提高我的 psql 查询性能
Improve my psql query performance
我有一个执行奇数的 psql 查询。我在现在用于搜索的 2 列上定义了一个 GIN 索引：
Indexes:
"pk_products" PRIMARY KEY, btree (id)
"fk_affiliate_affiliate_product_id" UNIQUE, btree (affiliate_id, affiliate_product_id)
"idx_products" btree (merchant_id)
"idx_products_affiliates" btree (affiliate_id)
"idx_products_brand_id" btree (brand_id)
"idx_products_ts" gin (to_tsvector('english'::regconfig, (COALESCE(title, ''::character varying)::text || ' '::text) || COALESCE(description, ''::text)))
如果我搜索一个短词，比如 4 个字符，我会得到一个快速查询：
EXPLAIN ANALYZE SELECT p.id, p.price, p.currency, p.images, p.merchant_id
FROM products AS p
WHERE deleted=false AND to_tsvector('english', p.title || coalesce(p.description, '')) @@ to_tsquery('blue:*')
LIMIT 30 OFFSET 0;
结果：
查询计划
Limit (cost=0.00..219.47 rows=30 width=49) (actual time=3.138..40.914 rows=30 loops=1)
-> Seq Scan on products p (cost=0.00..41120.86 rows=5621 width=49) (actual time=2.740..40.478 rows=30 loops=1)
Filter: ((NOT deleted) AND (to_tsvector('english'::regconfig, ((title)::text || COALESCE(description, ''::text))) @@ to_tsquery('blue:*'::text)))
Rows Removed by Filter: 153
Total runtime: 40.986 ms
(5 rows)
如果我使用更长的单词：
EXPLAIN ANALYZE SELECT p.id, p.price, p.currency, p.images, p.merchant_id
FROM products AS p
WHERE deleted=false AND to_tsvector('english', p.title || coalesce(p.description, '')) @@ to_tsquery('turquoise:*')
LIMIT 30 OFFSET 0;
时间增加：
查询计划
Limit (cost=0.00..219.47 rows=30 width=49) (actual time=1.097..1579.187 rows=30 loops=1)
-> Seq Scan on products p (cost=0.00..41120.86 rows=5621 width=49) (actual time=1.093..1579.129 rows=30 loops=1)
Filter: ((NOT deleted) AND (to_tsvector('english'::regconfig, ((title)::text || COALESCE(description, ''::text))) @@ to_tsquery('turquoise:*'::text)))
Rows Removed by Filter: 12697
Total runtime: 1579.287 ms
(5 rows)
如果我在单词中使用“-”，那么得到结果的时间会很长：
EXPLAIN ANALYZE SELECT p.id, p.price, p.currency, p.images, p.merchant_id
FROM products AS p
WHERE deleted=false AND to_tsvector('english', p.title || coalesce(p.description, '')) @@ to_tsquery('turquoise-blue:*')
LIMIT 30 OFFSET 0;
结果：
查询计划
Limit (cost=0.00..41120.86 rows=2 width=49) (actual time=31400.164..31400.164 rows=0 loops=1)
-> Seq Scan on products p (cost=0.00..41120.86 rows=2 width=49) (actual time=31400.158..31400.158 rows=0 loops=1)
Filter: ((NOT deleted) AND (to_tsvector('english'::regconfig, ((title)::text || COALESCE(description, ''::text))) @@ to_tsquery('turquoise-blue:*'::text)))
Rows Removed by Filter: 281510
Total runtime: 31400.247 ms
(5 rows)
非常感谢任何想法！谢谢！
编辑：
我觉得跟结果的数量有关，没有结果的查询耗时更长？
找到问题了。
像这样更改了索引：
"idx_products_ts" gin (to_tsvector('english'::regconfig, title::text || description))
现在查询速度真的很快！
EXPLAIN ANALYZE SELECT p.id, p.price, p.currency, p.images, p.merchant_id
FROM products AS p
WHERE deleted=false AND to_tsvector('english', p.title || p.description) @@ to_tsquery('blue:*')
LIMIT 30 OFFSET 0;
查询计划
