主页 / user-118438

KadekM's questions

Martin Hope
KadekM
Asked: 2018-06-14 07:27:08 +0800 CST
  • 0

以下使用非确定性函数的查询会产生两个相同的行 - CTE 被计算并重新使用。

WITH foo AS (SELECT uuid_generate_v4() AS id)
SELECT id FROM foo
UNION ALL
SELECT id FROM foo

| id 
| 741f4f69-416b-4b4c-9226-559527f4a84e
| 741f4f69-416b-4b4c-9226-559527f4a84e

这个执行有保证吗？postgres 可以决定将它内联到

SELECT id FROM (SELECT uuid_generate_v4() AS id) t
UNION ALL
SELECT id FROM (SELECT uuid_generate_v4() AS id) t

| id 
| 262f0006-b4d5-440a-a86a-d79cd2684458
| c24b5835-9c5c-4d6f-b649-8e510129015b

这会产生两个不同的 ID（因此会产生不同的结果）？

SQL 标准规定了什么？Postgres 是否提供任何额外的保证？如果不能保证，重用依赖于非确定性过程的语句的标准方法是什么？

sql-standard postgresql
Martin Hope
KadekM
Asked: 2018-06-14 02:36:06 +0800 CST
  • 1

这显然会导致：

SELECT *, uuid_generate_v4() FROM fruits;

apple      |f6b9a8a5-31c9-4917-9649-f366b7c12492
orange     |a55da08f-95a0-4f33-b536-866a3eae71e9

我知道我能做到

SELECT * FROM fruits LEFT OUTER JOIN uuid_generate_v4() ON TRUE;

但是有没有更好的方法来分配相同的 uuid？我最感兴趣的是概念层面，uuid() 函数只是一个例子。

postgresql
Martin Hope
KadekM
Asked: 2017-11-15 01:08:02 +0800 CST
  • 1

基于Northwind架构：

假设我们有这样的查询（请忽略它的实际用处，因为涉及到更多实际用例）

SELECT unnest(r.agg) FROM (
  SELECT array_agg(x) AS agg
  FROM (
     SELECT *
     FROM northwind.u_suppliers s LEFT JOIN northwind.u_products p ON s.guid = p.supplier
     ) x
  GROUP BY x.supplier
  ) r LIMIT 1;

结果是

(9Y_03ogA1mOixVtJTqB8zih9Lu5-DB-l--fYvnHhyMQ,22,"Dirk Luchte","Accounting Manager",6j7qZ3Pjad5irKAM1MKX1WU7hvkgIOhZoe9OeRfTX4Q,"(12345) 1210",,northwindSupplier,jp9VvzFBbklHelD_ybS1oDsJbILda1duNhMtCdGBZqg,0,do4_k8zCsOog5tYROlsouF81Uj9I5LbTFawvuPw9Q-0,"2017-10-11 08:45:14.982",47,"Zaanse koeken",9Y_03ogA1mOixVtJTqB8zih9Lu5-DB-l--fYvnHhyMQ,AkOUoKYOpHFWIkNcoPX3CLtTl4n_ACN6TwWxqd3gzYQ,"10 - 4 oz boxes",9.5,36,0,0,f,,northwindProduct)

这是一个记录。

从中创建一个包含所有列和未嵌套记录作为行的表的方法是什么？

所以会有：

u_suppliers_column1 | ... | u_suppliers.columnN | u_products.column1 | ... | u_products.columnN
------------------------------------------------------------------------------------------------
9Y_03og...          | 22 | "Dirk Luchte" ...

