PostgreSQL 查询匹配列文本字符串非常慢，即使有索引

Question

我有一个包含航空公司时刻表数据的表，目前大约有 4.5 亿行。

我无法弄清楚为什么我的数据库上的以下查询需要这么长时间才能运行（几分钟）。如果我正确阅读了查询计划，它似乎会挂在一个针对完全索引列的特定相等子句上，所以我不确定我还能做些什么来解决这个问题。

其他查询的运行时间似乎没有那么长，因此它可能与查询本身有关，而不是与硬件/其他方面有关。任何帮助将不胜感激！

有问题的表：

   Column    |            Type             | Collation | Nullable | Default
-------------+-----------------------------+-----------+----------+---------
 pubdate     | date                        |           | not null |
 flowndate   | date                        |           | not null |
 carcode     | character varying(3)        |           | not null |
 fltno       | character varying(5)        |           | not null |
 acfttype    | character varying(4)        |           | not null |
 svctype     | character varying(5)        |           |          |
 traffrest   | character varying(1)        |           |          |
 depstn      | character varying(3)        |           | not null |
 arrstn      | character varying(3)        |           | not null |
 depdatetime | timestamp without time zone |           | not null |
 deputcvar   | smallint                    |           | not null |
 arrdatetime | timestamp without time zone |           | not null |
 arrutcvar   | smallint                    |           | not null |
 seats       | integer                     |           |          |
 acftclass   | character varying(3)        |           |          |
Indexes:
    "index_dailyflights_arrstn" btree (arrstn)
    "index_dailyflights_depstn" btree (depstn)
    "index_dailyflights_flowndate" btree (flowndate)
    "index_dailyflights_pubdate" btree (pubdate DESC)

查询：

SELECT carCode, COUNT(*) AS departures
FROM truesight_data.oag_dailyflights df
WHERE flowndate <@ '[2016-01-01,2017-10-31]'::daterange
AND (flowndate <= ((DATE_TRUNC('MONTH', pubDate) + INTERVAL '1 MONTH - 1 day')::date) OR PubDate = (SELECT MAX(PubDate) FROM truesight_data.oag_dailyflights))
AND depStn = 'PIT'
AND (svcType IN ('cargo','combi') OR acftClass = 'WB')
GROUP BY carCode
ORDER BY COUNT(*) DESC;

还有这个查询的EXPLAIN (ANALYZE, BUFFERS)：

               QUERY PLAN                                                       
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 Sort  (cost=140256.14..140256.18 rows=18 width=11) (actual time=22013.297..22013.297 rows=2 loops=1)
   Sort Key: (count(*)) DESC
   Sort Method: quicksort  Memory: 25kB
   Buffers: shared hit=985 read=142487
   I/O Timings: read=20089.741
   ->  GroupAggregate  (cost=140255.45..140255.76 rows=18 width=11) (actual time=22013.281..22013.284 rows=2 loops=1)
         Group Key: carcode
         Buffers: shared hit=985 read=142487
         I/O Timings: read=20089.741
         ->  Sort  (cost=140255.45..140255.49 rows=18 width=3) (actual time=22013.262..22013.263 rows=27 loops=1)
               Sort Key: carcode
               Sort Method: quicksort  Memory: 26kB
               Buffers: shared hit=985 read=142487
               I/O Timings: read=20089.741
               ->  Index Scan using index_dailyflights_depstn on oag_dailyflights df  (cost=0.57..140255.07 rows=18 width=3) (actual time=13947.770..22013.193 rows=27 loops=1)
                     Index Cond: ((depstn)::text = 'PIT'::text)
                     Filter: ((flowndate <@ '[2016-01-01,2017-11-01)'::daterange) AND (((svctype)::text = ANY ('{cargo,combi}'::text[])) OR ((acftclass)::text = 'WB'::text)) AND current_status(df.*))
                     Rows Removed by Filter: 616867
                     Buffers: shared hit=985 read=142487
                     I/O Timings: read=20089.741
 Planning Time: 0.460 ms
 Execution Time: 22013.389 ms
(22 rows)

再次感谢您的帮助——很高兴回答您可能遇到的任何其他问题。

Answer 1

您有一个复杂的where 条件。我能提出的唯一建议是在(depStn, flowndate) 上建立一个复合索引。这可能会有所帮助，具体取决于flowndate 条件的选择性。