使用 row_to_json 的 Postgres 递归查询

Posted

技术标签:

【中文标题】使用 row_to_json 的 Postgres 递归查询【英文标题】:Postgres recursive query with row_to_json 【发布时间】:2014-10-29 23:20:04 【问题描述】:

我在 postgres 9.3.5 中有一个如下所示的表:

CREATE TABLE customer_area_node
(
  id bigserial NOT NULL,
  customer_id integer NOT NULL,
  parent_id bigint,
  name text,
  description text,

  CONSTRAINT customer_area_node_pkey PRIMARY KEY (id)
)

我查询:

WITH RECURSIVE c AS (
       SELECT *, 0 as level, name as path FROM customer_area_node WHERE customer_id = 2 and parent_id is null
       UNION ALL
       SELECT customer_area_node.*, 
       c.level + 1 as level, 
       c.path || '/' || customer_area_node.name as path
  FROM customer_area_node 
  join c ON customer_area_node.parent_id = c.id
)
SELECT * FROM c ORDER BY path;

这似乎适用于构建诸如 building1/floor1/room1、building1/floor1/room2 等路径。

我希望能够轻松地将其转换为代表树结构的 json,我被告知我可以使用 row_to_json 来完成。

作为一种合理的选择,我可以通过任何其他方式将数据格式化为更有效的机制,这样我实际上可以轻松地将其转换为实际的树结构,而无需在 / 上添加大量 string.splits。

是否有一种相当简单的方法可以使用 row_to_json 做到这一点?

【问题讨论】:

你能提供样本数据吗? 【参考方案1】:

普通递归 CTE 无法做到这一点,因为几乎不可能在其层次结构的深处设置 json 值。但是你可以反过来做:从叶子开始构建树,直到它的根:

-- calculate node levels
WITH RECURSIVE c AS (
    SELECT *, 0 as lvl
    FROM customer_area_node
    -- use parameters here, to select the root first
    WHERE customer_id = 2 AND parent_id IS NULL
  UNION ALL
    SELECT customer_area_node.*, c.lvl + 1 as lvl
    FROM customer_area_node 
    JOIN c ON customer_area_node.parent_id = c.id
),
-- select max level
maxlvl AS (
  SELECT max(lvl) maxlvl FROM c
),
-- accumulate children
j AS (
    SELECT c.*, json '[]' children -- at max level, there are only leaves
    FROM c, maxlvl
    WHERE lvl = maxlvl
  UNION ALL
    -- a little hack, because PostgreSQL doesn't like aggregated recursive terms
    SELECT (c).*, array_to_json(array_agg(j)) children
    FROM (
      SELECT c, j
      FROM j
      JOIN c ON j.parent_id = c.id
    ) v
    GROUP BY v.c
)
-- select only root
SELECT row_to_json(j) json_tree
FROM j
WHERE lvl = 0;

这甚至适用于 PostgreSQL 9.2+

SQLFiddle

更新:一个变体,它也应该处理流氓叶节点(位于 1 和 max-level 之间的级别):

WITH RECURSIVE c AS (
    SELECT *, 0 as lvl
    FROM   customer_area_node
    WHERE  customer_id = 1 AND parent_id IS NULL
  UNION ALL
    SELECT customer_area_node.*, c.lvl + 1
    FROM   customer_area_node 
    JOIN   c ON customer_area_node.parent_id = c.id
),
maxlvl AS (
  SELECT max(lvl) maxlvl FROM c
),
j AS (
    SELECT c.*, json '[]' children
    FROM   c, maxlvl
    WHERE  lvl = maxlvl
  UNION ALL
    SELECT   (c).*, array_to_json(array_agg(j) || array(SELECT r
                                                        FROM   (SELECT l.*, json '[]' children
                                                                FROM   c l, maxlvl
                                                                WHERE  l.parent_id = (c).id
                                                                AND    l.lvl < maxlvl
                                                                AND    NOT EXISTS (SELECT 1
                                                                                   FROM   c lp
                                                                                   WHERE  lp.parent_id = l.id)) r)) children
    FROM     (SELECT c, j
              FROM   c
              JOIN   j ON j.parent_id = c.id) v
    GROUP BY v.c
)
SELECT row_to_json(j) json_tree
FROM   j
WHERE  lvl = 0;

