# An Alternative Elixir Driver for MongoDB

[![Hex Docs](](
[![Last Updated](](

## Features

- supports MongoDB versions 3.2, 3.4, 3.6, 4.x, 5.x
- connection pooling ([through DBConnection 2.x](
- streaming cursors
- performant ObjectID generation
- aggregation pipeline
- replica sets
- support for SCRAM-SHA-256 (MongoDB 4.x)
- support for GridFS ([See](
- support for change streams api ([See](
- support for bulk writes ([See](
- support for driver sessions ([See](
- support for driver transactions ([See](
- support for command monitoring ([See](
- support for retryable reads ([See](
- support for retryable writes ([See](
- support for simple structs using the Mongo.Encoder protocol
- support for complex and nested documents using the `Mongo.Collection` macros
- support for streaming protocol ([See](

## Usage

### Installation

Add `mongodb_driver` to your mix.exs `deps`.

defp deps do
  [{:mongodb_driver, "~> 0.9.0"}]

Then run `mix deps.get` to fetch dependencies.

### Simple Connection to MongoDB

# Starts an unpooled connection
{:ok, conn} = Mongo.start_link(url: "mongodb://localhost:27017/my-database")

# Gets an enumerable cursor for the results
cursor = Mongo.find(conn, "test-collection", %{})

|> Enum.to_list()
|> IO.inspect

To specify a username and password, use the `:username`, `:password`, and `:auth_source` options.

# Starts an unpooled connection
{:ok, conn} =
    Mongo.start_link(url: "mongodb://localhost:27017/db-name",
                     username: "test_user",
                     password: "hunter2",
                     auth_source: "admin_test")

# Gets an enumerable cursor for the results
cursor = Mongo.find(conn, "test-collection", %{})

|> Enum.to_list()
|> IO.inspect

For secure requests, you may need to add some more options; see the "AWS, TLS and Erlang SSL ciphers" section below.

Failing operations return a `{:error, error}` tuple where `error` is a
`Mongo.Error` object:

   code: 13435,
   error_labels: [],
   host: nil,
   message: "not master and slaveOk=false",
   resumable: true,
   retryable_reads: true,
   retryable_writes: true

## Examples

### Find

Using `$and`

Mongo.find(:mongo, "users", %{"$and" => [%{email: ""}, %{first_name: "first_name"}]})

Using `$or`

Mongo.find(:mongo, "users", %{"$or" => [%{email: ""}, %{first_name: "first_name"}]})

Using `$in`

Mongo.find(:mongo, "users", %{email: %{"$in" => ["", ""]}})

### Inserts

To insert a single document:

Mongo.insert_one(top, "users", %{first_name: "John", last_name: "Smith"})

To insert a list of documents:

Mongo.insert_many(top, "users", [
  %{first_name: "John", last_name: "Smith"},
  %{first_name: "Jane", last_name: "Doe"}

## Data Representation

Since BSON documents are ordered Elixir maps cannot be used to fully represent them. This driver chose to accept both maps and lists of key-value pairs when encoding but will only decode documents to lists. This has the side-effect that it's impossible to discern empty arrays from empty documents. Additionally, the driver will accept both atoms and strings for document keys but will only decode to strings. BSON symbols can only be decoded.

    BSON                Elixir
    ----------          ------
    double              0.0
    string              "Elixir"
    document            [{"key", "value"}] | %{"key" => "value"} (1)
    binary              %BSON.Binary{binary: <<42, 43>>, subtype: :generic}
    UUID                %BSON.Binary{binary: <<42, 43>>, subtype: :uuid}
    UUID (old style)    %BSON.Binary{binary: <<42, 43>>, subtype: :uuid_old}
    object id           %BSON.ObjectId{value: <<...>>}
    boolean             true | false
    UTC datetime        %DateTime{}
    null                nil
    regex               %BSON.Regex{pattern: "..."}
    JavaScript          %BSON.JavaScript{code: "..."}
    timestamp           #BSON.Timestamp<value:ordinal>"
    integer 32          42
    integer 64          #BSON.LongNumber<value>
    symbol              "foo" (2)
    min key             :BSON_min
    max key             :BSON_max
    decimal128          Decimal{}

## Writing your own encoding info

If you want to write a custom struct to your mongo collection - you can do that
by implementing `Mongo.Encoder` protocol for your module. The output should be a map,
which will be passed to the Mongo database.


