Motor 1.1

Motor depends on PyMongo 3.4 or later. It wraps the latest PyMongo code which support the new server features introduced in MongoDB 3.4. (It is a coincidence that the latest MongoDB and PyMongo versions are the same number.)

Highlights include:


Starting in PyMongo 3.4, bson.code.Code.scope may return None, as the default scope is None instead of {}.


PyMongo 3.4+ attempts to create sockets non-inheritable when possible (i.e. it sets the close-on-exec flag on socket file descriptors). Support is limited to a subset of POSIX operating systems (not including Windows) and the flag usually cannot be set in a single atomic operation. CPython 3.4+ implements PEP 446, creating all file descriptors non-inheritable by default. Users that require this behavior are encouraged to upgrade to CPython 3.4+.

Motor 1.0

Motor now depends on PyMongo 3.3 and later. The move from PyMongo 2 to 3 brings a large number of API changes, read Motor 1.0 Migration Guide and the PyMongo 3 changelog carefully.

MotorReplicaSetClient is removed

In Motor 1.0, MotorClient is the only class. Connect to a replica set with a “replicaSet” URI option or parameter:

MotorClient(host, port, replicaSet="my-rs")

New features

New classes MotorGridFSBucket and AsyncIOMotorGridFSBucket conform to the GridFS API Spec for MongoDB drivers. These classes supersede the old MotorGridFS and AsyncIOMotorGridFS. See GridFS changes below, especially note the breaking change in GridFSHandler.

Serve GridFS files over HTTP using aiohttp and AIOHTTPGridFS.

MotorClient changes


  •; clients have opened themselves automatically on demand since version 0.2.
  • MotorClient.seeds, use pymongo.uri_parser.parse_uri() on your MongoDB URI.
  • MotorClient.alive


Unix domain socket paths must be quoted with urllib.parse.quote_plus() (or urllib.quote_plus in Python 2) before they are included in a URI:

path = '/tmp/mongodb-27017.sock'
MotorClient('mongodb://%s' % urllib.parse.quote_plus(path))

Changes to find() and find_one()

The following find/find_one options have been renamed:

These renames only affect your code if you passed these as keyword arguments, like find(fields=['fieldname']). If you passed only positional parameters these changes are not significant for your application.

  • spec -> filter
  • fields -> projection
  • partial -> allow_partial_results

The following find/find_one options have been added:

  • cursor_type (see CursorType for values)
  • oplog_replay
  • modifiers

The following find/find_one options have been removed:

  • network_timeout (use max_time_ms() instead)
  • read_preference (use with_options() instead)
  • tag_sets (use one of the read preference classes from read_preferences and with_options() instead)
  • secondary_acceptable_latency_ms (use the localThresholdMS URI option instead)
  • max_scan (use the new modifiers option instead)
  • snapshot (use the new modifiers option instead)
  • tailable (use the new cursor_type option instead)
  • await_data (use the new cursor_type option instead)
  • exhaust (use the new cursor_type option instead)
  • as_class (use with_options() with CodecOptions instead)
  • compile_re (BSON regular expressions are always decoded to Regex)

The following find/find_one options are deprecated:

  • manipulate

The following renames need special handling.

  • timeout -> no_cursor_timeout - By default, MongoDB closes a cursor after 10 minutes of inactivity. In previous Motor versions, you disabled the timeout by passing timeout=False to MotorCollection.find() or MotorGridFS.find(). The timeout parameter has been renamed to no_cursor_timeout, it defaults to False, and you must now pass no_cursor_timeout=True to disable timeouts.


The old GridFS classes MotorGridFS and AsyncIOMotorGridFS are deprecated in favor of MotorGridFSBucket and AsyncIOMotorGridFSBucket, which comply with MongoDB’s cross-language driver spec for GridFS.

The old classes are still supported, but will be removed in Motor 2.0.