这个应该在 PostgreSQL 9.2+ 上也可以工作,但是,我无法测试。 (我现在只能在 9.5+ 上测试)。

这些解决方案可以处理任何分层表中的任何列,但总是将int 类型为lvl JSON 属性附加到它们的输出中。

http://rextester.com/YNU7932

【讨论】:

我发现所有叶子都必须有相同的水平才能工作。 @Macario 在 SQLFiddle 示例中,有多个级别的叶子(例如 1.3.71.3.6.9)并收集所有节点。 @Macario 我遇到了与@pozs 报告的相同问题。我用1.4.10 创建了一个示例SQLFiddle,显示了这种情况,它只有2 个叶子,而不是像其他所有叶子一样的3 个叶子。因此,使用此 SQL,您必须使所有分支的深度完全相同。 抱歉迟到了,但我也可以找到解决这些叶节点的方法。【参考方案2】:

很抱歉,答案很晚,但我想我找到了一个优雅的解决方案,可以成为这个问题的公认答案。

基于@pozs 发现的令人敬畏的“小技巧”,我想出了一个解决方案:

用很少的代码解决“流氓离开”的情况(利用 NOT EXISTS 谓词) 避免整个级别计算/条件的东西
WITH RECURSIVE customer_area_tree("id", "customer_id", "parent_id", "name", "description", "children") AS (
  -- tree leaves (no matching children)
  SELECT c.*, json '[]'
  FROM customer_area_node c
  WHERE NOT EXISTS(SELECT * FROM customer_area_node AS hypothetic_child WHERE hypothetic_child.parent_id = c.id)

  UNION ALL

  -- pozs's awesome "little hack"
  SELECT (parent).*, json_agg(child) AS "children"
  FROM (
    SELECT parent, child
    FROM customer_area_tree AS child
    JOIN customer_area_node parent ON parent.id = child.parent_id
  ) branch
  GROUP BY branch.parent
)
SELECT json_agg(t)
FROM customer_area_tree t
LEFT JOIN customer_area_node AS hypothetic_parent ON(hypothetic_parent.id = t.parent_id)
WHERE hypothetic_parent.id IS NULL

更新

Tested with very simple data,它确实有效,但正如 posz 在评论中指出的那样,with his sample data,一些流氓叶子节点被遗忘了。但是,我发现with even more complex data,之前的答案也不起作用,因为只有具有“最大级别”叶节点的共同祖先的流氓叶节点被捕获(当“1.2.5.8”不存在时,“1.2. 4”和“1.2.5”不存在,因为它们没有任何“最大级别”叶节点的共同祖先)。

所以这是一个新的提议,通过提取NOT EXISTS 子请求并使其成为内部UNION,利用UNION 重复数据删除能力(利用 jsonb 比较能力),将 posz 的工作与我的工作混合:

<!-- language: sql -->
WITH RECURSIVE
c_with_level AS (

    SELECT *, 0 as lvl
    FROM   customer_area_node
    WHERE  parent_id IS NULL

    UNION ALL

    SELECT child.*, parent.lvl + 1
    FROM   customer_area_node child
    JOIN   c_with_level parent ON parent.id = child.parent_id
),
maxlvl AS (
  SELECT max(lvl) maxlvl FROM c_with_level
),
c_tree AS (
    SELECT c_with_level.*, jsonb '[]' children
    FROM   c_with_level, maxlvl
    WHERE  lvl = maxlvl

    UNION 
    (
        SELECT (branch_parent).*, jsonb_agg(branch_child)
        FROM (
            SELECT branch_parent, branch_child
            FROM c_with_level branch_parent
            JOIN c_tree branch_child ON branch_child.parent_id = branch_parent.id
        ) branch
        GROUP BY branch.branch_parent

        UNION

        SELECT c.*, jsonb '[]' children
        FROM   c_with_level c
        WHERE  NOT EXISTS (SELECT 1 FROM c_with_level hypothetical_child WHERE hypothetical_child.parent_id = c.id)
    )
)
SELECT jsonb_pretty(row_to_json(c_tree)::jsonb)
FROM c_tree
WHERE lvl = 0;

在http://rextester.com/SMM38494 上测试;)

【讨论】:

