选择孤立行（以通用方式）

Question

我有一个表my_tables，它在几个表中被引用为外键。

我想选择my_table其他表中未引用的所有行。

AFAIK，应该可以以通用方式（带有一些内省魔法）来做到这一点。

Answer 1

引用手册：

外键必须引用作为主键或形成唯一约束的列。

所以这不一定限于PK。但是如果我们从 开始pg_constraint，我们会自动获得指向目标表的所有 FK 约束。无需提供任何键列 - 除非您想限制为某些 FK。

使用对象标识符类型regclass和表别名，我们可以保持函数简短，结果安全无歧义：

基本查询

SELECT format(E'SELECT * FROM %s t\nWHERE  NOT EXISTS (SELECT FROM %s'
            , c.confrelid::regclass
            , string_agg(format('%s WHERE %s = %s)', c.conrelid::regclass, src.cols, tgt.cols)
                       , E'\nAND    NOT EXISTS (SELECT FROM '))
FROM   pg_constraint c
     , cardinality(c.conkey) AS col_ct
     , LATERAL (
   SELECT concat(CASE WHEN col_ct > 1 THEN '(' END
               , string_agg(quote_ident(attname), ', ' ORDER BY fld.ord) -- original order
               , CASE WHEN col_ct > 1 THEN ')' END) AS cols
   FROM   unnest(c.conkey) WITH ORDINALITY fld(attnum, ord)             -- possibly n cols
   JOIN   pg_catalog.pg_attribute a ON (a.attrelid, a.attnum) = (c.conrelid, fld.attnum)
   ) src
     , LATERAL (
   SELECT concat(CASE WHEN col_ct > 1 THEN '(' END     -- parentheses for multiple columns
               , string_agg('t.' || quote_ident(attname), ', t.' ORDER BY fld.ord)
               , CASE WHEN col_ct > 1 THEN ')' END) AS cols
   FROM   unnest(c.confkey) WITH ORDINALITY fld(attnum, ord)
   JOIN   pg_catalog.pg_attribute a ON (a.attrelid, a.attnum) = (c.confrelid, fld.attnum)
   ) tgt
WHERE  c.confrelid = 'my_table'::regclass -- target table name, optionally schema-qualified
AND    c.contype = 'f'  -- FK constraints
GROUP  BY c.confrelid;

生成以下形式的查询：

SELECT * FROM my_table t
WHERE  NOT EXISTS (SELECT FROM schema1.tbl1 WHERE col1 = t.id)
AND    NOT EXISTS (SELECT FROM "tB-l2" WHERE ("COL2", col3) = (t.col4, t.col5));

它返回当前未被任何FK 约束引用的所有行。

如果cardinality(c.conkey) > 1那样的话，也可以安全地假设cardinality(c.confkey) > 1. 所以只计算一次来决定是否加括号。

全自动化

要动态地对任何输入表进行此操作，请创建一个采用表的行值的多态函数：

CREATE OR REPLACE FUNCTION f_orphans(_tbl anyelement)
  RETURNS SETOF anyelement AS
$func$
BEGIN

RETURN QUERY EXECUTE (  -- exactly the query from above
SELECT format(E'SELECT * FROM %s t\nWHERE  NOT EXISTS (SELECT FROM %s'
            , c.confrelid::regclass
            , string_agg(format('%s WHERE %s = %s)', c.conrelid::regclass, src.cols, tgt.cols)
                       , E'\nAND    NOT EXISTS (SELECT FROM '))
FROM   pg_constraint c
     , cardinality(c.conkey) AS col_ct
     , LATERAL (
   SELECT concat(CASE WHEN col_ct > 1 THEN '(' END
               , string_agg(quote_ident(attname), ', ' ORDER BY fld.ord)
               , CASE WHEN col_ct > 1 THEN ')' END) AS cols
   FROM   unnest(c.conkey) WITH ORDINALITY fld(attnum, ord)
   JOIN   pg_catalog.pg_attribute a ON (a.attrelid, a.attnum) = (c.conrelid, fld.attnum)
   ) src
     , LATERAL (
   SELECT concat(CASE WHEN col_ct > 1 THEN '(' END
               , string_agg('t.' || quote_ident(attname), ', t.' ORDER BY fld.ord)
               , CASE WHEN col_ct > 1 THEN ')' END) AS cols
   FROM   unnest(c.confkey) WITH ORDINALITY fld(attnum, ord)
   JOIN   pg_catalog.pg_attribute a ON (a.attrelid, a.attnum) = (c.confrelid, fld.attnum)
   ) tgt
WHERE  c.confrelid = pg_typeof(_tbl)::text::regclass  -- input goes here!
AND    c.contype = 'f'
GROUP  BY c.confrelid
);

END
$func$  LANGUAGE plpgsql;

打电话（重要）！

SELECT * FROM f_orphans(NULL::my_table);

或者：

SELECT * FROM f_orphans(NULL::myschema.my_table);

有关的：

• PostgreSQL：将表作为函数中的参数传递
• 检索所有 PK 和 FK
• SQL order by query 结果以任何任意（但可重现）的方式
• 如何保留未嵌套数组中元素的原始顺序？