BREAKING CHANGE: The overridable method get_gridfs_file of GridFSHandler now takes a MotorGridFSBucket, not a MotorGridFS. It also takes an additional request parameter.


New method MotorGridIn.abort().

In a Python 3.5 native coroutine, the “async with” statement calls close() automatically:

async def upload():
    my_db = MotorClient().test
    fs = MotorGridFSBucket(my_db)
    async with await fs.new_file() as gridin:
        await gridin.write(b'First part\n')
        await gridin.write(b'Second part')

    # gridin is now closed automatically.


MotorGridOut is now an async iterable, so reading a chunk at a time is much simpler with a Python 3 native coroutine:

async def read_file(file_id):
    fs = motor.motor_tornado.MotorGridFS(db)
    gridout = await fs.get(file_id)

    async for chunk in gridout:



The Motor asyncio API is now fully documented, side by side with the Motor Tornado API.

New Developer Guide added.

Motor 0.7

For asynchronous I/O Motor now uses a thread pool, which is faster and simpler than the prior implementation with greenlets. It no longer requires the greenlet package, and now requires the futures backport package on Python 2.

This version updates the PyMongo dependency from 2.8.0 to 2.9.x, and wraps PyMongo 2.9’s new APIs.

Most of Motor 1.0’s API is now implemented, and APIs that will be removed in Motor 1.0 are now deprecated and raise warnings. See the Motor 1.0 Migration Guide to prepare your code for Motor 1.0.

MotorClient changes

The get_database method is added for getting a MotorDatabase instance with its options configured differently than the MotorClient’s.

New read-only attributes:

MotorReplicaSetClient changes

The get_database() method is added for getting a MotorDatabase instance with its options configured differently than the MotorReplicaSetClient’s.

New read-only attributes:

  • codec_options
  • local_threshold_ms

MotorDatabase changes

The get_collection() method is added for getting a MotorCollection instance with its options configured differently than the MotorDatabase’s.

The connection property is deprecated in favor of a new read-only attribute client.

New read-only attribute:

MotorCollection changes

The with_options() method is added for getting a MotorCollection instance with its options configured differently than this MotorCollection’s.

New read-only attribute:

The following methods wrap PyMongo’s implementation of the standard CRUD API Spec for MongoDB Drivers:

These new methods do not apply SON Manipulators.


MOTOR-124: an import deadlock in Python 2 and Tornado 3 led to an AutoReconnect exception with some replica sets.

Motor 0.6.2

Fix “from motor import *” for Python 3.

Motor 0.6.1

Fix source distribution, which hadn’t included the “frameworks” submodules.

Motor 0.6

This is a bugfix release. Fixing these bugs has introduced tiny API changes that may affect some programs.

motor_asyncio and motor_tornado submodules

These modules have been moved from:



  • motor_asyncio/
  • motor_tornado/

Motor had to make this change in order to omit the motor_asyncio submodule entirely and avoid a spurious SyntaxError being printed when installing in Python 2. The change should be invisible to application code.

Database and collection names with leading underscores

A database or collection whose name starts with an underscore can no longer be accessed as a property:

# Now raises AttributeError.
db = MotorClient()._mydatabase
collection = db._mycollection
subcollection = collection._subcollection

Such databases and collections can still be accessed dict-style:

# Continues to work the same as previous Motor versions.
db = MotorClient()['_mydatabase']
collection = db['_mycollection']

To ensure a “sub-collection” with a name that includes an underscore is accessible, Motor collections now allow dict-style access, the same as Motor clients and databases always have:

# New in Motor 0.6
subcollection = collection['_subcollection']

These changes solve problems with iPython code completion and the Python 3 ABC abstract base class.

Motor 0.5


Motor can now integrate with asyncio, as an alternative to Tornado. My gratitude to Rémi Jolin, Andrew Svetlov, and Nikolay Novik for their huge contributions to Motor’s asyncio integration.

