Using the Twisted Web Client

Overview

This document describes how to use the HTTP client included in Twisted Web. After reading it, you should be able to make HTTP and HTTPS requests using Twisted Web. You will be able to specify the request method, headers, and body and you will be able to retrieve the response code, headers, and body.

A number of higher-level features are also explained, including proxying, automatic content encoding negotiation, and cookie handling.

Prerequisites

This document assumes that you are familiar with Deferreds and Failures , and producers and consumers . It also assumes you are familiar with the basic concepts of HTTP, such as requests and responses, methods, headers, and message bodies. The HTTPS section of this document also assumes you are somewhat familiar with SSL and have read about using SSL in Twisted .

The Agent

Issuing Requests

The twisted.web.client.Agent class is the entry point into the client API. Requests are issued using the request method, which takes as parameters a request method, a request URI, the request headers, and an object which can produce the request body (if there is to be one). The agent is responsible for connection setup. Because of this, it requires a reactor as an argument to its initializer. An example of creating an agent and issuing a request using it might look like this:

request.py

from twisted.internet import reactor
from twisted.web.client import Agent
from twisted.web.http_headers import Headers

agent = Agent(reactor)

d = agent.request(
    b"GET",
    b"http://httpbin.com/anything",
    Headers({"User-Agent": ["Twisted Web Client Example"]}),
    None,
)


def cbResponse(ignored):
    print("Response received")


d.addCallback(cbResponse)


def cbShutdown(ignored):
    reactor.stop()


d.addBoth(cbShutdown)

reactor.run()

As may be obvious, this issues a new GET request for / to the web server on example.com . Agent is responsible for resolving the hostname into an IP address and connecting to it on port 80 (for HTTP URIs), port 443 (for HTTPS URIs), or on the port number specified in the URI itself. It is also responsible for cleaning up the connection afterwards. This code sends a request which includes one custom header, User-Agent . The last argument passed to Agent.request is None , though, so the request has no body.

Sending a request which does include a body requires passing an object providing twisted.web.iweb.IBodyProducer to Agent.request . This interface extends the more general IPushProducer by adding a new length attribute and adding several constraints to the way the producer and consumer interact.

  • The length attribute must be a non-negative integer or the constant twisted.web.iweb.UNKNOWN_LENGTH . If the length is known, it will be used to specify the value for the Content-Length header in the request. If the length is unknown the attribute should be set to UNKNOWN_LENGTH . Since more servers support Content-Length , if a length can be provided it should be.

  • An additional method is required on IBodyProducer implementations: startProducing . This method is used to associate a consumer with the producer. It should return a Deferred which fires when all data has been produced.

  • IBodyProducer implementations should never call the consumer’s unregisterProducer method. Instead, when it has produced all of the data it is going to produce, it should only fire the Deferred returned by startProducing .

For additional details about the requirements of IBodyProducer implementations, see the API documentation.

Here’s a simple IBodyProducer implementation which writes an in-memory string to the consumer:

bytesprod.py

from zope.interface import implementer

from twisted.internet.defer import succeed
from twisted.web.iweb import IBodyProducer


@implementer(IBodyProducer)
class BytesProducer:
    def __init__(self, body):
        self.body = body
        self.length = len(body)

    def startProducing(self, consumer):
        consumer.write(self.body)
        return succeed(None)

    def pauseProducing(self):
        pass

    def stopProducing(self):
        pass

This producer can be used to issue a request with a body:

sendbody.py

from bytesprod import BytesProducer

from twisted.internet import reactor
from twisted.web.client import Agent
from twisted.web.http_headers import Headers

agent = Agent(reactor)
body = BytesProducer(b"hello, world")
d = agent.request(
    b"POST",
    b"http://httpbin.org/post",
    Headers(
        {
            "User-Agent": ["Twisted Web Client Example"],
            "Content-Type": ["text/x-greeting"],
        }
    ),
    body,
)


def cbResponse(ignored):
    print("Response received")


d.addCallback(cbResponse)


def cbShutdown(ignored):
    reactor.stop()


d.addBoth(cbShutdown)

reactor.run()

