是否可以使用MongoDB聚合来连接子查询？

Question

是否存在与使用单个 MongoDB 聚合的以下 SQL 语句等效的 MongoDB 聚合查询？ （JOIN 的两个操作数是两个也包含连接操作的子查询。）

SELECT 
    T1.C1, T2.C2 
FROM
    (SELECT 
         T3.C1 C1, T4.C2 C2 
     FROM 
         T3 
     JOIN 
         T4 ON T3.C0 = T4.C0) T1
JOIN
    (SELECT 
         T5.C1 C1, T6.C2 C2 
     FROM 
         T5 
     JOIN 
         T6 ON T5.C0 = T6.C0) T2 ON T1.C1 = T2.C1 
                                 AND T1.C2 = T2.C2;

Answer 1

简短的回答 - 不，lookup要求“from”是一个集合。

正如 WD 所指出的，SQL 模式在 NoSQL 世界中不起作用，但数据库就是数据库 - 如果数据在那里，你总是可以获取它。逻辑将被颠倒 - 你会做

SELECT T3.C1, T6.C2 FROM T3 and complex lookups/unwinds from T4, T5 and T6

。

考虑到OP的片段，可以公平地假设这个问题更具学术性而不是实用性。从这个角度来看——仍然是“不”。不要将数据设计/反规范化到您发现自己处于需要运行此类查询的位置的程度。

Answer 2

这并不能完全解决OP问题。子查询，但展示了如果绝对必要的话，如何使用“临时”集合来代替子查询。请注意，在原始 SQL 中，只有表

T3

.

T4

、

T5

和

T6

实际存在；

T1

和

T2

是子查询结构。此外，原始 SQL 并不是特别有用，因为它只显示匹配的列；我们对这些行从何而来一无所知，因此我们将随身携带

old_id

只是为了展示。

SQL 子查询实际上很可能隐式地执行我们在此处显式显示的操作：创建临时表

T1

和

T2

并且不将其保存在内存中。

db.XT3.drop();
var r = [
    {C0: "A", C1: "R0-1", C2: "R0-2"}
    ,{C0: "A", C1: "R1-1", C2: "R1-2"}
    ,{C0: "B", C1: "R0-1", C2: "R0-2"}
    ,{C0: "C", C1: "R0-1", C2: "R0-2"}
];
var nn = 0; r.forEach(function(d) { d['_id'] = nn++ }); // auto incr ID
db.XT3.insertMany(r);

db.XT4.drop();
var r = [
    {C0: "A", C1: "R0-1", C2: "R0-2"}
    ,{C0: "A", C1: "R2-1", C2: "R2-2"}
    ,{C0: "B", C1: "R0-1", C2: "R0-2"}
    ,{C0: "no_match", C1: "R0-1", C2: "R0-2"}
];
r.forEach(function(d) { d['_id'] = nn++ }); // auto incr ID
db.XT4.insertMany(r);


db.XT5.drop();
var r = [
    {C0: "X", C1: "nope", C2: "nope"}
    ,{C0: "X", C1: "R1-1", C2: "R1-2"}
    ,{C0: "Y", C1: "R0-1", C2: "R0-2"}
    ,{C0: "Z", C1: "R0-1", C2: "R0-2"}
];
r.forEach(function(d) { d['_id'] = nn++ }); // auto incr ID
db.XT5.insertMany(r);

db.XT6.drop();
var r = [
    {C0: "X", C1: "R0-1", C2: "R0-2"}
    ,{C0: "Y", C1: "R0-1", C2: "R0-2"}
    ,{C0: "oiwejfioj", C1: "R0-1", C2: "R0-2"}
];
r.forEach(function(d) { d['_id'] = nn++ }); // auto incr ID
db.XT6.insertMany(r);



db.TMP1.drop();
db.TMP2.drop();

c=db.XT3.aggregate([
    {$lookup: {"from": "XT4",
           let: { id: "$C0" },
           pipeline: [
           {$match: {$expr: {$eq: [ "$C0", "$$id" ]} }}
           ,{$project: {_id:false,C0:false}} // no need to carry them
           ],
           as: "T_3_4"
          }}

    // This ALSO acts to filter out items from XT3 with NO match to XT4:
    ,{$unwind: '$T_3_4'}

    // $unwind will dupe _id of inbound XT3 record; must get rid of it
    // in prep for $out; keep it around for debugging/tracking as 'old_id':
    ,{$project: {_id:false, old_id:'$_id', C1:'$C1', C2:'$T_3_4.C2'}}
    ,{$out: "TMP1"}
]);

db.XT5.aggregate([
    {$lookup: {"from": "XT6",
           let: { id: "$C0" },
           pipeline: [
           {$match: {$expr: {$eq: [ "$C0", "$$id" ]} }}
           ,{$project: {_id:false,C0:false}} // no need to carry them
           ],
           as: "T_5_6"
          }}
    ,{$unwind: '$T_5_6'}
    ,{$project: {_id:false, old_id:'$_id', C1:'$C1', C2:'$T_5_6.C2'}}    
    ,{$out: "TMP2"}
]);

c = db.TMP1.aggregate([
    {$lookup: {"from": "TMP2",
           let: { c1: "$C1", c2: "$C2" },

           // ON T1.C1 = T2.C1 AND T1.C2 = T2.C2;
           pipeline: [
           {$match: {$expr: {$and: [
               {$eq: [ "$$c1", "$C1" ]},
               {$eq: [ "$$c2", "$C2" ]}
           ]} }}
           ,{$project: {_id:false}}
           ],
           as: "X"
          }}

    // Filter out non-matches; comment out this stage to double check
    // docs where T1.C1 = T2.C1 AND T1.C2 = T2.C2 is NOT satisfied:
    ,{$match: {$expr: {$ne:[0, {$size:'$X'}]} }}
]);
emit(c);

产量

{
  _id: ObjectId("655e1e5a4e762532ed9ba7aa"),
  old_id: 0,
  C1: 'R0-1',
  C2: 'R0-2',
  X: [
    {
      old_id: 10,
      C1: 'R0-1',
      C2: 'R0-2'
    }
  ]
}
{
  _id: ObjectId("655e1e5a4e762532ed9ba7ac"),
  old_id: 1,
  C1: 'R1-1',
  C2: 'R0-2',
  X: [
    {
      old_id: 9,
      C1: 'R1-1',
      C2: 'R0-2'
    }
  ]
}
{
  _id: ObjectId("655e1e5a4e762532ed9ba7ae"),
  old_id: 2,
  C1: 'R0-1',
  C2: 'R0-2',
  X: [
    {
      old_id: 10,
      C1: 'R0-1',
      C2: 'R0-2'
    }
  ]
}