Python 3.5

Motor is now compatible with Python 3.5, which required some effort. Motor not only supports users’ coroutines, it uses coroutines to implement some of its own features, like open() and put(). There is no single way to return a value from a Python 3.5 native coroutine or a Python 2 generator-based coroutine, so Motor internal coroutines that return values were rewritten. (See commit message dc19418c for an explanation.)

async and await

Motor now supports Python 3.5 native coroutines, written with the async and await syntax:

async def f():
    await collection.insert({'_id': 1})

Cursors from find(), aggregate(), or find() can be iterated elegantly and very efficiently in native coroutines with async for:

async def f():
    async for doc in collection.find():


MotorCollection.aggregate() now returns a cursor by default, and the cursor is returned immediately without a yield. The old syntax is no longer supported:

# Motor 0.4 and older, no longer supported.
cursor = yield collection.aggregate(pipeline, cursor={})
while (yield cursor.fetch_next):
    doc = cursor.next_object()

In Motor 0.5, simply do:

# Motor 0.5: no "cursor={}", no "yield".
cursor = collection.aggregate(pipeline)
while (yield cursor.fetch_next):
    doc = cursor.next_object()

Or with Python 3.5 and later:

# Motor 0.5, Python 3.5.
async for doc in collection.aggregate(pipeline):

MongoDB versions 2.4 and older do not support aggregation cursors. For compatibility with older MongoDBs, aggregate() now takes an argument cursor=False, and returns a Future that you can yield to get all the results in one document:

# Motor 0.5 with MongoDB 2.4 and older.
reply = yield collection.aggregate(cursor=False)
for doc in reply['results']:


Motor 0.5 deprecates a large number of APIs that will be removed in version 1.0:

  • ~MotorClient.port
  • ~MotorClient.document_class
  • ~MotorClient.tz_aware
  • ~MotorClient.secondary_acceptable_latency_ms
  • ~MotorClient.tag_sets
  • ~MotorClient.uuid_subtype
  • ~MotorClient.disconnect
  • ~MotorClient.alive
  • ~MotorReplicaSetClient.document_class
  • ~MotorReplicaSetClient.tz_aware
  • ~MotorReplicaSetClient.secondary_acceptable_latency_ms
  • ~MotorReplicaSetClient.tag_sets
  • ~MotorReplicaSetClient.uuid_subtype
  • ~MotorReplicaSetClient.alive
  • ~MotorDatabase.secondary_acceptable_latency_ms
  • ~MotorDatabase.tag_sets
  • ~MotorDatabase.uuid_subtype
  • ~MotorCollection.secondary_acceptable_latency_ms
  • ~MotorCollection.tag_sets
  • ~MotorCollection.uuid_subtype

Cursor slicing

Cursors can no longer be indexed like cursor[n] or sliced like cursor[start:end], see MOTOR-84. If you wrote code like this:

cursor = collection.find()[i]
yield cursor.fetch_next
doc = cursor.next_object()

Then instead, write:

cursor = collection.find().skip(i).limit(-1)
yield cursor.fetch_next
doc = cursor.next_object()

The negative limit ensures the server closes the cursor after one result, saving Motor the work of closing it. See cursor.limit.

SSL hostname validation error

When you use Motor with Tornado and SSL hostname validation fails, Motor used to raise a ConnectionFailure with a useful messsage like “hostname ‘X’ doesn’t match ‘Y’”. The message is now empty and Tornado logs a warning instead.

Configuring uuid_subtype

You can now get and set uuid_subtype on MotorClient, MotorReplicaSetClient, and MotorDatabase instances, not just on MotorCollection.

Motor 0.4.1

Fix MOTOR-66, deadlock when initiating MotorReplicaSetClient connection from multiple operations at once.

Motor 0.4

Supports MongoDB 3.0. In particular, supports MongoDB 3.0’s new SCRAM-SHA-1 authentication mechanism and updates the implementations of MotorClient.database_names() and MotorDatabase.collection_names().

