PostgreSQL 中的递归 JSON 生成

Question

我在 PostgreSQL 9.5 服务器中有以下表：

值得注意的结构是location 理论上是无限递归的。我需要从根location 生成一条 JSON 消息，递归到所有子位置；每个location 都有一些属性，一个inventory 项目数组和一个子locations 数组。

如何为此创建高性能查询？我正在查看各种 PostgreSQL JSON 函数、LATERAL 关键字、CTE，并且有点困惑。我已经用非递归查询完成了 JSON 输出，但不确定如何干净地处理递归。

这是一个示例输出：

{
  "id": 1000,
  "name": "By Location",
  "type": "SITE",
  "locations": [
    {
      "id": 1005,
      "name": "Storage A",
      "type": "STOR",
      "locations": [ ...(same schema as parent)... ],
      "inventories": [ ...(see below for schema)... ]
    },
    {
      "id": 1017,
      "name": "Storage B",
      "name": "COLD",
      "locations": [ ...(same schema as parent)... ],
      "inventories": [...(see below for schema)... ]
    }
  ],
  "inventories": [
    {
      "id": 5340,
      "product_id": 9120,
      "name": "Product X",
      "thumb": "https://example.com/api/images/nnnn.jpg",
      "sort_order": 1,
      "par_level": 3.5,
      "created": 1452898800,
      "updated": 1453071600,
      "measures": [
        {"id": 3498, "quantity": 2.25, "created": 1453071600, "updated": 1453071600},
        {"id": 3456, "quantity": 3.25, "created": 1452898800, "updated": 1452898800}
      ]
    }
  ]
}

Answer 1

让我们把它分成几块。首先，您将使用嵌套子查询来创建嵌套数组。公用表表达式可能会有所帮助。

其他技巧是 row_to_json 和 json_agg。

第一个问题是 row_to_json 需要表作为参数来返回正确的标签。

select json_agg(locations) from locations

将为每一行返回一个 json 对象。要仅使用某些字段，您需要创建一个类型并强制转换为它，或者使用 CTE 和上面的语法。在大多数情况下，我会使用 CTE。

所以你最终会得到类似的东西：

WITH lowlevel1 AS 
( 
       SELECT a, 
          b, 
          c 
       FROM   tab1) ,lowlevel2 AS 
( 
       SELECT b, 
          c, 
          d 
       FROM   tab2) ,midlevel1 AS 
( 
        SELECT          e, 
                f, 
                g, 
                json_agg(lowlevel1) AS lab1, 
                json_agg(lowlevel2) AS lab2 
        FROM            tab3 
        LEFT OUTER JOIN lowlevel1 
        ON              tab3.id = lowlevel1.parent 
        LEFT OUTER JOIN lowlevel2 
        ON              tab3.id = lovlevel2.parent)
SELECT row_to_json(midlevel1) from midlevel1

或者在最后一行使用json_agg(midlevel1) 而不是row_to_json(midlevel1) 返回一个包含所有行的数组。

CTE 还支持使用 RECURSIVE 修饰符进行回归。但是，这会返回一个包含回归结果的表，而不是嵌套的 JSON 结构。因此，您可能需要明确编码所需的嵌套级别。

如果一个元素不存在 Postgres 将返回 null。例如，没有子位置的列表将返回"locations":[null]。要将其替换为更有意义的结果，可以使用 case when <> then <> else '[]' end 或 if <> then <> else '[]' end。第一个是“搜索案例”，其中每个测试都是一个布尔表达式。

Answer 2

我最终创建了三个函数。可能做得更少，但这些函数可以在其他查​​询中重用。基本上，JSON 输出中应该有一个值数组的任何位置都由一个函数处理，该函数返回一个记录集，该记录集得到json_agg()'ed。

CREATE OR REPLACE FUNCTION get_measures_by_inventory_as_json(invid UUID, del TIMESTAMP WITH TIME ZONE DEFAULT now())
RETURNS TABLE(inventory_id UUID, measure_json JSON)
AS $$
  -- returns a JSONified record per measure tied to an inventory record
  SELECT m.inventory_id, json_build_object(
    'id', m.id, 
    'quantity', m.quantity,
    'read', TRUNC(EXTRACT(EPOCH FROM m.read_date)),
    'created', TRUNC(EXTRACT(EPOCH FROM m.created)), 
    'updated', TRUNC(EXTRACT(EPOCH FROM m.updated)),
    'deleted', TRUNC(EXTRACT(EPOCH FROM m.deleted))
  )
  FROM measure m
  WHERE m.inventory_id = invid
  AND (m.deleted >= del);
$$
LANGUAGE sql;