defmodule CustomStruct do
  @fields [:a, :b, :c, :id]
  @enforce_keys @fields
  defstruct @fields
  defimpl Mongo.Encoder do
    def encode(%{a: a, b: b, id: id}) do
        _id: id,
        a: a,
        b: b,
        custom_encoded: true

So, given the struct:

%CustomStruct{a: 10, b: 20, c: 30, id: "5ef27e73d2a57d358f812001"}

it will be written to database, as:

  "a": 10,
  "b": 20,
  "custom_encoded": true,
  "_id": "5ef27e73d2a57d358f812001"

## Collections

While using the `Mongo.Encoder` protocol give you the possibility to encode your structs into maps the opposite way to decode those maps into structs is missing. To handle it you can use the `Mongo.Collection` which provides some boilerplate code for a better support of structs while using the MongoDB driver

- automatic load and dump function
- reflection functions
- type specification
- support for embedding one and many structs
- support for `after load` function
- support for `before dump` function
- support for id generation
- support for default values
- support for derived values

When using the MongoDB driver only maps and keyword lists are used to represent documents.
If you prefer to use structs instead of the maps to give the document a stronger meaning or to emphasize
its importance, you have to create a `defstruct` and fill it from the map manually:

defmodule Label do
  defstruct name: "warning", color: "red"

iex> label_map = Mongo.find_one(:mongo, "labels", %{})
  %{"name" => "warning", "color" => "red"}
iex> label = %Label{name: label_map["name"], color: label_map["color"]}

We have defined a module `Label` as `defstruct`, then we get the first label document
the collection `labels`. The function `find_one` returns a map. We convert the map manually and
get the desired struct. If we want to save a new structure, we have to do the reverse. We convert the struct into a map:

iex> label = %Label{}
iex> label_map = %{"name" =>, "color" => label.color}
iex> {:ok, _} = Mongo.insert_one(:mongo, "labels", label_map)

Alternatively, you can also remove the `__struct__` key from `label`. The MongoDB driver automatically
converts the atom keys into strings (Or use the `Mongo.Encode` protocol)

iex>  Map.drop(label, [:__struct__])
%{color: :red, name: "warning"}

If you use nested structures, the work becomes a bit more complex. In this case, you have to use the inner structures
convert manually, too. If you take a closer look at the necessary work, two basic functions can be derived:

- `load` Conversion of the map into a struct.
- `dump` Conversion of the struct into a map.

`Mongo.Collection` provides the necessary macros to automate this boilerplate code. The above example can be rewritten as follows:

defmodule Label do
    use Mongo.Collection

    document do
      attribute :name, String.t(), default: "warning"
      attribute :color, String.t(), default: :red

This results in the following module:

defmodule Label do

    defstruct [name: "warning", color: "red"]

    @type t() :: %Label{String.t(), String.t()}

    def new()...
    def load(map)...
    def dump(%Label{})...
    def __collection__(:attributes)...
    def __collection__(:types)...
    def __collection__(:collection)...
    def __collection__(:id)...


You can now create new structs with the default values and use the conversion functions between map and structs:

iex(1)> x =
%Label{color: :red, name: "warning"}
iex(2)> m = Label.dump(x)
%{color: :red, name: "warning"}
iex(3)> Label.load(m, true)
%Label{color: :red, name: "warning"}

The `load/2` function distinguishes between keys of type binarys `load(map, false)` and keys of type atoms `load(map, true)`. The default is `load(map, false)`:

iex(1)> m = %{"color" => :red, "name" => "warning"}
iex(2)> Label.load(m)
%Label{color: :red, name: "warning"}

If you would now expect atoms as keys, the result of the conversion is not correct in this case:

iex(3)> Label.load(m, true)
%Label{color: nil, name: nil}

The background is that MongoDB always returns binarys as keys and structs use atoms as keys.

For more information look at the module documentation `Mongo.Collection`.

Of course, using the `Mongo.Collection` is not free. When loading and saving, the maps are converted into structures, which increases CPU usage somewhat. When it comes to speed, it is better to use the maps directly.

## Using the Repo Module

For convenience, you can also `use` the `Mongo.Repo` module in your application to configure the MongoDB application.

Simply create a new module and include the `use Mongo.Repo` macro:

defmodule MyApp.Repo do
  use Mongo.Repo,
    otp_app: :my_app,
    topology: :mongo

To configure the MongoDB add the configuration to your `config.exs`:

config :my_app, MyApp.Repo,
  url: "mongodb://localhost:27017/my-app-dev",
  timeout: 60_000,
  idle_interval: 10_000,
  queue_target: 5_000

Finally, we can add the `Mongo` instance to our application supervision tree:

  children = [
    # ...
    {Mongo, MyApp.Repo.config()},
    # ...

