MongoDB Aggregate Query Equivalent To PostsgreSQL

Question

This question has two parts. The collection structure is:

_id: MongoID,
agent_id: string,
result: string,
created_on: ISO DATE,
...other fields...

Part one:
Desired Output: One result for each agent_id and result combination with a count: TUPLE representation with Equivalent SQL using PostgreSQL.

( "1234", "Success", 4 ),
( "1234", "Failure", 4 ),
( "4567", "Success", 3 ),
( "7896", "Failure", 2 ),
.....

SELECT agent_id, result, count(*)
FROM table
GROUP BY agent_id, result
HAVING created_on >= now()::date;

I have come up with the below mongo query....I think I have a conceptual or syntax error. The docs say to use $match early in the pipeline:, but although the $match limits the query when I run it by itself, as soon as I add the $group I get way to many results. Also I can't seem to understand how to group by more than one field. How can I edit the below query to get results like the SQL query above?

db.collection.aggregate(
  { $match : 
    { created_on: 
        { $gte: new Date('08-13-2012') //some arbitrary date
    }
  }, $group:
    { _id:"$agent_id" }, 
   $project:
  {_id:0, agent_id:1, result:1}
})

Part 2) The first result set would be adequate, but not optimal. With PostgreSQL I can achieve a result set like:

( "1234", { "Success", "Failure" }, { 4, 3 } ),
( "4567", { "Success", "Failure" }, { 3, 0 } ),
( "7896", { "Success", "Failure" }, { 0, 2 } )

I can do this in Postgresql with the array data type and a set_to_array function (custom function). The Pg specific SQL is:

SELECT agent_id, set_to_array(result), set_to_array( count(*) )
FROM table
GROUP BY agent_id, result
HAVING created_on >= now()::date;

I believe the equivalent data structure in mongodb would look like :

[
   { "1234", [ { "success": 4 }, { "failure": 4 } ] },
   { "4567", [ { "success": 3 }, { "failure": 0 } ] },
   { "7896", [ { "success": 0 }, { "failure": 0 } ] }
]

Is it possible to achieve these desired compressed results with mongodb aggregate framework ?

Answer 1

Here you go:

Created some test data:

db.test.insert({agent_id:"1234", result:"Failure", created_on:new Date()}); db.test.insert({agent_id:"1234", result:"Success", created_on:new Date()}); db.test.insert({agent_id:"1234", result:"Failure", created_on:new Date()}); db.test.insert({agent_id:"1234", result:"Success", created_on:new Date()}); db.test.insert({agent_id:"1234", result:"Failure", created_on:new Date()}); db.test.insert({agent_id:"1234", result:"Success", created_on:new Date()}); db.test.insert({agent_id:"1234", result:"Success", created_on:new Date()}); db.test.insert({agent_id:"1324", result:"Success", created_on:new Date()}); db.test.insert({agent_id:"1324", result:"Success", created_on:new Date()}); db.test.insert({agent_id:"1324", result:"Success", created_on:new Date()}); db.test.insert({agent_id:"1324", result:"Success", created_on:new Date()}); db.test.insert({agent_id:"1324", result:"Failure", created_on:new Date()}); db.test.insert({agent_id:"1324", result:"Failure", created_on:new Date()}); db.test.insert({agent_id:"1324", result:"Failure", created_on:new Date()}); db.test.insert({agent_id:"1324", result:"Failure", created_on:new Date()}); db.test.insert({agent_id:"1324", result:"Failure", created_on:new Date()}); db.test.insert({agent_id:"1324", result:"Failure", created_on:new Date()}); db.test.insert({agent_id:"1324", result:"Failure", created_on:new Date()});

db.test.aggregate(
  {
    $match:{ /* filter out the things you want to aggregate */
      created_on:{$gte:new Date(1000000)}
    }
  }, 
  {
    $group: {_
      _id: { /* the things you want to group on go in the _id */
        agent_id:"$agent_id", 
        result:"$result"
      }, 
      count:{$sum:1} /* simple count */
    }
  }, 
  {
    $project: { /* take the id out into the separate fields for your tuple. */
      _id:0, 
      agent_id:"$_id.agent_id", 
      result:"$_id.result", 
      count:"$count"
    }
  });

Gives:

{
"result" : [
    {
        "count" : 7,
        "agent_id" : "1324",
        "result" : "Failure"
    },
    {
        "count" : 4,
        "agent_id" : "1324",
        "result" : "Success"
    },
    {
        "count" : 4,
        "agent_id" : "1234",
        "result" : "Success"
    },
    {
        "count" : 3,
        "agent_id" : "1234",
        "result" : "Failure"
    }
],
"ok" : 1
} 

Adding part 2--pretty similar to part 1, but the counting is a bit more complicated; basically you count only if it matches what you want to count:

db.test.aggregate(
  {
    $match: { 
      created_on: {$gte:new Date(1000000)}
    }
  }, 
  {
    $group: {
      _id: { 
        agent_id:"$agent_id"
      }, 
      failure: {
        $sum:{
          $cond:[
            {$eq:["$result","Failure"]}, 
            1, 
            0
          ]
        }
      }, 
      success: {
        $sum: { 
          $cond:[
            {$eq:["$result","Success"]}, 
            1, 
            0
          ]
        }
      } 
    } 
  }, 
  {
    $project: {
      _id: 0, 
      agent_id: "$_id.agent_id", 
      failure: "$failure", 
      success: "$success"
    }
  });

Gives:

{
"result" : [
    {
        "failure" : 7,
        "success" : 4,
        "agent_id" : "1324"
    },
    {
        "failure" : 3,
        "success" : 4,
        "agent_id" : "1234"
    }
],
"ok" : 1
}