该片段仅用于说明目的——当然，在这种情况下，创建聚合只是为了再次取消嵌套是不明智的，但为了简单和可重现性，我决定这样写。

postgresql
Martin Hope
KadekM
Asked: 2017-03-01 12:36:45 +0800 CST
  • 5

我有这样的查询：

SELECT t0.id
FROM platform_conversations t0
LEFT OUTER JOIN contacts t1 ON t1.id = t0.contact_id
WHERE t0.user_id = 5340
AND (
     t0.participant ILIKE '%baa%'  -- (1)
     OR t1.first_name ILIKE '%baa%' -- (2)
     )
LIMIT 50;

和

CREATE INDEX ix_conversations_participant
  ON platform_conversations USING GIN (participant gin_trgm_ops);

CREATE INDEX ix_trgm_contacts_search
  ON contacts USING GIN (first_name gin_trgm_ops);

并且无法弄清楚为什么索引不与OR条件一起使用。如果我只使用 (1)，或只使用 (2)，或AND改为使用，它们都会被使用。

这是计划：

仅供 (1)

Limit  (cost=12.43..203.68 rows=50 width=37) (actual time=0.037..0.037 rows=0 loops=1)
  ->  Bitmap Heap Scan on platform_conversations t0  (cost=12.43..222.80 rows=55 width=37) (actual time=0.030..0.030 rows=0 loops=1)
        Recheck Cond: ((participant)::text ~~* '%baa%'::text)
        Filter: (user_id = 5340)
        ->  Bitmap Index Scan on ix_conversations_participant  (cost=0.00..12.42 rows=55 width=0) (actual time=0.012..0.012 rows=0 loops=1)
              Index Cond: ((participant)::text ~~* '%baa%'::text)
Planning time: 0.397 ms
Execution time: 0.092 ms

仅供 (2)

Limit  (cost=16.71..446.06 rows=50 width=37) (actual time=0.034..0.034 rows=0 loops=1)
  ->  Nested Loop  (cost=16.71..471.82 rows=53 width=37) (actual time=0.028..0.028 rows=0 loops=1)
        ->  Bitmap Heap Scan on contacts t1  (cost=16.29..158.99 rows=37 width=37) (actual time=0.022..0.022 rows=0 loops=1)
              Recheck Cond: ((first_name)::text ~~* '%baa%'::text)
              ->  Bitmap Index Scan on ix_trgm_contacts_search  (cost=0.00..16.28 rows=37 width=0) (actual time=0.016..0.016 rows=0 loops=1)
                    Index Cond: ((first_name)::text ~~* '%baa%'::text)
        ->  Index Scan using ix_platform_conversations_contact_id on platform_conversations t0  (cost=0.42..8.45 rows=1 width=74) (never executed)
              Index Cond: ((contact_id)::text = (t1.id)::text)
              Filter: (user_id = 5340)
Planning time: 0.840 ms
Execution time: 0.121 ms

慢一个（用或）：

Limit  (cost=25771.43..47419.63 rows=50 width=37) (actual time=6652.292..6652.292 rows=0 loops=1)
  ->  Hash Left Join  (cost=25771.43..72531.55 rows=108 width=37) (actual time=6652.282..6652.282 rows=0 loops=1)
        Hash Cond: ((t0.contact_id)::text = (t1.id)::text)
        Filter: (((t0.participant)::text ~~* '%baa%'::text) OR ((t1.first_name)::text ~~* '%baa%'::text))
        Rows Removed by Filter: 553477
        ->  Seq Scan on platform_conversations t0  (cost=0.00..20627.29 rows=553512 width=87) (actual time=0.045..1741.899 rows=553477 loops=1)
              Filter: (user_id = 5340)
              Rows Removed by Filter: 386
        ->  Hash  (cost=17578.19..17578.19 rows=384819 width=46) (actual time=2372.508..2372.508 rows=384819 loops=1)
              Buckets: 65536  Batches: 16  Memory Usage: 2359kB
              ->  Seq Scan on contacts t1  (cost=0.00..17578.19 rows=384819 width=46) (actual time=0.014..1168.218 rows=384819 loops=1)
Planning time: 0.741 ms
Execution time: 6652.333 ms

它不能只对 t1、t2 进行位图索引扫描，然后对它们进行位图 OR 吗？为什么 OR 这样的问题？

Postgres 9.6（无法创建标签，没有足够的代表）

postgresql postgresql-9.6