Limit (cost=103.64..183.47 rows=30 width=49) (actual time=15.588..15.644 rows=30 loops=1)
-> Bitmap Heap Scan on products p (cost=103.64..15084.42 rows=5630 width=49) (actual time=15.586..15.633 rows=30 loops=1)
Recheck Cond: (to_tsvector('english'::regconfig, ((title)::text || description)) @@ to_tsquery('blue:*'::text))
Filter: (NOT deleted)
-> Bitmap Index Scan on idx_products_ts (cost=0.00..102.23 rows=5630 width=0) (actual time=12.955..12.955 rows=26747 loops=1)
Index Cond: (to_tsvector('english'::regconfig, ((title)::text || description)) @@ to_tsquery('blue:*'::text))
Total runtime: 15.714 ms
(7 rows)
并且：
EXPLAIN ANALYZE SELECT p.id, p.price, p.currency, p.images, p.merchant_id
FROM products AS p
WHERE deleted=false AND to_tsvector('english', p.title || p.description) @@ to_tsquery('turquoise-blue:*')
LIMIT 30 OFFSET 0;
查询计划
Limit (cost=108.02..116.02 rows=2 width=49) (actual time=26.234..26.234 rows=0 loops=1)
-> Bitmap Heap Scan on products p (cost=108.02..116.02 rows=2 width=49) (actual time=26.226..26.226 rows=0 loops=1)
Recheck Cond: (to_tsvector('english'::regconfig, ((title)::text || description)) @@ to_tsquery('turquoise-blue:*'::text))
Filter: (NOT deleted)
-> Bitmap Index Scan on idx_products_ts (cost=0.00..108.02 rows=2 width=0) (actual time=26.209..26.209 rows=0 loops=1)
Index Cond: (to_tsvector('english'::regconfig, ((title)::text || description)) @@ to_tsquery('turquoise-blue:*'::text))
Total runtime: 26.433 ms
(7 rows)
从总运行时间：31400.247 毫秒到总运行时间：26.433 毫秒的巨大改进。
我有一个执行奇数的 psql 查询。我在现在用于搜索的 2 列上定义了一个 GIN 索引：
Indexes:
"pk_products" PRIMARY KEY, btree (id)
"fk_affiliate_affiliate_product_id" UNIQUE, btree (affiliate_id, affiliate_product_id)
"idx_products" btree (merchant_id)
"idx_products_affiliates" btree (affiliate_id)
"idx_products_brand_id" btree (brand_id)
"idx_products_ts" gin (to_tsvector('english'::regconfig, (COALESCE(title, ''::character varying)::text || ' '::text) || COALESCE(description, ''::text)))
如果我搜索一个短词，比如 4 个字符，我会得到一个快速查询：
EXPLAIN ANALYZE SELECT p.id, p.price, p.currency, p.images, p.merchant_id
FROM products AS p
WHERE deleted=false AND to_tsvector('english', p.title || coalesce(p.description, '')) @@ to_tsquery('blue:*')
LIMIT 30 OFFSET 0;
结果：
查询计划
Limit (cost=0.00..219.47 rows=30 width=49) (actual time=3.138..40.914 rows=30 loops=1)
-> Seq Scan on products p (cost=0.00..41120.86 rows=5621 width=49) (actual time=2.740..40.478 rows=30 loops=1)
Filter: ((NOT deleted) AND (to_tsvector('english'::regconfig, ((title)::text || COALESCE(description, ''::text))) @@ to_tsquery('blue:*'::text)))
Rows Removed by Filter: 153
Total runtime: 40.986 ms
(5 rows)
如果我使用更长的单词：
EXPLAIN ANALYZE SELECT p.id, p.price, p.currency, p.images, p.merchant_id
FROM products AS p
WHERE deleted=false AND to_tsvector('english', p.title || coalesce(p.description, '')) @@ to_tsquery('turquoise:*')
LIMIT 30 OFFSET 0;
时间增加：
查询计划
Limit (cost=0.00..219.47 rows=30 width=49) (actual time=1.097..1579.187 rows=30 loops=1)
