Postgres无限自我加入

Posted 2021-03-26

所以我有一篇文章和文章的“评论”..

评论允许人们回复..你可以回复答案..等等，这意味着最深的树根将是N

快速模拟表格的样子

Comments(id, news_id, user_id, body, likes)

Replies(id, parent_id) --> id here is = Comments.id

User(id, username, password)

News(id, title, body, image)

有没有办法查询Postgres DB给我一些类似的结果

因此，Replies表中具有null parent_id的任何内容都是“主要”注释（也就是不是回复）..如果children字段在其自身内填充（即回复的回复），我会很高兴

Postgres甚至可以实现这一点吗？或者我应该拿起所有Replies加入他们与Comments然后迭代通过每一个试图找到它适当的desitination？

顺便说一句，我正在使用GoLang作为我的后端和Gorm包来访问我的postgres数据库

编辑：我正在使用此查询

with recursive commentss as (
  select r.id, r.parent, array[r.id] as all_parents, 
         c.body, u.username 
    from replies r 
          inner join comments c 
                  on c.id = r.id 
                join users u 
                  on u.id = c.user_refer 
   where (parent <> '') IS NOT TRUE 
   union all 
  select r.id, r.parent, c.all_parents || r.id, 
         co.body, u.username 
    from replies r 
          join comments co 
            on co.id = r.id 
          join users u 
            on u.id = co.user_refer 
          join commentss c 
            on r.parent = c.id 
               and r.id <> ALL (c.all_parents)
  ) 
   select * from commentss order by all_parents;

结果是：

这是更接近的..然而我需要的是返回一个看起来像JSON对象

comments: [
  {
    comment_id: ...,
    username: ...,
    comment_body: ....,
    comment_likes: ....,
    children: [...]
  },
  {
    .....
  }
]

comments对象中的第一个项目将是不是回复的注释，而children字段应该填充回复的注释..并且children内的注释也应该填充children以回复该回复

答案

希望这是你的预期结果。 （我在这里做了类似的事情：https://stackoverflow.com/a/52076212/3984221）

demo: db<>fiddle

表comments：

id  body          user_id  likes  
--  ------------  -------  -----  
a   foo           1        1      
b   foofoo        1        232    
c   foofoofoo     1        23232  
d   fooFOO        1        53     
e   cookies       1        864    
f   bar           1        44     
g   barbar        1        54     
h   barBAR        1        222    
i   more cookies  1        1      

表replies

id  parent_id  
--  ---------  
a   (null)     
b   a          
c   b          
d   a          
e   (null)     
f   (null)     
g   f          
h   f          
i   (null)     

结果：

{
    "comments": [{
        "children": [],
        "username": "Mike Tyson",
        "comment_id": "i",
        "comment_body": "more cookies",
        "comment_likes": 1
    },
    {
        "children": [{
            "children": [],
            "username": "Mike Tyson",
            "comment_id": "b",
            "comment_body": "foofoo",
            "comment_likes": 232
        },
        {
            "children": [{
                "children": [],
                "username": "Mike Tyson",
                "comment_id": "c",
                "comment_body": "foofoofoo",
                "comment_likes": 23232
            }],
            "username": "Mike Tyson",
            "comment_id": "d",
            "comment_body": "fooFOO",
            "comment_likes": 53
        }],
        "username": "Mike Tyson",
        "comment_id": "a",
        "comment_body": "foo",
        "comment_likes": 1
    },
    {
        "children": [],
        "username": "Mike Tyson",
        "comment_id": "e",
        "comment_body": "cookies",
        "comment_likes": 864
    },
    {
        "children": [{
            "children": [],
            "username": "Mike Tyson",
            "comment_id": "g",
            "comment_body": "barbar",
            "comment_likes": 54
        },
        {
            "children": [],
            "username": "Mike Tyson",
            "comment_id": "h",
            "comment_body": "barBAR",
            "comment_likes": 222
        }],
        "username": "Mike Tyson",
        "comment_id": "f",
        "comment_body": "bar",
        "comment_likes": 44
    }]
}

查询：

递归：

WITH RECURSIVE parent_tree AS (
    SELECT 
        id, 
        NULL::text[] as parent_id,
        array_append('{comments}'::text[], (row_number() OVER ())::text) as path, 
        rc.children  
    FROM replies r
    LEFT JOIN LATERAL (SELECT parent_id, ARRAY_AGG(id) as children FROM replies WHERE parent_id = r.id GROUP BY parent_id) rc ON rc.parent_id = r.id
    WHERE r.parent_id IS NULL 

    UNION

    SELECT 
        r.id, 
        array_append(pt.parent_id, r.parent_id), 
        array_append(array_append(pt.path, 'children'), (row_number() OVER (PARTITION BY pt.parent_id))::text),
        rc.children      
    FROM parent_tree pt
    JOIN replies r ON r.id = ANY(pt.children)
    LEFT JOIN LATERAL (SELECT parent_id, ARRAY_AGG(id) as children FROM replies WHERE parent_id = r.id GROUP BY parent_id) rc ON rc.parent_id = r.id
), json_objects AS (
   SELECT c.id, jsonb_build_object('children', '[]'::jsonb, 'comment_id', c.id, 'username', u.name, 'comment_body', c.body, 'comment_likes', c.likes) as jsondata
   FROM comments c
   LEFT JOIN users u ON u.id = c.user_id
)
SELECT 
    parent_id, 
    path,
    jsondata
FROM parent_tree pt 
LEFT JOIN json_objects jo ON pt.id = jo.id
ORDER BY parent_id NULLS FIRST

唯一的递归部分是在CTE parent_tree内。在这里，我正在寻找父母并建立一条道路。稍后在正确的位置插入json数据需要此路径。

第二个CTE（json_objects）为每个注释构建一个json对象，其中包含一个空子数组，以便稍后插入子项。

LATERAL join在回复表中搜索当前id的子项，并给出一个带有id的数组。

最后的ORDER BY条款很重要。通过这种方式，可以确保所有上层节点都位于下层节点（它们的子节点）之前。否则，对全局json对象的输入可能会在以后失败，因为必要的父级在适当的时刻不能存在。

构建最终的JSON对象：

CREATE OR REPLACE FUNCTION json_tree() RETURNS jsonb AS $$
DECLARE
    _json_output jsonb;
    _temprow record;
BEGIN

    SELECT 
        jsonb_build_object('comments', '[]'::jsonb) 
    INTO _json_output;

    FOR _temprow IN
        -- <query above>
    LOOP
        SELECT jsonb_insert(_json_output, _temprow.path, _temprow.jsondata) INTO _json_output;
    END LOOP;

    RETURN _json_output;
END;
$$ LANGUAGE plpgsql;

无法在递归中构建json对象，因为在查询中jsondata对象不是全局变量。因此，如果我在一个递归分支中将b作为子项添加到a中，它将不存在于另一个分支中，我将c添加为子项。

因此有必要生成一个全局变量。这可以在一个函数中完成。使用计算出的路径和子对象，将最终的json组合在一起非常简单：循环遍历结果集并将json对象添加到全局对象的路径中。