额外的磁盘读取。

table_b_idx的存在是为了加快查询的EXISTS子句。

我还建议您使用COUNT(*)而不是COUNT(id)。

Answer 2

尝试将一些过滤逻辑移动到连接中

SELECT
    company_id
  , COUNT(CASE
        WHEN approval_status = 2 AND
            is_locked = TRUE AND
            b.tba_id IS NOT NULL
        THEN id
    END)
FROM table_a
LEFT JOIN (
    SELECT DISTINCT tba_id 
    FROM table_b
    ) b on b.tba_id = table_a.id
GROUP BY
    company_id