使用 2 列主键合并 table 中的记录

Question

我已经为我的问题准备了一个简单的测试用例 -

在游戏中，玩家 ID 和名称存储在 table users:

CREATE TABLE users (
        uid SERIAL PRIMARY KEY,
        name varchar(255) NOT NULL
);

并且玩家可以在 table reviews 中与 2 列 PK:

CREATE TABLE reviews (
        uid integer NOT NULL CHECK (uid <> author) REFERENCES users ON DELETE CASCADE,
        author integer NOT NULL REFERENCES users(uid) ON DELETE CASCADE,
        review varchar(255),
        PRIMARY KEY(uid, author)
);

这里两个 table 都填充了示例数据：

INSERT INTO users (uid, name) VALUES (1, 'User 1');
INSERT INTO users (uid, name) VALUES (2, 'User 2');
INSERT INTO users (uid, name) VALUES (3, 'User 3');
INSERT INTO users (uid, name) VALUES (4, 'User 4');

INSERT INTO reviews (uid, author, review) VALUES (1, 2, 'User 2 says: 1 is nice');
INSERT INTO reviews (uid, author, review) VALUES (1, 3, 'User 3 says: 1 is nice');
INSERT INTO reviews (uid, author, review) VALUES (1, 4, 'User 4 says: 1 is nice');

INSERT INTO reviews (uid, author, review) VALUES (2, 1, 'User 1 says: 2 is nice');
INSERT INTO reviews (uid, author, review) VALUES (2, 3, 'User 3 says: 2 is nice');
INSERT INTO reviews (uid, author, review) VALUES (2, 4, 'User 4 says: 2 is ugly');

INSERT INTO reviews (uid, author, review) VALUES (3, 1, 'User 1 says: 3 is nice');
INSERT INTO reviews (uid, author, review) VALUES (3, 2, 'User 2 says: 3 is ugly');
INSERT INTO reviews (uid, author, review) VALUES (3, 4, 'User 4 says: 3 is ugly');

INSERT INTO reviews (uid, author, review) VALUES (4, 1, 'User 1 says: 4 is ugly');
INSERT INTO reviews (uid, author, review) VALUES (4, 2, 'User 2 says: 4 is ugly');
INSERT INTO reviews (uid, author, review) VALUES (4, 3, 'User 3 says: 4 is ugly');

当我的移动应用程序注意到同一玩家正在使用多个用户 ID 时，它会将记录与如下所示的自定义存储函数合并。

合并（到 out_uid）时，用户对他自己的评论将被删除，任何由此产生的重叠评论也应删除。

(致合并记录的背景：这个真的很有必要，因为我运行另外一款游戏玩家评论多年，用户一直缠着我-为什么他们的评论和游戏当他们通过 Facebook、通过 Google+、通过 Apple Game Center 登录时，统计数据是不同的...)

由于没有UPDATE ... ON CONFLICT DO NOTHING - 我尝试在自定义存储函数中使用以下两个INSERT ... SELECT ... ON CONFLICT DO NOTHING来帮助自己：

CREATE OR REPLACE FUNCTION merge_users(
                in_uids integer[],
                OUT out_uid integer
        ) RETURNS integer AS
$func$
BEGIN
        SELECT
                MIN(uid)
        INTO STRICT
                out_uid 
        FROM users
        WHERE uid = ANY(in_uids);

        -- delete self-reviews
        DELETE FROM reviews
        WHERE uid = out_uid
        AND author = ANY(in_uids);

        DELETE FROM reviews
        WHERE author = out_uid
        AND uid = ANY(in_uids);

        -- try to copy as many reviews OF this user as possible
        INSERT INTO reviews (
                uid,
                author,
                review
        ) SELECT
                out_uid,        -- change to out_uid
                author,
                review
        FROM reviews
        WHERE uid <> out_uid
        AND uid = ANY(in_uids)
        ON CONFLICT DO NOTHING;

        DELETE FROM reviews
        WHERE uid <> out_uid
        AND uid = ANY(in_uids);

        -- try to copy as many reviews BY this user as possible
        INSERT INTO reviews (
                uid,
                author,
                review
        ) SELECT
                uid,
                out_uid,        -- change to out_uid
                review
        FROM reviews
        WHERE author <> out_uid
        AND author = ANY(in_uids)
        ON CONFLICT DO NOTHING;

        DELETE FROM reviews
        WHERE author <> out_uid
        AND author = ANY(in_uids);

        DELETE FROM users
        WHERE uid <> out_uid
        AND uid = ANY(in_uids);
END
$func$ LANGUAGE plpgsql;

不幸的是，有问题 - 请运行 2 个命令查看它们：

test=> SELECT out_uid FROM merge_users(ARRAY[1,2]);
 out_uid 
---------
       1
(1 row)

test=> SELECT out_uid FROM merge_users(ARRAY[1,2,3,4]);
ERROR:  new row for relation "reviews" violates check constraint "reviews_check"
DETAIL:  Failing row contains (1, 1, User 4 says: 3 is ugly).
CONTEXT:  SQL statement "INSERT INTO reviews (
                uid,
                author,
                review
        ) SELECT
                uid,
                out_uid,        -- change to out_uid
                review
        FROM reviews
        WHERE author <> out_uid
        AND author = ANY(in_uids)
        ON CONFLICT DO NOTHING"
PL/pgSQL function merge_users(integer[]) line 38 at SQL statement

所以删除自评好像不行，求助

另外我想知道是否有比我使用 INSERT ... SELECT ... ON CONFLICT DO NOTHING 的技巧更好的合并 reviews 记录的方法。

为了方便起见，我创建了一个 SQL Fiddle。

我也在非常有帮助的 pgsql-general 邮件列表中问过这个问题。

Answer 1

我想我会通过以下方式解决这个问题：

正在删除所有基于组合用户 ID 的自我评论。
将其余部分合并在一起。

我认为这是失败的第一部分。试试这个 delete:

DELETE FROM reviews
WHERE uid = ANY(in_uids) AND author = ANY(in_uids);

也就是老uids的任意组合都是一道题。我不确定 in_uids 是否包含所有等效的 uid，但我的想法是整个等效的 class 用于此目的。

使用 2 列主键合并 table 中的记录

Merging records in a table with 2-columns primary key

sql

postgresql

primary-key

plpgsql

postgresql-9.4