Updates PyMongo dependency from 2.7.1 to 2.8, therefore inheriting PyMongo 2.7.2’s bug fixes and PyMongo 2.8’s bug fixes and features.

Fixes a connection-pool timeout when waitQueueMultipleMS is set and two bugs in replica set monitoring.

The copy_database method has been removed. It was overly complex and no one used it, see MOTOR-56. You can still use the MotorDatabase.command() method directly. The only scenario not supported is copying a database from one host to another, if the remote host requires authentication. For this, use PyMongo’s copy_database method, or, since PyMongo’s copy_database will be removed in a future release too, use the mongo shell.

Motor 0.3.3

Fix MOTOR-45, a stack-context leak in domain name resolution that could lead to an infinite loop and rapid memory leak.

Document Motor’s Requirements in detail.

Motor 0.3.2

Fix MOTOR-44, a socket leak in MotorClient.copy_database and MotorReplicaSetClient.copy_database.

Motor 0.3.1

Fix MOTOR-43, a TypeError when using GridFSHandler with a timezone-aware MotorClient.

Fix GridFS examples that hadn’t been updated for Motor 0.2’s new syntax.

Fix a unittest that hadn’t been running.

Motor 0.3

No new features.

  • Updates PyMongo dependency from 2.7 to 2.7.1, therefore inheriting PyMongo 2.7.1’s bug fixes.
  • Motor continues to support Python 2.6, 2.7, 3.3, and 3.4, but now with single-source. 2to3 no longer runs during installation with Python 3.
  • nosetests is no longer required for regular Motor tests.
  • Fixes a mistake in the docstring for aggregate().

Motor 0.2.1

Fixes two bugs:

  • MOTOR-32: The documentation for MotorCursor.close() claimed it immediately halted execution of MotorCursor.each(), but it didn’t.
  • MOTOR-33: An incompletely iterated cursor’s __del__ method sometimes got stuck and cost 100% CPU forever, even though the application was still responsive.

Motor 0.2

This version includes API changes that break backward compatibility with applications written for Motor 0.1. For most applications, the migration chores will be minor. In exchange, Motor 0.2 offers a cleaner style, and it wraps the new and improved PyMongo 2.7 instead of 2.5.

Changes in Dependencies

Motor now requires PyMongo 2.7.0 exactly and Tornado 3 or later. It drops support for Python 2.5 since Tornado 3 has dropped it.

Motor continues to work with Python 2.6 through 3.4. It still requires Greenlet.

API Changes


The open_sync method has been removed from MotorClient and MotorReplicaSetClient. Clients now connect to MongoDB automatically on first use. Simply delete the call to open_sync from your application.

If it’s important to test that MongoDB is available before continuing your application’s startup, use IOLoop.run_sync:

loop = tornado.ioloop.IOLoop.current()
client = motor.MotorClient(host, port)
except pymongo.errors.ConnectionFailure:
    print "Can't connect"

Similarly, calling is now optional. MotorGridIn and MotorGridFS no longer have an open method at all.


Motor 0.2 takes advantage of Tornado’s tidy new coroutine syntax:

# Old style:
document = yield motor.Op(collection.find_one, {'_id': my_id})

# New style:
document = yield collection.find_one({'_id': my_id})

To make this possible, Motor asynchronous methods (except MotorCursor.each()) now return a Future.

Using Motor with callbacks is still possible: If a callback is passed, it will be executed with the (result, error) of the operation, same as in Motor 0.1:

def callback(document, error):
    if error:
        logging.error("Oh no!")
        print document

collection.find_one({'_id': my_id}, callback=callback)

If no callback is passed, a Future is returned that resolves to the method’s result or error:

document = yield collection.find_one({'_id': my_id})

motor.Op works the same as before, but it’s deprecated.