In addition, the convenient configuration, the `Mongo.Repo` module will also include query functions to use with your
`Mongo.Collection` modules.

For more information check out the `Mongo.Repo` module documentation and the `Mongo` module documentation.

## Logging

You config the logging output by adding in your config file this line

config :mongodb_driver, log: true

The attribute `log` supports `true`, `false` or a log level like `:info`. The default value is `false`. If you turn
logging on, then you will see log output (command, collection, parameters):

[info] CMD find "my-collection" [filter: [name: "Helga"]] db=2.1ms

## Telemetry

The driver uses the [:telemetry]( package to emit the execution duration
for each command. The event name is `[:mongodb_driver, :execution]` and the driver uses the following meta data:

metadata = %{
type: :mongodb_driver,
command: command,
params: parameters,
collection: collection,
options: Keyword.get(opts, :telemetry_options, [])

:telemetry.execute([:mongodb_driver, :execution], %{duration: duration}, metadata)

In a Phoenix application with installed Phoenix Dashboard the metrics can be used by defining a metric in the Telemetry module:

        tags: [:collection, :command],
        unit: {:microsecond, :millisecond}

Then you see for each collection the execution time for each different command in the Dashboard metric page.

## Connection Pooling

The driver supports pooling by DBConnection (2.x). By default `mongodb_driver` will start a single
connection, but it also supports pooling with the `:pool_size` option. For 3 connections add the `pool_size: 3` option to `Mongo.start_link` and to all
function calls in `Mongo` using the pool:

# Starts an pooled connection
{:ok, conn} = Mongo.start_link(url: "mongodb://localhost:27017/db-name", pool_size: 3)

# Gets an enumerable cursor for the results
cursor = Mongo.find(conn, "test-collection", %{})

|> Enum.to_list()
|> IO.inspect

If you're using pooling it is recommended to add it to your application supervisor:

def start(_type, _args) do
  import Supervisor.Spec

  children = [
    worker(Mongo, [[name: :mongo, database: "test", pool_size: 3]])

  opts = [strategy: :one_for_one, name: MyApp.Supervisor]
  Supervisor.start_link(children, opts)

Due to the mongodb specification, an additional connection is always set up for the monitor process.

## Replica Sets

By default, the driver will discover the deployment's topology and will connect
to the replica set automatically, using either the seed list syntax or the URI
syntax. Assuming the deployment has nodes at ``,
`` and ``, either of the following
invocations will discover the entire deployment:

{:ok, pid} = Mongo.start_link(database: "test", seeds: [""])

{:ok, pid} = Mongo.start_link(url: "mongodb://")

To ensure that the connection succeeds even when some of the nodes are not
available, it is recommended to list all nodes in both the seed list and the
URI, as follows:

{:ok, pid} = Mongo.start_link(database: "test", seeds: ["", "", ""])

{:ok, pid} = Mongo.start_link(url: "mongodb://,,")

Using an SRV URI also discovers all nodes of the deployment automatically.

## Auth Mechanisms

For versions of Mongo 3.0 and greater, the auth mechanism defaults to SCRAM.
If you'd like to use [MONGODB-X509](
authentication, you can specify that as a `start_link` option.

{:ok, pid} = Mongo.start_link(database: "test", auth_mechanism: :x509)

## AWS, TLS and Erlang SSL Ciphers

Some MongoDB cloud providers (notably AWS) require a particular TLS cipher that isn't enabled
by default in the Erlang SSL module. In order to connect to these services,
you'll want to add this cipher to your `ssl_opts`:

{:ok, pid} = Mongo.start_link(database: "test",
      ssl_opts: [
        ciphers: ['AES256-GCM-SHA384'],
        cacertfile: "...",
        certfile: "...")

## Change Streams

Change streams are available in replica set and sharded cluster deployments
and tell you about changes of documents in collections. They work like endless

The special thing about change streams is that they are resumable: in case of
a resumable error, no exception is propagated to the application, but instead
the cursor is re-scheduled at the last successful location.

