Navigation
This is an upcoming (i.e. in progress) version of the manual.

$accumulator (aggregation)

Definition

$accumulator

New in version 4.4.

Defines a custom accumulator operator. Accumulators are operators that maintain their state (e.g. totals, maximums, minimums, and related data) as documents progress through the pipeline. Use the $accumulator operator to execute your own JavaScript functions to implement behavior not supported by the MongoDB Query Language. See also $function.

Important

Executing JavaScript inside of an aggregation operator may decrease performance. Only use the $accumulator operator if the provided pipeline operators cannot fulfill your application’s needs.

$accumulator is available in the following pipeline stages:

Syntax

The $accumulator operator has the following syntax:

{
  $accumulator: {
    init: <code>,
    initArgs: <array expression>,        // Optional
    accumulate: <code>,
    accumulateArgs: <array expression>,
    merge: <code>,
    finalize: <code>,                    // Optional
    lang: <string>
  }
}
Field Type Description
init String or Code

Function used to initialize the state. The init function receives its arguments from the initArgs array expression. You can specify the function definition as either BSON type Code or String.

The init function has the following form:

function (<initArg1>, <initArg2>, ...) {
  ...
  return <initialState>
}
initArgs Array

Optional. Arguments passed to the init function.

initArgs has the following form:

[ <initArg1>, <initArg2>, ... ]

Important

When used in a $bucketAuto stage, initArgs cannot refer to the group key (i.e., you cannot use the $<fieldName> syntax). Instead, in a $bucketAuto stage, you can only specify constant values in initArgs.

accumulate String or Code

Function used to accumulate documents. The accumulate function receives its arguments from the current state and accumulateArgs array expression. The result of the accumulate function becomes the new state. You can specify the function definition as either BSON type Code or String.

The accumulate function has the following form:

function(state, <accumArg1>, <accumArg2>, ...) {
  ...
  return <newState>
}
accumulateArgs Array

Arguments passed to the accumulate function. You can use accumulateArgs to specify what field value(s) to pass to the accumulate function.

accumulateArgs has the following form:

[ <accumArg1>, <accumArg2>, ... ]
merge String or Code

Function used to merge two internal states. merge must be either a String or Code BSON type. merge returns the combined result of the two merged states. For information on when the merge function is called, see Merge Two States with $merge.

The merge function has the following form:

function (<state1>, <state2>) {
  <logic to merge state1 and state2>
  return <newState>
}
finalize String or Code

Optional. Function used to update the result of the accumulation.

The finalize function has the following form:

function (state) {
  ...
  return <finalState>
}
lang String

The language used in the $accumulator code.

Important

Currently, the only supported value for lang is js.

Behavior

The following steps outline how the $accumulator operator processes documents:

  1. The operator begins at an initial state, defined by the init function.
  2. For each document, the operator updates the state based on the accumulate function. The accumulate function’s first argument is the current state, and additional arguments are be specified in the accumulateArgs array.
  3. When the operator needs to merge multiple intermediate states, it executes the merge function. For more information on when the merge function is called, see Merge Two States with $merge.
  4. If a finalize function has been defined, once all documents have been processed and the state has been updated accordingly, finalize converts the state to a final output.

Merge Two States with $merge

As part of its internal operations, the $accumulator operator may need to merge two separate, intermediate states. The merge function specifies how the operator should merge two states.

For example, $accumulator may need to combine two states when:

  • $accumulator is run on a sharded cluster. The operator needs to merge the results from each shard to obtain the final result.

  • A single $accumulator operation exceeds its specified memory limit. If you specify the allowDiskUse option, the operator stores the in-progress operation on disk and finishes the operation in memory. Once the operation finishes, the results from disk and memory are merged together using the merge function.

Note

The merge function always merges two states at a time. In the event that more than two states must be merged, the resulting merge of two states is merged with a single state. This process repeats until all states are merged.

Javascript Enabled

To use $accumulator, you must have server-side scripting enabled.

If you do not use $accumulator (or $function, $where, or mapReduce), disable server-side scripting:

See also ➤ Run MongoDB with Secure Configuration Options.