WaitOp and WaitAllOps have been removed. Code that used them can now yield a Future or a list of them. Consider this function written for Tornado 2 and Motor 0.1:

def get_some_documents():
    cursor = collection.find().sort('_id').limit(2)
    cursor.to_list(callback=(yield gen.Callback('key')))
        documents = yield motor.WaitOp('key')
        print documents
    except Exception, e:
        print e

The function now becomes:

def f():
    cursor = collection.find().sort('_id').limit(2)
    future = cursor.to_list(2)
        documents = yield future
        print documents
    except Exception, e:
        print e

Similarly, a function written like so in the old style:

def get_two_documents_in_parallel(collection):
        {'_id': 1}, callback=(yield gen.Callback('one')))

        {'_id': 2}, callback=(yield gen.Callback('two')))

        doc_one, doc_two = yield motor.WaitAllOps(['one', 'two'])
        print doc_one, doc_two
    except Exception, e:
        print e

Now becomes:

def get_two_documents_in_parallel(collection):
    future_0 = collection.find_one({'_id': 1})
    future_1 = collection.find_one({'_id': 2})

        doc_one, doc_two = yield [future_0, future_1]
        print doc_one, doc_two
    except Exception, e:
        print e


Any calls to MotorCursor.to_list() that omitted the length argument must now include it:

result = yield collection.find().to_list(100)

None is acceptable, meaning “unlimited.” Use with caution.

Connection Pooling

MotorPool has been rewritten. It supports the new options introduced in PyMongo 2.6, and drops all Motor-specific options.

MotorClient and MotorReplicaSetClient have an option max_pool_size. It used to mean “minimum idle sockets to keep open”, but its meaning has changed to “maximum sockets open per host.” Once this limit is reached, operations will pause waiting for a socket to become available. Therefore the default has been raised from 10 to 100. If you pass a value for max_pool_size make sure it’s large enough for the expected load. (Sockets are only opened when needed, so there’s no cost to having a max_pool_size larger than necessary. Err towards a larger value.) If you’ve been accepting the default, continue to do so.

max_pool_size is now synonymous with Motor’s special max_concurrent option, so max_concurrent has been removed.

max_wait_time has been renamed waitQueueTimeoutMS for consistency with PyMongo. If you pass max_wait_time, rename it and multiply by 1000.

The MotorPoolTimeout exception is gone; catch PyMongo’s ConnectionFailure instead.


Motor can take advantage of Tornado 3’s asynchronous resolver interface. By default, Motor still uses blocking DNS, but you can enable non-blocking lookup with a threaded resolver:


Or install pycares and use the c-ares resolver:



The MotorCursor.tail method has been removed. It was complex, diverged from PyMongo’s feature set, and encouraged overuse of MongoDB capped collections as message queues when a purpose-built message queue is more appropriate. An example of tailing a capped collection is provided instead: Motor Tailable Cursor Example.


is_locked has been removed since calling it from Motor would be bizarre. If you called MotorClient.is_locked like:

locked = yield motor.Op(client.is_locked)

you should now do:

result = yield client.admin.current_op()
locked = bool(result.get('fsyncLock', None))

The result is True only if an administrator has called fsyncLock on the mongod. It is unlikely that you have any use for this.


get_gridfs_file() now returns a Future instead of accepting a callback.

New Features

The introduction of a Futures-based API is the most pervasive new feature. In addition Motor 0.2 includes new features from PyMongo 2.6 and 2.7:

Bugfixes threw an error if called without a callback.

MotorCursor.to_list ignored SON manipulators. (Thanks to Eren Güven for the report and the fix.)

The full list is in Jira.

Motor 0.1.2

Fixes innocuous unittest failures when running against Tornado 3.1.1.

Motor 0.1.1

Fixes issue MOTOR-12 by pinning its PyMongo dependency to PyMongo version 2.5.0 exactly.

Motor relies on some of PyMongo’s internal details, so changes to PyMongo can break Motor, and a change in PyMongo 2.5.1 did. Eventually PyMongo will expose stable hooks for Motor to use, but for now I changed Motor’s dependency from PyMongo>=2.4.2 to PyMongo==2.5.0.