CREATE OR REPLACE FUNCTION get_inventories_by_location_as_json(locid UUID, del TIMESTAMP WITH TIME ZONE DEFAULT now())
RETURNS TABLE(location_id UUID, inventory_json JSON)
AS $$
  -- returns a JSONified set of inventory items, with product info and measures, given a location
  SELECT i.location_id, json_build_object(
    'id', i.id,
    'product_id', p.id,
    'name', p.name,
    'mass_quantity', p.mass_quantity,
    'mass_unit', um.code,
    'count_unit', uc.code,
    'thumb', p.product_picture_uri,
    'sort_order', i.sort_order,
    'par_level', i.par_level,
    'created', TRUNC(EXTRACT(EPOCH FROM i.created)),
    'updated', TRUNC(EXTRACT(EPOCH FROM i.updated)),
    'deleted', TRUNC(EXTRACT(EPOCH FROM i.deleted)),
    'measures', COALESCE((SELECT json_agg(measure_json) FROM get_measures_by_inventory_as_json(i.id)), '[]')::json
  )
  FROM inventory i 
  INNER JOIN product p ON i.product_id = p.id
  LEFT JOIN unit um ON p.mass_unit_id = um.id
  LEFT JOIN unit uc ON p.count_unit_id = uc.id
  WHERE i.location_id = locid
  AND i.deleted >= del
  AND p.deleted >= del;
$$
LANGUAGE sql;



CREATE OR REPLACE FUNCTION get_inventories_recursive_as_json(locid UUID[], del TIMESTAMP WITH TIME ZONE DEFAULT now())
RETURNS JSON
AS $$
  -- returns JSONified location info and inventories in that location
  -- and recurses into child locations, showing the same
  SELECT json_agg(loc) FROM (
    SELECT l.id, array_agg(c.id), json_build_object(
    'id', l.id,
    'name', l.name,
    'type', t.code,
    'locations', get_inventories_recursive_as_json(array_agg(c.id)),
    'inventories', COALESCE((SELECT json_agg(inventory_json) FROM get_inventories_by_location_as_json(l.id)),'[]')::json
    ) AS loc
    FROM location l
    LEFT OUTER JOIN location c ON l.id = c.parent_id
    INNER JOIN location_type t ON l.location_type_id = t.id
    WHERE l.id = ANY(locid)
    AND l.deleted >= del
    GROUP BY l.id, l.name, t.code
  ) AS out;
$$
LANGUAGE sql;

尝试通过 CTE 来完成，这本来会非常优雅，但无法弄清楚如何在不发生与无法在递归中聚合相关的错误的情况下做到这一点。

WITH RECURSIVE locations AS (
  WITH inventories AS (
    WITH measures AS (
     SELECT m.inventory_id, json_agg(json_build_object(
        'id', m.id, 
        'quantity', m.quantity,
        'read', TRUNC(EXTRACT(EPOCH FROM m.read_date)),
        'created', TRUNC(EXTRACT(EPOCH FROM m.created)), 
        'updated', TRUNC(EXTRACT(EPOCH FROM m.updated)),
        'deleted', TRUNC(EXTRACT(EPOCH FROM m.deleted))
      )) as measures
      FROM measure m
      GROUP BY m.inventory_id
    )
    SELECT i.location_id, json_agg(json_build_object(
      'id', i.id,
      'product_id', p.id,
      'name', p.name,
      'mass_quantity', p.mass_quantity,
      'mass_unit', um.code,
      'count_unit', uc.code,
      'thumb', p.product_picture_uri,
      'sort_order', i.sort_order,
      'par_level', i.par_level,
      'created', TRUNC(EXTRACT(EPOCH FROM i.created)),
      'updated', TRUNC(EXTRACT(EPOCH FROM i.updated)),
      'deleted', TRUNC(EXTRACT(EPOCH FROM i.deleted)),
      'measures', COALESCE(m.measures, '[]')
    )) AS inventories
    FROM inventory i 
    INNER JOIN product p ON i.product_id = p.id
    LEFT JOIN unit um ON p.mass_unit_id = um.id
    LEFT JOIN unit uc ON p.count_unit_id = uc.id
    LEFT JOIN measures m ON i.id = m.inventory_id 
    GROUP BY i.location_id
  )

  SELECT null as id, null as name, null as type, null as inventories
  FROM location l
  INNER JOIN location_type t ON l.location_type_id = t.id
  LEFT OUTER JOIN inventories i ON l.id = i.location_id
  GROUP BY l.parent_id

  /*UNION ALL

  SELECT p.id, p.parent_id, p.name, t.code, COALESCE(i.inventories, '[]')::jsonb AS inventories, json_agg(row_to_json(c.*))
  FROM location p
  INNER JOIN location_type t ON p.location_type_id = t.id
  LEFT OUTER JOIN inventories i ON p.id = i.location_id
  INNER JOIN locations c ON p.id = c.parent_id
  GROUP BY p.id, p.name, t.code, COALESCE(i.inventories, '[]')::jsonb*/

)
SELECT * FROM locations