Examples

Use $accumulator to Implement the $avg Operator

Note

This example walks through using the $accumulator operator to implement the $avg operator, which is already supported by MongoDB. The goal of this example is not to implement new functionality, but to illustrate the behavior and syntax of the $accumulator operator with familiar logic.

From the mongo shell, create a sample collection named books with the following documents:

db.books.insertMany([
  { "_id" : 8751, "title" : "The Banquet", "author" : "Dante", "copies" : 2 },
  { "_id" : 8752, "title" : "Divine Comedy", "author" : "Dante", "copies" : 1 },
  { "_id" : 8645, "title" : "Eclogues", "author" : "Dante", "copies" : 2 },
  { "_id" : 7000, "title" : "The Odyssey", "author" : "Homer", "copies" : 10 },
  { "_id" : 7020, "title" : "Iliad", "author" : "Homer", "copies" : 10 }
])

The following operation groups the documents by author, and uses $accumulator to compute the average number of copies across books for each author:

db.books.aggregate([
{
  $group :
  {
    _id : "$author",
    avgCopies:
    {
      $accumulator:
      {
        init: function() {                        // Set the initial state
          return { count: 0, sum: 0 }
        },

        accumulate: function(state, numCopies) {  // Define how to update the state
          return {
            count: state.count + 1,
            sum: state.sum + numCopies
          }
        },

        accumulateArgs: ["$copies"],              // Argument required by the accumulate function

        merge: function(state1, state2) {         // When the operator performs a merge,
          return {                                // add the fields from the two states
            count: state1.count + state2.count,
            sum: state1.sum + state2.sum
          }
        },

        finalize: function(state) {               // After collecting the results from all documents,
          return (state.sum / state.count)        // calculate the average
        },
        lang: "js"
      }
    }
  }
}
])

Result

This operation returns the following result:

{ "_id" : "Dante", "avgCopies" : 1.6666666666666667 }
{ "_id" : "Homer", "avgCopies" : 10 }

Behavior

The $accumulator defines an initial state where count and sum are both set to 0. For each document that the $accumulator processes, it updates the state by:

  • Incrementing the count by 1 and
  • Adding the values of the document’s copies field to the sum. The accumulate function can access the copies field because it is passed in the accumulateArgs field.

With each document that is processed, the accumulate function returns the updated state.

Once all documents have been processed, the finalize function divides the sum of the copies by the count of documents to obtain the average. This removes the need to keep a running computed average, since the finalize function receives the cumulative sum and count of all documents.

Comparison with $avg

This operation is equivalent to the following pipeline, which uses the $avg operator:

db.books.aggregate([
{
  $group : {
    _id : "$author",
    avgCopies: { $avg: "$copies" }
  }
}
])

Use initArgs to Vary the Initial State by Group

You can use the initArgs option in to vary the initial state of $accumulator. This can be useful if you want to, for example:

  • Use the value of a field which is not in your state to affect your state, or
  • Set the initial state to a different value based on the group being processed.

From the mongo shell, create a sample collection named restaurants with the following documents:

db.restaurants.insertMany([
  { "_id" : 1, "name" : "Food Fury", "city" : "Bettles", "cuisine" : "American" },
  { "_id" : 2, "name" : "Meal Macro", "city" : "Bettles", "cuisine" : "Chinese" },
  { "_id" : 3, "name" : "Big Crisp", "city" : "Bettles", "cuisine" : "Latin" },
  { "_id" : 4, "name" : "The Wrap", "city" : "Onida", "cuisine" : "American" },
  { "_id" : 5, "name" : "Spice Attack", "city" : "Onida", "cuisine" : "Latin" },
  { "_id" : 6, "name" : "Soup City", "city" : "Onida", "cuisine" : "Chinese" },
  { "_id" : 7, "name" : "Crave", "city" : "Pyote", "cuisine" : "American" },
  { "_id" : 8, "name" : "The Gala", "city" : "Pyote", "cuisine" : "Chinese" }
])

Suppose an application allows users to query this data to find restaurants. It may be useful to show more results for the city where the user lives. For this example, we assume that the user’s city is called in a variable called userProfileCity.