The following example will never stop, thus it is a good idea to use a process
for reading from change streams:

seeds = ["", "", ""]
{:ok, top} = Mongo.start_link(database: "my-db", seeds: seeds, appname: "getting rich")
cursor =  Mongo.watch_collection(top, "accounts", [], fn doc -> IO.puts "New Token #{inspect doc}" end, max_time: 2_000 )
cursor |> Enum.each(fn doc -> IO.puts inspect doc end)

An example with a spawned process that sends messages to the monitor process:

def for_ever(top, monitor) do
    cursor = Mongo.watch_collection(top, "users", [], fn doc -> send(monitor, {:token, doc}) end)
    cursor |> Enum.each(fn doc -> send(monitor, {:change, doc}) end)

spawn(fn -> for_ever(top, self()) end)

For more information see `Mongo.watch_collection/5`

## Indexes

To create indexes you can call the function `Mongo.create_indexes/4`:

indexes =  [[key: [files_id: 1, n: 1], name: "files_n_index", unique: true]]
Mongo.create_indexes(topology_pid, "my_collection", indexes, opts)

You specify the `indexes` parameter as a keyword list with all options described in the documentation of the [createIndex]( command.

For more information see:

- `Mongo.create_indexes/4`
- `Mongo.drop_index/4`

## Bulk Writes

The motivation for bulk writes lies in the possibility of optimization, the same operations
to group. Here, a distinction is made between disordered and ordered bulk writes.
In disordered, inserts, updates, and deletes are grouped as individual commands
sent to the database. There is no influence on the order of the execution.
A good use case is the import of records from one CSV file.
The order of the inserts does not matter.

For ordered bulk writers, order compliance is important to keep.
In this case, only the same consecutive operations are grouped.

Currently, all bulk writes are optimized in memory. This is unfavorable for large bulk writes.
In this case, one can use streaming bulk writes that only have a certain set of
group operation in memory and when the maximum number of operations
has been reached, operations are written to the database. The size can be specified.

Using ordered bulk writes. In this example we first insert some dog's name, add an attribute `kind`
and change all dogs to cats. After that we delete three cats. This example would not work with
unordered bulk writes.


bulk = "bulk"
       |> OrderedBulk.insert_one(%{name: "Greta"})
       |> OrderedBulk.insert_one(%{name: "Tom"})
       |> OrderedBulk.insert_one(%{name: "Waldo"})
       |> OrderedBulk.update_one(%{name: "Greta"}, %{"$set": %{kind: "dog"}})
       |> OrderedBulk.update_one(%{name: "Tom"}, %{"$set": %{kind: "dog"}})
       |> OrderedBulk.update_one(%{name: "Waldo"}, %{"$set": %{kind: "dog"}})
       |> OrderedBulk.update_many(%{kind: "dog"}, %{"$set": %{kind: "cat"}})
       |> OrderedBulk.delete_one(%{kind: "cat"})
       |> OrderedBulk.delete_one(%{kind: "cat"})
       |> OrderedBulk.delete_one(%{kind: "cat"})

result = Mongo.BulkWrite.write(:mongo, bulk, w: 1)

In the following example we import 1.000.000 integers into the MongoDB using the stream api:

We need to create an insert operation for each number. Then we call the ``
function to import it. This function returns a stream function which accumulate
all inserts operations until the limit `1000` is reached. In this case the operation group is send to
MongoDB. So using the stream api you can reduce the memory using while
importing big volume of data.

|> i -> Mongo.BulkOps.get_insert_one(%{number: i}) end)
|> Mongo.UnorderedBulk.write(:mongo, "bulk", 1_000)

For more information see:

- `Mongo.UnorderedBulk`
- `Mongo.OrderedBulk`
- `Mongo.BulkWrite`
- `Mongo.BulkOps`

and have a look at the test units as well.

## GridFS

The driver supports the GridFS specifications. You create a `Mongo.GridFs.Bucket`
struct and with this struct you can upload and download files. For example:

    bucket =
    upload_stream = Upload.open_upload_stream(bucket, "test.jpg")
    src_filename = "./test/data/test.jpg"!(src_filename, [], 512) |> Stream.into(upload_stream) |>

    file_id =

In the example a new bucket with default values is used to upload a file from the file system (`./test/data/test.jpg`) to the MongoDB (using the name `test.jpg`). The `upload_stream` struct contains the id of the new file which can be used to download the stored file. The following code fragments downloads the file by using the `file_id`.

    dest_filename = "/tmp/my-test-file.jps"

    with {:ok, stream} <- Mongo.GridFs.Download.open_download_stream(bucket, file_id) do
      |> Stream.into(!(dest_filename))

For more information see:

- [Mongo.GridFs.Bucket](
- [Mongo.GridFs.Download](
- [Mongo.GridFs.Upload](

## Transactions

Since MongoDB 4.x, transactions for multiple write operations are possible. Transaction uses sessions, which
just contain a transaction number for each transaction. The `Mongo.Session` is responsible for the
details, and you can use a convenient api for transactions:


{:ok, ids} = Mongo.transaction(top, fn ->
{:ok, %InsertOneResult{:inserted_id => id1}} = Mongo.insert_one(top, "dogs", %{name: "Greta"})
{:ok, %InsertOneResult{:inserted_id => id2}} = Mongo.insert_one(top, "dogs", %{name: "Waldo"})
{:ok, %InsertOneResult{:inserted_id => id3}} = Mongo.insert_one(top, "dogs", %{name: "Tom"})
{:ok, [id1, id2, id3]}
end, w: 1)

The `Mongo.transaction/3` function supports nesting. This allows the functions to be called from each other and all write operations
are still in the same transaction. The session is stored in the process dictionary under the key `:session`. The surrounding
`Mongo.transaction/3` call creates the session and starts the transaction, storing the session in the process dictionary, commits or
aborts the transaction. All other `Mongo.transaction/3` calls just call the function parameter without other actions.

def insert_dog(top, name) do
  Mongo.insert_one(top, "dogs", %{name: name})

def insert_dogs(top) do
  Mongo.transaction(top, fn ->
    insert_dog(top, "Tom")
    insert_dog(top, "Bell")
    insert_dog(top, "Fass")

:ok = Mongo.transaction(top, fn ->
    insert_dog(top, "Greta")

It is also possible to get more control over the progress of the transaction:

alias Mongo.Session

{:ok, session} = Session.start_session(top, :write, [])
:ok = Session.start_transaction(session)

Mongo.insert_one(top, "dogs", %{name: "Greta"}, session: session)
Mongo.insert_one(top, "dogs", %{name: "Waldo"}, session: session)
Mongo.insert_one(top, "dogs", %{name: "Tom"}, session: session)

:ok = Session.commit_transaction(session)
:ok = Session.end_session(top, session)
For more information see `Mongo.Session` and have a look at the test units as well.

### Aborting a transaction

You have some options to abort a transaction. The simplest possibility is to return an `:error`. For nested
function calls, the `Mongo.abort_transaction/1` function call that throws an exception is suitable.
That means, you can just generate a `raise :should_not_happen` exception as well.

## Command Monitoring

You can watch all events that are triggered while the driver send requests and processes responses. You can use the
`Mongo.EventHandler` as a starting point. It logs the events from the topic `:commands` (by ignoring the `:isMaster` command)
to ``:

iex> Mongo.EventHandler.start()
iex> {:ok, conn} = Mongo.start_link(url: "mongodb://localhost:27017/test")
{:ok, #PID<0.226.0>}
 iex> Mongo.find_one(conn, "test", %{})
                                      [info] Received command: %Mongo.Events.CommandStartedEvent{command: [find: "test", ...
                                                                                                                 [info] Received command: %Mongo.Events.CommandSucceededEvent{command_name: :find, ...

## Testing

Latest MongoDB is used while running the tests. Replica set of three nodes is created and runs all test except the socket and ssl test. If you want to
run the test cases against other MongoDB deployments or older versions, you can use the [mtools]( for deployment and run the test cases locally:

pyenv global 3.6
pip3 install --upgrade pip
pip3 install 'mtools[all]'
export PATH=to-your-mongodb/bin/:$PATH
ulimit -S -n 2048 ## in case of Mac OS X
mlaunch init --setParameter enableTestCommands=1 --replicaset --name "rs_1"
mix test --exclude ssl --exclude socket

The SSL test suite is disabled by default.

### Enable the SSL Tests

`mix test --exclude ssl`

### Enable SSL on Your MongoDB Server

$ openssl req -newkey rsa:2048 -new -x509 -days 365 -nodes -out mongodb-cert.crt -keyout mongodb-cert.key
$ cat mongodb-cert.key mongodb-cert.crt > mongodb.pem
$ mongod --sslMode allowSSL --sslPEMKeyFile /path/to/mongodb.pem

- For `--sslMode` you can use one of `allowSSL` or `preferSSL`
- You can enable any other options you want when starting `mongod`

## Special Thanks

Special thanks to [JetBrains]( for providing a free JetBrains Open Source license for their complete toolbox.

## Copyright and License

Copyright 2015 Eric Meadows-Jönsson and Justin Wood \
Copyright 2019 - 2022 Michael Maier

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at [](

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
See the License for the specific language governing permissions and
limitations under the License.