If you want to upload a file or you just have some data in a string, you don’t have to copy StringProducer though. Instead, you can use FileBodyProducer . This IBodyProducer implementation works with any file-like object (so use it with a StringIO if your upload data is already in memory as a string); the idea is the same as StringProducer from the previous example, but with a little extra code to only send data as fast as the server will take it.

filesendbody.py

from io import BytesIO

from twisted.internet import reactor
from twisted.web.client import Agent, FileBodyProducer
from twisted.web.http_headers import Headers

agent = Agent(reactor)
body = FileBodyProducer(BytesIO(b"hello, world"))
d = agent.request(
    b"GET",
    b"http://example.com/",
    Headers(
        {
            "User-Agent": ["Twisted Web Client Example"],
            "Content-Type": ["text/x-greeting"],
        }
    ),
    body,
)


def cbResponse(ignored):
    print("Response received")


d.addCallback(cbResponse)


def cbShutdown(ignored):
    reactor.stop()


d.addBoth(cbShutdown)

reactor.run()

FileBodyProducer closes the file when it no longer needs it.

If the connection or the request take too much time, you can cancel the Deferred returned by the Agent.request method. This will abort the connection, and the Deferred will errback with CancelledError .

Receiving Responses

So far, the examples have demonstrated how to issue a request. However, they have ignored the response, except for showing that it is a Deferred which seems to fire when the response has been received. Next we’ll cover what that response is and how to interpret it.

Agent.request , as with most Deferred -returning APIs, can return a Deferred which fires with a Failure . If the request fails somehow, this will be reflected with a failure. This may be due to a problem looking up the host IP address, or it may be because the HTTP server is not accepting connections, or it may be because of a problem parsing the response, or any other problem which arises which prevents the response from being received. It does not include responses with an error status.

If the request succeeds, though, the Deferred will fire with a Response . This happens as soon as all the response headers have been received. It happens before any of the response body, if there is one, is processed. The Response object has several attributes giving the response information: its code, version, phrase, and headers, as well as the length of the body to expect. In addition to these, the Response also contains a reference to the request that it is a response to; one particularly useful attribute on the request is absoluteURI : The absolute URI to which the request was made. The Response object has a method which makes the response body available: deliverBody . Using the attributes of the response object and this method, here’s an example which displays part of the response to a request:

response.py

from pprint import pformat

from twisted.internet import reactor
from twisted.internet.defer import Deferred
from twisted.internet.protocol import Protocol
from twisted.web.client import Agent
from twisted.web.http_headers import Headers


class BeginningPrinter(Protocol):
    def __init__(self, finished):
        self.finished = finished
        self.remaining = 1024 * 10

    def dataReceived(self, bytes):
        if self.remaining:
            display = bytes[: self.remaining]
            print("Some data received:")
            print(display)
            self.remaining -= len(display)

    def connectionLost(self, reason):
        print("Finished receiving body:", reason.getErrorMessage())
        self.finished.callback(None)


agent = Agent(reactor)
d = agent.request(
    b"GET",
    b"http://httpbin.com/anything/",
    Headers({"User-Agent": ["Twisted Web Client Example"]}),
    None,
)


def cbRequest(response):
    print("Response version:", response.version)
    print("Response code:", response.code)
    print("Response phrase:", response.phrase)
    print("Response headers:")
    print(pformat(list(response.headers.getAllRawHeaders())))
    finished = Deferred()
    response.deliverBody(BeginningPrinter(finished))
    return finished


d.addCallback(cbRequest)


def cbShutdown(ignored):
    reactor.stop()


d.addBoth(cbShutdown)

reactor.run()

The BeginningPrinter protocol in this example is passed to Response.deliverBody and the response body is then delivered to its dataReceived method as it arrives. When the body has been completely delivered, the protocol’s connectionLost method is called. It is important to inspect the Failure passed to connectionLost . If the response body has been completely received, the failure will wrap a twisted.web.client.ResponseDone exception. This indicates that it is known that all data has been received. It is also possible for the failure to wrap a twisted.web.http.PotentialDataLoss exception: this indicates that the server framed the response such that there is no way to know when the entire response body has been received. Only HTTP/1.0 servers should behave this way. Finally, it is possible for the exception to be of another type, indicating guaranteed data loss for some reason (a lost connection, a memory error, etc).

