Uploading Data

Streaming Multipart Data Encoder

Requests has support for multipart uploads, but the API means that using that functionality to build exactly the Multipart upload you want can be difficult or impossible. Additionally, when using Requests’ Multipart upload functionality all the data must be read into memory before being sent to the server. In extreme cases, this can make it impossible to send a file as part of a multipart/form-data upload.

The toolbelt contains a class that allows you to build multipart request bodies in exactly the format you need, and to avoid reading files into memory. An example of how to use it is like this:

import requests
from requests_toolbelt.multipart.encoder import MultipartEncoder

m = MultipartEncoder(
    fields={'field0': 'value', 'field1': 'value',
            'field2': ('filename', open('file.py', 'rb'), 'text/plain')}
    )

r = requests.post('http://httpbin.org/post', data=m,
                  headers={'Content-Type': m.content_type})

The MultipartEncoder has the .to_string() convenience method, as well. This method renders the multipart body into a string. This is useful when developing your code, allowing you to confirm that the multipart body has the form you expect before you send it on.

The toolbelt also provides a way to monitor your streaming uploads with the MultipartEncoderMonitor.

class requests_toolbelt.multipart.encoder.MultipartEncoder(fields, boundary=None, encoding='utf-8')

The MultipartEncoder object is a generic interface to the engine that will create a multipart/form-data body for you.

The basic usage is:

import requests
from requests_toolbelt import MultipartEncoder

encoder = MultipartEncoder({'field': 'value',
                            'other_field', 'other_value'})
r = requests.post('https://httpbin.org/post', data=encoder,
                  headers={'Content-Type': encoder.content_type})

If you do not need to take advantage of streaming the post body, you can also do:

r = requests.post('https://httpbin.org/post',
                  data=encoder.to_string(),
                  headers={'Content-Type': encoder.content_type})

If you want the encoder to use a specific order, you can use an OrderedDict or more simply, a list of tuples:

encoder = MultipartEncoder([('field', 'value'),
                            ('other_field', 'other_value')])

Changed in version 0.4.0.

You can also provide tuples as part values as you would provide them to requests’ files parameter.

encoder = MultipartEncoder({
    'field': ('file_name', b'{"a": "b"}', 'application/json',
              {'X-My-Header': 'my-value'})
])

Warning

This object will end up directly in httplib. Currently, httplib has a hard-coded read size of 8192 bytes. This means that it will loop until the file has been read and your upload could take a while. This is not a bug in requests. A feature is being considered for this object to allow you, the user, to specify what size should be returned on a read. If you have opinions on this, please weigh in on this issue.

Monitoring Your Streaming Multipart Upload

If you need to stream your multipart/form-data upload then you’re probably in the situation where it might take a while to upload the content. In these cases, it might make sense to be able to monitor the progress of the upload. For this reason, the toolbelt provides the MultipartEncoderMonitor. The monitor wraps an instance of a MultipartEncoder and is used exactly like the encoder. It provides a similar API with some additions:

  • The monitor accepts a function as a callback. The function is called every time requests calls read on the monitor and passes in the monitor as an argument.
  • The monitor tracks how many bytes have been read in the course of the upload.

You might use the monitor to create a progress bar for the upload. Here is an example using clint which displays the progress bar.

To use the monitor you would follow a pattern like this:

import requests
from requests_toolbelt.multipart import encoder

def my_callback(monitor):
    # Your callback function
    pass

e = encoder.MultipartEncoder(
    fields={'field0': 'value', 'field1': 'value',
            'field2': ('filename', open('file.py', 'rb'), 'text/plain')}
    )
m = encoder.MultipartEncoderMonitor(e, my_callback)

r = requests.post('http://httpbin.org/post', data=m,
                  headers={'Content-Type': m.content_type})

If you have a very simple use case you can also do:

import requests
from requests_toolbelt.multipart.encoder import MultipartEncoderMonitor

def my_callback(monitor):
    # Your callback function
    pass

m = MultipartEncoderMonitor.from_fields(
    fields={'field0': 'value', 'field1': 'value',
            'field2': ('filename', open('file.py', 'rb'), 'text/plain')},
    callback=my_callback
    )

r = requests.post('http://httpbin.org/post', data=m,
                  headers={'Content-Type': m.content_type})
class requests_toolbelt.multipart.encoder.MultipartEncoderMonitor(encoder, callback=None)

An object used to monitor the progress of a MultipartEncoder.