-> Seq Scan on products p (cost=0.00..41120.86 rows=5621 width=49) (actual time=1.093..1579.129 rows=30 loops=1)
Filter: ((NOT deleted) AND (to_tsvector('english'::regconfig, ((title)::text || COALESCE(description, ''::text))) @@ to_tsquery('turquoise:*'::text)))
Rows Removed by Filter: 12697
Total runtime: 1579.287 ms
(5 rows)
如果我在单词中使用“-”，那么得到结果的时间会很长：
EXPLAIN ANALYZE SELECT p.id, p.price, p.currency, p.images, p.merchant_id
FROM products AS p
WHERE deleted=false AND to_tsvector('english', p.title || coalesce(p.description, '')) @@ to_tsquery('turquoise-blue:*')
LIMIT 30 OFFSET 0;
结果：
查询计划
Limit (cost=0.00..41120.86 rows=2 width=49) (actual time=31400.164..31400.164 rows=0 loops=1)
-> Seq Scan on products p (cost=0.00..41120.86 rows=2 width=49) (actual time=31400.158..31400.158 rows=0 loops=1)
Filter: ((NOT deleted) AND (to_tsvector('english'::regconfig, ((title)::text || COALESCE(description, ''::text))) @@ to_tsquery('turquoise-blue:*'::text)))
Rows Removed by Filter: 281510
Total runtime: 31400.247 ms
(5 rows)
非常感谢任何想法！谢谢！
编辑：
我觉得跟结果的数量有关，没有结果的查询耗时更长？
找到问题了。
像这样更改了索引：
"idx_products_ts" gin (to_tsvector('english'::regconfig, title::text || description))
现在查询速度真的很快！
EXPLAIN ANALYZE SELECT p.id, p.price, p.currency, p.images, p.merchant_id
FROM products AS p
WHERE deleted=false AND to_tsvector('english', p.title || p.description) @@ to_tsquery('blue:*')
LIMIT 30 OFFSET 0;
查询计划
Limit (cost=103.64..183.47 rows=30 width=49) (actual time=15.588..15.644 rows=30 loops=1)
-> Bitmap Heap Scan on products p (cost=103.64..15084.42 rows=5630 width=49) (actual time=15.586..15.633 rows=30 loops=1)
Recheck Cond: (to_tsvector('english'::regconfig, ((title)::text || description)) @@ to_tsquery('blue:*'::text))
Filter: (NOT deleted)
-> Bitmap Index Scan on idx_products_ts (cost=0.00..102.23 rows=5630 width=0) (actual time=12.955..12.955 rows=26747 loops=1)
Index Cond: (to_tsvector('english'::regconfig, ((title)::text || description)) @@ to_tsquery('blue:*'::text))
Total runtime: 15.714 ms
(7 rows)
并且：
EXPLAIN ANALYZE SELECT p.id, p.price, p.currency, p.images, p.merchant_id
FROM products AS p
WHERE deleted=false AND to_tsvector('english', p.title || p.description) @@ to_tsquery('turquoise-blue:*')
LIMIT 30 OFFSET 0;
查询计划
Limit (cost=108.02..116.02 rows=2 width=49) (actual time=26.234..26.234 rows=0 loops=1)
-> Bitmap Heap Scan on products p (cost=108.02..116.02 rows=2 width=49) (actual time=26.226..26.226 rows=0 loops=1)
Recheck Cond: (to_tsvector('english'::regconfig, ((title)::text || description)) @@ to_tsquery('turquoise-blue:*'::text))
Filter: (NOT deleted)
-> Bitmap Index Scan on idx_products_ts (cost=0.00..108.02 rows=2 width=0) (actual time=26.209..26.209 rows=0 loops=1)
Index Cond: (to_tsvector('english'::regconfig, ((title)::text || description)) @@ to_tsquery('turquoise-blue:*'::text))
Total runtime: 26.433 ms
(7 rows)
从总运行时间：31400.247 毫秒到总运行时间：26.433 毫秒的巨大改进。