postgres - 避免创建重复的空列

postgres - avoid creating duplicate null columns

我在 Postgres 上有这个 table 模式:

> \d+ users_types_brands

                   Table "public.users_types_brands"
     Column     |            Type             | Collation | Nullable |                    Default                     | Storage | Stats target | Description 
----------------+-----------------------------+-----------+----------+------------------------------------------------+---------+--------------+-------------
 id             | integer                     |           | not null | nextval('users_types_brands_id_seq'::regclass) | plain   |              | 
 inserted_at    | timestamp without time zone |           |          | now()                                          | plain   |              | 
 updated_at     | timestamp without time zone |           |          | now()                                          | plain   |              | 
 users_types_id | bigint                      |           |          |                                                | plain   |              | 
 brand_id       | bigint                      |           | not null |                                                | plain   |              | 
 tasks_type_id  | integer                     |           |          |                                                | plain   |              | 
Indexes:
    "users_types_brands_pkey" PRIMARY KEY, btree (id)
    "users_types_brands_users_types_id_brand_id_tasks_type_id_index" UNIQUE, btree (users_types_id, brand_id, tasks_type_id)
Foreign-key constraints:
    "users_types_brands_users_types_id_fkey" FOREIGN KEY (users_types_id) REFERENCES users_types(id)
Access method: heap

现在 table 看起来像这样:

my_db=# select * from users_types_brands;
 id |        inserted_at         |         updated_at         | users_types_id | brand_id | tasks_type_id 
----+----------------------------+----------------------------+----------------+----------+---------------
 12 | 2021-10-24 16:43:12.244026 | 2021-10-24 16:43:12.244026 |              2 |      112 |             8
 14 | 2021-10-24 17:03:12.012874 | 2021-10-24 17:03:12.012874 |              2 |      111 |             9
(2 rows)
                   

当然,我不能像这样插入一行:

my_db=# insert into users_types_brands (users_types_id, brand_id, tasks_type_id) values (2, 112, 8);
ERROR:  duplicate key value violates unique constraint "users_types_brands_users_types_id_brand_id_tasks_type_id_index"
DETAIL:  Key (users_types_id, brand_id, tasks_type_id)=(2, 112, 8) already exists.

但我可以这样做几次:

my_db=# insert into users_types_brands (users_types_id, brand_id) values (2, 112);
INSERT 0 1

并得到这个:

my_db=# select * from users_types_brands;
 id |        inserted_at         |         updated_at         | users_types_id | brand_id | tasks_type_id 
----+----------------------------+----------------------------+----------------+----------+---------------
 12 | 2021-10-24 16:43:12.244026 | 2021-10-24 16:43:12.244026 |              2 |      112 |             8
 14 | 2021-10-24 17:03:12.012874 | 2021-10-24 17:03:12.012874 |              2 |      111 |             9
 16 | 2021-10-24 17:15:58.295428 | 2021-10-24 17:15:58.295428 |              2 |      112 |              
 17 | 2021-10-24 17:16:36.99971  | 2021-10-24 17:16:36.99971  |              2 |      112 |              
(4 rows)

现在,根据业务规则,tasks_type_id 可以为 null

但是我怎样才能避免创建像最后两行那样的重复行呢?一个 null tasks_type_id 可以,但不能有两个或更多。

有没有人遇到过这个问题?

您可以创建 Partial Unique Index. It will allow a single row with the same users_types_id and brand_id and null tasks_type_id, but only a single one. (See Demo)

create unique index tasks_type_id_just_1_unique
    on users_types_brands (users_types_id, brand_id)
  where tasks_type_id is null;

这个问题有两种基本的解决方案,但都有各自的缺点。

1.Using 部分索引,正如 Belayer 指出的那样。缺点是对于非空值,您将需要另一个部分索引,因为该部分索引将忽略非空值并仅覆盖具有空值的行。

CREATE UNIQUE INDEX "index_for_nulls" ON "table" ( "field_a", "field_b" ) WHERE "field_c" IS NULL;
CREATE UNIQUE INDEX "index_for_non_nulls" ON "table" ( "field_a", "field_b", "field_c" ) WHERE "field_c" IS NOT NULL;

2.Using 在索引定义中合并以避免空值。这样索引将覆盖所有行,但如果您不使用索引中定义的确切语句,规划器将不会使用完整索引

CREATE UNIQUE INDEX "index" ON "table" ( "field_a", "field_b", ( COALESCE( "field_c", -1 ) );