The MultipartEncoder should only be responsible for preparing and streaming the data. For anyone who wishes to monitor it, they shouldn’t be using that instance to manage that as well. Using this class, they can monitor an encoder and register a callback. The callback receives the instance of the monitor.

To use this monitor, you construct your MultipartEncoder as you normally would.

from requests_toolbelt import (MultipartEncoder,
                               MultipartEncoderMonitor)
import requests

def callback(encoder, bytes_read):
    # Do something with this information
    pass

m = MultipartEncoder(fields={'field0': 'value0'})
monitor = MultipartEncoderMonitor(m, callback)
headers = {'Content-Type': montior.content_type}
r = requests.post('https://httpbin.org/post', data=monitor,
                  headers=headers)

Alternatively, if your use case is very simple, you can use the following pattern.

from requests_toolbelt import MultipartEncoderMonitor
import requests

def callback(encoder, bytes_read):
    # Do something with this information
    pass

monitor = MultipartEncoderMonitor.from_fields(
    fields={'field0': 'value0'}, callback
    )
headers = {'Content-Type': montior.content_type}
r = requests.post('https://httpbin.org/post', data=monitor,
                  headers=headers)

Streaming Data from a Generator

There are cases where you, the user, have a generator of some large quantity of data and you already know the size of that data. If you pass the generator to requests via the data parameter, requests will assume that you want to upload the data in chunks and set a Transfer-Encoding header value of chunked. Often times, this causes the server to behave poorly. If you want to avoid this, you can use the StreamingIterator. You pass it the size of the data and the generator.

import requests
from requests_toolbelt.streaming_iterator import StreamingIterator

generator = some_function()  # Create your generator
size = some_function_size()  # Get your generator's size
content_type = content_type()  # Get the content-type of the data

streamer = StreamingIterator(size, generator)
r = requests.post('https://httpbin.org/post', data=streamer,
                  headers={'Content-Type': content_type})

The streamer will handle your generator for you and buffer the data before passing it to requests.

Changed in version 0.4.0: File-like objects can be passed instead of a generator.

If, for example, you need to upload data being piped into standard in, you might otherwise do:

import requests
import sys

r = requests.post(url, data=sys.stdin)

This would stream the data but would use a chunked transfer-encoding. If instead, you know the length of the data that is being sent to stdin and you want to prevent the data from being uploaded in chunks, you can use the StreamingIterator to stream the contents of the file without relying on chunking.

import requests
from requests_toolbelt.streaming_iterator import StreamingIterator
import sys

stream = StreamingIterator(size, sys.stdin)
r = requests.post(url, data=stream,
                  headers={'Content-Type': content_type})
class requests_toolbelt.streaming_iterator.StreamingIterator(size, iterator, encoding='utf-8')

This class provides a way of allowing iterators with a known size to be streamed instead of chunked.

In requests, if you pass in an iterator it assumes you want to use chunked transfer-encoding to upload the data, which not all servers support well. Additionally, you may want to set the content-length yourself to avoid this but that will not work. The only way to preempt requests using a chunked transfer-encoding and forcing it to stream the uploads is to mimic a very specific interace. Instead of having to know these details you can instead just use this class. You simply provide the size and iterator and pass the instance of StreamingIterator to requests via the data parameter like so:

from requests_toolbelt import StreamingIterator

import requests

# Let iterator be some generator that you already have and size be
# the size of the data produced by the iterator

r = requests.post(url, data=StreamingIterator(size, iterator))

You can also pass file-like objects to StreamingIterator in case requests can’t determize the filesize itself. This is the case with streaming file objects like stdin or any sockets. Wrapping e.g. files that are on disk with StreamingIterator is unnecessary, because requests can determine the filesize itself.

Naturally, you should also set the Content-Type of your upload appropriately because the toolbelt will not attempt to guess that for you.