This document describes the current stable version of Kombu (5.3). For development docs, go here.



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:

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

The accept argument can also include MIME-types.

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:

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

# Allow serialization of django models:
    lambda o: [o._meta.label,],
    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.

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 PICKLE_PROTOCOL environment variable or by changing the global 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:

    >>> producer = Producer(channel,
    ...                     exchange=exchange,
    ...                     serializer='yaml')
  2. Set the serialization option per message:

    >>> 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 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:

>>> with open('~/my_picture.jpg', 'rb') as fh:
...     producer.publish(,

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

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:

from setuptools import setup

        'kombu.serializers': [
            'my_serializer = my_module.serializer:register_args'

Then the module my_module.serializer would look like:

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.

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 pickle.loads() in such a way:

import pickle
from io import BytesIO
from kombu import serialization

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

    'my_pickle', pickle.dumps, loads,