I'm afraid,如果不对级别进行特殊处理,您最终会在输出中出现许多单独的“分支”,因为不同级别中存在许多叶子。您的示例数据对于每个节点只有 0 或 1 个子节点,这就是它没有显示的原因。 感谢您的观察!我一直在做一些工作,我发现你处理流氓叶节点的技巧也不是很有效,因为在你的例子中,如果你没有节点“1.2.5.8”,分支“1.2 " 永远不会被捕获,因此最终结果中不存在 "1.2.4" 和 "1.2.5"。这是因为只有当恶意叶节点具有“最大级别”叶节点的共同祖先时,您才会捕获它们。我找到了解决方案,我会在一分钟内编辑我的答案;) 对于任何感兴趣的人,我将 David 的第二次尝试修改为 return nested JSON objects。在 Python 中,我想将 JSON 反序列化为嵌套字典 - 为什么在代码中执行 SQL 中可以完成的操作;)我也希望对 javascript 中的数据结构进行类似的访问。谢谢@DavidGuillot! 这种方法很好,但不包括存在多个根项且一棵树比另一棵树更深的情况。它在根级别生成具有不同子级的重复项:sqlfiddle.com/#!17/022f80/8【参考方案3】:

进一步开发了 pozs 的答案,以获得带有子树的递归叶子。所以这个答案真的返回了完整的树。

CREATE OR REPLACE FUNCTION pg_temp.getTree(bigint) 
    RETURNS TABLE( 
            id bigint,
            customer_id integer,
            parent_id bigint,
            name text,
            description text,
            children json
        ) 
        AS $$   

        WITH RECURSIVE relations AS ( 
            SELECT 
                can.id,
                can.customer_id,
                can.parent_id,
                can.name, 
                can.description,
                0 AS depth 
                FROM customer_area_node can 
                WHERE can.id = $1 
            UNION ALL 
            SELECT 
                can.id,
                can.customer_id,
                can.parent_id,
                can.name, 
                can.description,
                relations.depth + 1 
                FROM customer_area_node can
                JOIN relations ON can.parent_id = relations.id AND can.id != can.parent_id
        ),     

        maxdepth AS ( 
            SELECT max(depth) maxdepth FROM relations 
        ), 

        rootTree as ( 
            SELECT r.* FROM 
                relations r, maxdepth 
                WHERE depth = maxdepth 
            UNION ALL 
            SELECT r.* FROM 
                relations r, rootTree 
                WHERE r.id = rootTree.parent_id AND rootTree.id != rootTree.parent_id 
        ), 

        mainTree AS ( 
            SELECT 
                c.id,
                c.customer_id,
                c.parent_id,
                c.name, 
                c.description,
                c.depth, 
                json_build_array() children 
                FROM relations c, maxdepth 
                WHERE c.depth = maxdepth 
            UNION ALL 
            SELECT 
                (relations).*, 
                array_to_json( 
                    array_agg(mainTree) 
                    || 
                    array( 
                        SELECT t 
                            FROM ( 
                                SELECT 
                                    l.*, 
                                    json_build_array() children 
                                FROM relations l, maxdepth 
                                    WHERE l.parent_id = (relations).id 
                                    AND l.depth < maxdepth 
                                    AND l.id  NOT IN ( 
                                        SELECT id FROM rootTree 
                                    ) 
                            ) r 
                           JOIN pg_temp.getTree(r.id) t 
                            ON r.id = t.id 
                        )) 
                children 
    FROM ( 
        SELECT relations, mainTree 
            FROM relations 
        JOIN mainTree 
            ON ( 
                mainTree.parent_id = relations.id 
                AND mainTree.parent_id != mainTree.id 
            ) 
    ) v 
    GROUP BY v.relations 
    ) 

        SELECT 
            id,
            customer_id,
            parent_id,
            name, 
            description,
            children 
        FROM mainTree WHERE id = $1 
    $$ 
    LANGUAGE SQL; 

    SELECT * 
    FROM 
        customer_area_node can 
        JOIN pg_temp.getTree(can.id) t ON t.id = can.id 
    WHERE can.parent_id IS NULL;

【讨论】:

以上是关于使用 row_to_json 的 Postgres 递归查询的主要内容,如果未能解决你的问题,请参考以下文章

使用postgres json当键为空时删除行

如何将现有函数(包括聚合)包装到 Postgres 中的新函数中?

postgreSql聚合函数row_to_json初使用

在 row_to_json 函数中选择查询

使用自联接在 row_to_json() 中复制 JSON 元素

SQLAlchemy PostgreSQL row_to_json关系