The following aggregation pipeline groups the documents by city. The operation uses the $accumulator to display a different number of results from each city depending on whether the restaurant’s city matches the city in the user’s profile:

Note

To execute this example in the mongo shell, replace <userProfileCity> in the initArgs with a string containing an actual city value, such as Bettles.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
db.restaurants.aggregate([
{
  $group :
  {
    _id : { city: "$city" },
    restaurants:
    {
      $accumulator:
      {
        init: function(city, userProfileCity) {        // Set the initial state
          return {
            max: city === userProfileCity ? 3 : 1,     // If the group matches the user's city, return 3 restaurants
            restaurants: []                            // else, return 1 restaurant
          }
        },

        initArgs: ["$city", <userProfileCity>],        // Argument to pass to the init function

        accumulate: function(state, restaurantName) {  // Define how to update the state
          if (state.restaurants.length < state.max) {
            state.restaurants.push(restaurantName);
          }
          return state;
        },

        accumulateArgs: ["$name"],                     // Argument required by the accumulate function

        merge: function(state1, state2) {
          return {
            max: state1.max,
            restaurants: state1.restaurants.concat(state2.restaurants).slice(0, state1.max)
          }
        },

        finalize: function(state) {                   // Adjust the state to only return field we need
          return state.restaurants
        }

        lang: "js"
      }
    }
  }
}
])

Results

If the value of userProfileCity is Bettles, this operation returns the following result:

{ "_id" : { "city" : "Bettles" }, "restaurants" : { "restaurants" : [ "Food Fury", "Meal Macro", "Big Crisp" ] } }
{ "_id" : { "city" : "Onida" }, "restaurants" : { "restaurants" : [ "The Wrap" ] } }
{ "_id" : { "city" : "Pyote" }, "restaurants" : { "restaurants" : [ "Crave" ] } }

If the value of userProfileCity is Onida, this operation returns the following result:

{ "_id" : { "city" : "Bettles" }, "restaurants" : { "restaurants" : [ "Food Fury" ] } }
{ "_id" : { "city" : "Onida" }, "restaurants" : { "restaurants" : [ "The Wrap", "Spice Attack", "Soup City" ] } }
{ "_id" : { "city" : "Pyote" }, "restaurants" : { "restaurants" : [ "Crave" ] } }

If the value of userProfileCity is Pyote, this operation returns the following result:

{ "_id" : { "city" : "Bettles" }, "restaurants" : { "restaurants" : [ "Food Fury" ] } }
{ "_id" : { "city" : "Onida" }, "restaurants" : { "restaurants" : [ "The Wrap" ] } }
{ "_id" : { "city" : "Pyote" }, "restaurants" : { "restaurants" : [ "Crave", "The Gala" ] } }

If the value of userProfileCity is any other value, this operation returns the following result:

{ "_id" : { "city" : "Bettles" }, "restaurants" : { "restaurants" : [ "Food Fury" ] } }
{ "_id" : { "city" : "Onida" }, "restaurants" : { "restaurants" : [ "The Wrap" ] } }
{ "_id" : { "city" : "Pyote" }, "restaurants" : { "restaurants" : [ "Crave" ] } }

Behavior

The init function defines an initial state containing max and restaurants fields. The max field sets the maximum number of restaurants for that particular group. If the document’s city field matches userProfileCity, that group contains a maximum of 3 restaurants. Otherwise, if the document _id does not match userProfileCity, the group contains at most a single restaurant. The init function receives both the city userProfileCity arguments from the initArgs array.

For each document that the $accumulator processes, it pushes the name of the restaurant to the restaurants array, provided that name would not put the length of restaurants over the max value. With each document that is processed, the accumulate function returns the updated state.

The merge function defines how to merge two states. The function concatenates the restaurant arrays from each state together, and the length of the resulting array is limited using the slice() method to ensure that it does not exceed the max value.

Once all documents have been processed, the finalize function modifies the resulting state to only return the names of the restaurants. Without this function, the max field would also be included in the output, which does not fulfill any needs for the application.