kombu/docs/userguide/serialization.rst

.. _guide-serialization:

===============
 Serialization
===============

.. _serializers:

Serializers
===========

By default every message is encoded using `JSON`_, so sending
Python data structures like dictionaries and lists works.
`YAML`_, `msgpack`_ and Python's built-in `pickle` module is also supported,
and if needed you can register any custom serialization scheme you
want to use.


By default Kombu will only load JSON messages, so if you want
to use other serialization format you must explicitly enable
them in your consumer by using the ``accept`` argument:

.. code-block:: python

    Consumer(conn, [queue], accept=['json', 'pickle', 'msgpack'])

The accept argument can also include MIME-types.

.. _`JSON`: http://www.json.org/
.. _`YAML`: http://yaml.org/
.. _`msgpack`: https://msgpack.org/

Each option has its advantages and disadvantages.

`json` -- JSON is supported in many programming languages, is
    a standard part of Python, and is fairly fast to
    decode.

    The primary disadvantage to `JSON` is that it limits you to
    the following data types: strings, Unicode, floats, boolean,
    dictionaries, lists, decimals, DjangoPromise, datetimes, dates,
    time, bytes and UUIDs.

    For dates, datetimes, UUIDs and bytes the serializer will generate
    a dict that will later instruct the deserializer how to produce
    the right type.

    Also, binary data will be transferred using Base64 encoding, which
    will cause the transferred data to be around 34% larger than an
    encoding which supports native binary types. This will only happen
    if the bytes object can't be decoded into utf8.

    However, if your data fits inside the above constraints and
    you need cross-language support, the default setting of `JSON`
    is probably your best choice.

    If you need support for custom types, you can write serialize/deserialize
    functions and register them as follows:

    .. code-block:: python

        from kombu.utils.json import register_type
        from django.db.models import Model
        from django.apps import apps

        # Allow serialization of django models:
        register_type(
            Model,
            "model",
            lambda o: [o._meta.label, o.pk],
            lambda o: apps.get_model(o[0]).objects.get(pk=o[1]),
        )

`pickle` -- If you have no desire to support any language other than
    Python, then using the `pickle` encoding will gain you
    the support of all built-in Python data types (except class instances),
    smaller messages when sending binary files, and a slight speedup
    over `JSON` processing.

    .. admonition:: Pickle and Security

        The pickle format is very convenient as it can serialize
        and deserialize almost any object, but this is also a concern
        for security.

        Carefully crafted pickle payloads can do almost anything
        a regular Python program can do, so if you let your consumer
        automatically decode pickled objects you must make sure
        to limit access to the broker so that untrusted
        parties do not have the ability to send messages!

    By default Kombu uses pickle protocol 4, but this can be changed
    using the :envvar:`PICKLE_PROTOCOL` environment variable or by changing
    the global :data:`kombu.serialization.pickle_protocol` flag.

`yaml` -- YAML has many of the same characteristics as `json`,
    except that it natively supports more data types (including dates,
    recursive references, etc.)

    However, the Python libraries for YAML are a good bit slower
    than the libraries for JSON.

    If you need a more expressive set of data types and need to maintain
    cross-language compatibility, then `YAML` may be a better fit
    than the above.

To instruct `Kombu` to use an alternate serialization method,
use one of the following options.

    1. Set the serialization option on a per-producer basis:

        .. code-block:: pycon

            >>> producer = Producer(channel,
            ...                     exchange=exchange,
            ...                     serializer='yaml')

    2. Set the serialization option per message:

        .. code-block:: pycon

            >>> producer.publish(message, routing_key=rkey,
            ...                  serializer='pickle')

Note that a `Consumer` do not need the serialization method specified.
They can auto-detect the serialization method as the
content-type is sent as a message header.

.. _sending-raw-data:

Sending raw data without Serialization
======================================

In some cases, you don't need your message data to be serialized. If you
pass in a plain string or Unicode object as your message and a custom `content_type`, then `Kombu` will
not waste cycles serializing/deserializing the data.

You can optionally specify a `content_encoding`
for the raw data:

.. code-block:: pycon

    >>> with open('~/my_picture.jpg', 'rb') as fh:
    ...     producer.publish(fh.read(),
                             content_type='image/jpeg',
                             content_encoding='binary',
                             routing_key=rkey)

The `Message` object returned by the `Consumer` class will have a
`content_type` and `content_encoding` attribute.

.. _serialization-entrypoints:

Creating extensions using Setuptools entry-points
=================================================

A package can also register new serializers using Setuptools
entry-points.

The entry-point must provide the name of the serializer along
with the path to a tuple providing the rest of the args:
``encoder_function, decoder_function, content_type, content_encoding``.

An example entrypoint could be:

.. code-block:: python

    from setuptools import setup

    setup(
        entry_points={
            'kombu.serializers': [
                'my_serializer = my_module.serializer:register_args'
            ]
        }
    )


Then the module ``my_module.serializer`` would look like:

.. code-block:: python

    register_args = (my_encoder, my_decoder, 'application/x-mimetype', 'utf-8')


When this package is installed the new 'my_serializer' serializer will be
supported by Kombu.


.. admonition:: Buffer Objects

    The decoder function of custom serializer must support both strings
    and Python's old-style buffer objects.

    Python pickle and json modules usually don't do this via its ``loads``
    function, but you can easily add support by making a wrapper around the
    ``load`` function that takes file objects instead of strings.

    Here's an example wrapping :func:`pickle.loads` in such a way:

    .. code-block:: python

        import pickle
        from io import BytesIO
        from kombu import serialization


        def loads(s):
            return pickle.load(BytesIO(s))

        serialization.register(
            'my_pickle', pickle.dumps, loads,
            content_type='application/x-pickle2',
            content_encoding='binary',
        )