帮助使用 INSERT、ON CONFLICT 和 RETURNING 进行递归 CTE 查询

Kev*_*son 5 postgresql postgresql-9.5

我正在尝试编写查询以将文件名插入下表 path

CREATE SEQUENCE path_id_seq;
CREATE TABLE path (
    id INTEGER PRIMARY KEY DEFAULT NEXTVAL('path_id_seq'),
    parent_id INTEGER,
    name TEXT NOT NULL);

CREATE UNIQUE INDEX path_parent_id_name_ix
    ON path (COALESCE(parent_id, 0), name);

ALTER TABLE path
    ADD CONSTRAINT ios_path_parent_id_fk
    FOREIGN KEY (parent_id) REFERENCES path (id)
    ON DELETE CASCADE
    DEFERRABLE INITIALLY DEFERRED;
Run Code Online (Sandbox Code Playgroud)

例如,给定文件名Assets/images/a.pngAssets/images/b.png我想插入以下行。

id | parent_id |       name        
----+-----------+-------------------
  1 |         ¤ | Assets
  2 |         1 | images
  3 |         2 | a.png
  4 |         2 | b.png
Run Code Online (Sandbox Code Playgroud)

插入父对象很容易;

INSERT INTO path (name)
VALUES ('Assets')
ON CONFLICT (COALESCE(parent_id, 0), name)
DO UPDATE SET name = EXCLUDED.name -- this strange update is so that we get rows back
RETURNING id, parent_id, name;
Run Code Online (Sandbox Code Playgroud)

产生;

 id | parent_id |  name  
----+-----------+--------
  1 |         ¤ | Assets
Run Code Online (Sandbox Code Playgroud)

但是我似乎无法将递归 CTE 放在一起以在下一个 INSERT 语句中使用父代的 id 作为 parent_id。

它可能看起来像下面这个无效的查询。

WITH RECURSIVE names AS (
    SELECT name, lag(name, 1) OVER () as previous_name
    FROM (SELECT unnest(string_to_array('Assets/images/232.png', '/')) as name) names),

paths(id, name) AS (
    INSERT INTO path (name)
    SELECT name
    FROM names
    WHERE previous_name IS NULL
    ON CONFLICT (COALESCE(parent_id, 0), name)
    DO UPDATE SET name = EXCLUDED.name
    RETURNING id, name
    UNION
    INSERT INTO path (parent_id, name)
    SELECT paths.id, names.name
    FROM paths
    JOIN names ON names.previous_name = paths.name
    ON CONFLICT (COALESCE(parent_id, 0), name)
    DO UPDATE SET name = EXCLUDED.name
    RETURNING id, name)

SELECT *
FROM paths;
Run Code Online (Sandbox Code Playgroud)

但我似乎无法让它发挥作用。如果有人看到一种更简单的方法来做到这一点,我们将不胜感激。

小智 4

您不需要递归 CTE,您可以直接nextval()在结果中使用unnest(),然后使用lag()从上一行获取值。为了处理现有行,我将使用 unnest() 的结果和路径表之间的联接来执行此操作,检测新行和现有行。

对于新行,将生成一个新 ID,并且仅插入这些 ID:

insert into path (id, name, parent_id)
select id, name, parent_id
from (
  select *,
         case 
           when is_new
             then lag(id) over (order by level) 
             else existing_parent
         end as parent_id
  from (
    SELECT case 
              when p.id is null then nextval('path_id_seq') 
              else p.id
           end as id, 
           x.name, x.level, p.parent_id as existing_parent, 
           p.id is null as is_new        
    FROM unnest(string_to_array('Assets/images/232.png', '/')) with ordinality as x(name, level)
      LEFT JOIN path p on p.name = x.name
    order by level
  ) t1
) t2
where is_new;
Run Code Online (Sandbox Code Playgroud)

实例: http: //rextester.com/YGMT19565

如果您愿意的话,唯一的“缺点”是序列值不一定按照您期望的“顺序”分配。但由于生成的主键的唯一目的是唯一(而不是其他),这不应该成为问题。