Postgres 对常用词的全文搜索性能

Question

当有人尝试查询非常常见的内容时，我正在努力提高搜索性能。我有一个数据库，其中包含 530 万条记录及其邮寄地址，其中很大一部分的常用词是“road”、“rd”、“st”等......所以当有人搜索这个时，它需要很长一段时间。

如下所示，我尝试搜索不常见的东西（箭头）：

pulsar_dev=# EXPLAIN ANALYZE SELECT
                property->>'rollNumber',
                property->>'municipalAddress',
                property->>'municipalityDescription'
FROM
                properties_cmv
WHERE
                to_tsvector('simple', property->>'municipalAddress') ||
                to_tsvector('simple', property->>'municipalityDescription') ||
                to_tsvector('simple', property->>'countyDescription') @@ plainto_tsquery('arrowhead')
ORDER BY ts_rank(to_tsvector('simple', property->>'municipalAddress') ||
                to_tsvector('simple', property->>'municipalityDescription') ||
                to_tsvector('simple', property->>'countyDescription'), plainto_tsquery('arrowhead')) DESC;
                                                                                                                                                    QUERY PLAN

---------------------------------------------------------------------------------------------------------------------------------------------------------------------
----------------------------------------------------------------------------------------------------------------------------------------------
 Sort  (cost=4420.99..4424.11 rows=1248 width=23) (actual time=136.957..137.047 rows=490 loops=1)
   Sort Key: (ts_rank(((to_tsvector('simple'::regconfig, (property ->> 'municipalAddress'::text)) || to_tsvector('simple'::regconfig, (property ->> 'municipalityDesc
ription'::text))) || to_tsvector('simple'::regconfig, (property ->> 'countyDescription'::text))), plainto_tsquery('arrowhead'::text)))
   Sort Method: quicksort  Memory: 93kB
   ->  Bitmap Heap Scan on properties_cmv  (cost=25.69..4356.81 rows=1248 width=23) (actual time=0.350..136.566 rows=490 loops=1)
         Recheck Cond: (((to_tsvector('simple'::regconfig, (property ->> 'municipalAddress'::text)) || to_tsvector('simple'::regconfig, (property ->> 'municipalityDe
scription'::text))) || to_tsvector('simple'::regconfig, (property ->> 'countyDescription'::text))) @@ plainto_tsquery('arrowhead'::text))
         Heap Blocks: exact=39
         ->  Bitmap Index Scan on prop_address_idx  (cost=0.00..25.38 rows=1248 width=0) (actual time=0.072..0.072 rows=490 loops=1)
               Index Cond: (((to_tsvector('simple'::regconfig, (property ->> 'municipalAddress'::text)) || to_tsvector('simple'::regconfig, (property ->> 'municipali
tyDescription'::text))) || to_tsvector('simple'::regconfig, (property ->> 'countyDescription'::text))) @@ plainto_tsquery('arrowhead'::text))
 Planning time: 0.213 ms
 Execution time: 137.184 ms
(10 rows)

相当快，但是当我搜索“路”时，它一点也不快：

pulsar_dev=# EXPLAIN ANALYZE SELECT
                property->>'rollNumber',
                property->>'municipalAddress',
                property->>'municipalityDescription'
FROM
                properties_cmv
WHERE
                to_tsvector('simple', property->>'municipalAddress') ||
                to_tsvector('simple', property->>'municipalityDescription') ||
                to_tsvector('simple', property->>'countyDescription') @@ plainto_tsquery('road')
ORDER BY ts_rank(to_tsvector('simple', property->>'municipalAddress') ||
                to_tsvector('simple', property->>'municipalityDescription') ||
                to_tsvector('simple', property->>'countyDescription'), plainto_tsquery('road')) DESC;
                                                                                                                                                  QUERY PLAN

---------------------------------------------------------------------------------------------------------------------------------------------------------------------
-----------------------------------------------------------------------------------------------------------------------------------------
 Sort  (cost=25533.10..25560.73 rows=11051 width=23) (actual time=11065.051..11066.883 rows=10356 loops=1)
   Sort Key: (ts_rank(((to_tsvector('simple'::regconfig, (property ->> 'municipalAddress'::text)) || to_tsvector('simple'::regconfig, (property ->> 'municipalityDesc
ription'::text))) || to_tsvector('simple'::regconfig, (property ->> 'countyDescription'::text))), plainto_tsquery('road'::text)))
   Sort Method: quicksort  Memory: 1841kB
   ->  Bitmap Heap Scan on properties_cmv  (cost=117.67..24790.93 rows=11051 width=23) (actual time=1.911..11052.683 rows=10356 loops=1)
         Recheck Cond: (((to_tsvector('simple'::regconfig, (property ->> 'municipalAddress'::text)) || to_tsvector('simple'::regconfig, (property ->> 'municipalityDe
scription'::text))) || to_tsvector('simple'::regconfig, (property ->> 'countyDescription'::text))) @@ plainto_tsquery('road'::text))
         Heap Blocks: exact=1408
         ->  Bitmap Index Scan on prop_address_idx  (cost=0.00..114.91 rows=11051 width=0) (actual time=1.432..1.432 rows=10356 loops=1)
               Index Cond: (((to_tsvector('simple'::regconfig, (property ->> 'municipalAddress'::text)) || to_tsvector('simple'::regconfig, (property ->> 'municipali
tyDescription'::text))) || to_tsvector('simple'::regconfig, (property ->> 'countyDescription'::text))) @@ plainto_tsquery('road'::text))
 Planning time: 0.210 ms
 Execution time: 11069.142 ms
(10 rows)

如何提高第二个查询的性能？我还需要对结果进行排名，首先返回最相关的结果。

在 elasticsearch 上运行类似的测试以毫秒为单位返回。

Answer 1

我创建了一个新表并将连接的 tsvector 保存在一个列和索引中，这似乎提高了速度。