Postgresql，在postgres表

Question

我们在Postgres数据库中遇到有关选择数据的性能问题。

我有两张桌子，fleet_test和fleet_inspection_test。
fleet_test包含94列作为键：（fleet_id，usdot，country_code，contacts_count，....）和fleet_inspection_test包含9列作为键：（inspection_id，dot_number，insp_date ...）。
fleet_test有大约460k行，fleet_inspection_test大约有600万行。

我需要运行如下的查询，在数据库中运行大约需要4秒。

select sum(contacts_count) as sum from fleet_test
where country_code = 'US'
and usdot in (select  dot_number from fleet_inspection_test
            where insp_date between '2016-02-15' and '2016-12-15');

如果我只运行主查询，大约需要600 mesc：

select sum(contacts_count) as sum from fleet_test
where country_code = 'US'

我们为每列提供索引。不幸的是，结果并不那么好：

CREATE INDEX ix_fleet_test_contacts_count ON fleet_test USING btree (contacts_count);
CREATE INDEX ix_fleet_test_country_code ON fleet_test USING btree (country_code);
CREATE INDEX ix_fleet_test_usdot ON fleet_test USING btree (usdot);
CREATE INDEX ix_fleet_inspection_test_dot_number ON fleet_inspection_test USING btree (dot_number);
CREATE INDEX ix_fleet_inspection_test_insp_date ON fleet_inspection_test USING btree (insp_date);

查询计划

'Aggregate  (cost=198451.70..198451.71 rows=1 width=4) (actual time=12300.010..12300.011 rows=1 loops=1)'
'  Buffers: shared hit=89350'
'  ->  Hash Join  (cost=152330.84..198276.23 rows=70189 width=4) (actual time=10902.309..12168.892 rows=189740 loops=1)'
'        Hash Cond: (fleet_test.usdot = fleet_inspection_test.dot_number)'
'        Buffers: shared hit=89350'
'        ->  Seq Scan on fleet_test  (cost=0.00..43722.72 rows=444397 width=8) (actual time=0.015..562.385 rows=443616 loops=1)'
'              Filter: ((country_code)::text = 'US'::text)'
'              Rows Removed by Filter: 14726'
'              Buffers: shared hit=37993'
'        ->  Hash  (cost=151817.53..151817.53 rows=41065 width=4) (actual time=10898.319..10898.319 rows=356230 loops=1)'
'              Buckets: 524288 (originally 65536)  Batches: 1 (originally 1)  Memory Usage: 16620kB'
'              Buffers: shared hit=51357'
'              ->  HashAggregate  (cost=151406.88..151817.53 rows=41065 width=4) (actual time=10240.109..10604.871 rows=356231 loops=1)'
'                    Group Key: fleet_inspection_test.dot_number'
'                    Buffers: shared hit=51357'
'                    ->  Seq Scan on fleet_inspection_test  (cost=0.00..138905.67 rows=5000485 width=4) (actual time=126.081..5077.200 rows=5007065 loops=1)'
'                          Filter: ((insp_date >= '2016-02-15'::date) AND (insp_date <= '2016-12-15'::date))'
'                          Rows Removed by Filter: 829513'
'                          Buffers: shared hit=51357'
'Planning time: 1.092 ms'
'Execution time: 12302.407 ms'

这是我的查询中的EXPLAIN（ANALYZE，BUFFERS）示例。我的期望是在不到一秒的时间内完成执行时间。

这些是postgres db config的一些参数：
操作系统：CentOS
系统内存：16GB

work_mem = 768MB
shared_buffers = 2048MB

任何帮助，评论或想法都会非常感激！

提前谢谢。

Answer 1

试试这个（第3版）

    SELECT SUM(contacts_count) AS sum 
      FROM fleet_test AS t
      JOIN (SELECT DISTINCT dot_number 
              FROM fleet_inspection_test
             WHERE insp_date BETWEEN '2016-02-15' AND '2016-12-15') AS d
        ON (t.usdot = d.dot_number)
     WHERE t.country_code = 'US';