我遇到问题,使用btree_gin扩展名使postgres使用我的多列索引进行完整搜索。这是针对文章的搜索页面。使用btree_gin背后的想法是能够获得排序的'id'字段和magazine_id作为过滤器:
CREATE INDEX idx_gin_search ON article USING gin(id, magazine_id, search_vector_full) WITH (fastupdate = off);
Postgres决定在杂志上使用btree索引,然后过滤(=慢):
Executed SQL
SELECT ••• FROM article WHERE (( (article.search_vector) @@
(plainto_tsquery('pg_catalog.english', 'interesting'))) AND
article.magazine_id = 7) ORDER BY article.id ASC LIMIT 36
Time 13.4780406952 ms
QUERY PLAN
Limit (cost=2021.87..2021.96 rows=36 width=384) (actual time=9.782..9.787 rows=36 loops=1)
-> Sort (cost=2021.87..2027.49 rows=2248 width=384) (actual time=9.781..9.784 rows=36 loops=1)
Sort Key: id
Sort Method: top-N heapsort Memory: 53kB
-> Index Scan using idx_magazine_id on article (cost=0.29..1952.53 rows=2248 width=384) (actual time=0.035..8.924 rows=2249 loops=1)
Index Cond: (magazine_id = 7)
Filter: (search_vector @@ '''interesting'''::tsquery)
Rows Removed by Filter: 11413
Planning time: 4.600 ms
Execution time: 9.860 ms
然后,我发现甚至更不了解的是,它也拒绝在LIST页面上使用这个简单的btree索引来获取文章,其中它们只按降序列出x每页:
CREATE INDEX idx_btree_listing ON article USING btree(id DESC, magazine_id);
同样,它不使用多列索引:
Executed SQL
SELECT ••• FROM article WHERE article.magazine_id = 7
ORDER BY article.id DESC LIMIT 36
Time 1.4750957489 ms
QUERY PLAN
Limit (cost=0.29..7.48 rows=36 width=384) (actual time=0.034..0.115 rows=36 loops=1)
-> Index Scan Backward using idx_magazine_id on article (cost=0.29..2729.56 rows=13662 width=384) (actual time=0.031..0.107 rows=36 loops=1)
Filter: (magazine_id = 7)
Planning time: 1.354 ms
Execution time: 0.207 ms
编辑: 以上是一个开发设置,记录较少,只有1个杂志,因此速度快。以下是生产服务器上auto_explain生成的日志:
duration: 230.629 ms plan:
SELECT article.id, article.title, article.date, article.content FROM article WHERE article.magazine_id = 7 ORDER BY article.id DESC LIMIT 36
Limit (cost=0.42..43.67 rows=36 width=306) (actual time=229.876..229.995 rows=36 loops=1)
-> Index Scan Backward using idx_magazine_id on article (cost=0.42..239539.22 rows=199379 width=306) (actual time=229.866..229.968 rows=36 loops=1)
Filter: (article.magazine_id = 7)
Rows Removed by Filter: 116414
我将不胜感激,任何人都可以给我提供进一步调试的提示。
答案 0 :(得分:1)
多列索引中的第一列是id。你没有过滤id,所以postgres不会使用该索引。您不必过滤索引中的所有列,但过滤的列必须是索引中的前n列。
尝试尝试使用索引的变体,例如将id移到最后或从索引中省略id。