Just as protocols associated with a TCP connection are given a transport, so will be a protocol passed to deliverBody . Since it makes no sense to write more data to the connection at this stage of the request, though, the transport only provides IPushProducer . This allows the protocol to control the flow of the response data: a call to the transport’s pauseProducing method will pause delivery; a later call to resumeProducing will resume it. If it is decided that the rest of the response body is not desired, stopProducing can be used to stop delivery permanently; after this, the protocol’s connectionLost method will be called.

An important thing to keep in mind is that the body will only be read from the connection after Response.deliverBody is called. This also means that the connection will remain open until this is done (and the body read). So, in general, any response with a body must have that body read using deliverBody . If the application is not interested in the body, it should issue a HEAD request or use a protocol which immediately calls stopProducing on its transport.

If the body of the response isn’t going to be consumed incrementally, then readBody can be used to get the body as a byte-string. This function returns a Deferred that fires with the body after the request has been completed; cancelling this Deferred will close the connection to the HTTP server immediately.

responseBody.py

from pprint import pformat
from sys import argv

from twisted.internet.task import react
from twisted.web.client import Agent, readBody
from twisted.web.http_headers import Headers


def cbRequest(response):
    print("Response version:", response.version)
    print("Response code:", response.code)
    print("Response phrase:", response.phrase)
    print("Response headers:")
    print(pformat(list(response.headers.getAllRawHeaders())))
    d = readBody(response)
    d.addCallback(cbBody)
    return d


def cbBody(body):
    print("Response body:")
    print(body)


def main(reactor, url=b"http://httpbin.org/get"):
    agent = Agent(reactor)
    d = agent.request(
        b"GET", url, Headers({"User-Agent": ["Twisted Web Client Example"]}), None
    )
    d.addCallback(cbRequest)
    return d


react(main, argv[1:])

Customizing your HTTPS Configuration

Everything you’ve read so far applies whether the scheme of the request URI is HTTP or HTTPS .

Since version 15.0.0, Twisted’s Agent HTTP client will validate https URLs against your platform’s trust store by default.

Note

If you’re using an earlier version, please upgrade, as this is a critical security feature!

Note

Only Agent, and things that use it, validates HTTPS. Do not use getPage to retrieve HTTPS URLs; although we have not yet removed all the examples that use it, we discourage its use.

You should pip install twisted[tls] in order to get all the dependencies necessary to do TLS properly. The installation instructions gives more detail on optional dependencies and how to install them which may be of interest.

For some uses, you may need to customize Agent’s use of HTTPS; for example, to provide a client certificate, or to use a custom certificate authority for an internal network.

Here, we’re just going to show you how to inject the relevant TLS configuration into an Agent, so we’ll use the simplest possible example, rather than a more useful but complex one.

Agent’s constructor takes an optional second argument, which allows you to customize its behavior with respect to HTTPS. The object passed here must provide the twisted.web.iweb.IPolicyForHTTPS interface.

Using the very helpful badssl.com web API, we will construct a request that fails to validate, because the certificate has the wrong hostname in it. The following Python program should produce a verification error when run as python example.py https://wrong.host.badssl.com/.

import sys

from twisted.internet.task import react

from twisted.web.client import Agent, ResponseFailed

@react
def main(reactor):
    agent = Agent(reactor)
    requested = agent.request(b"GET", sys.argv[1].encode("ascii"))
    def gotResponse(response):
        print(response.code)
    def noResponse(failure):
        failure.trap(ResponseFailed)
        print(failure.value.reasons[0].getTraceback())
    return requested.addCallbacks(gotResponse, noResponse)

The verification error you see when running the above program is due to the mismatch between “wrong.host.badssl.com”, the expected hostname derived from the URL, and “badssl.com”, the hostname contained in the certificate presented by the server. In our pretend scenario, we want to construct an Agent that validates HTTPS certificates normally, _except_ for this one host. For "wrong.host.badssl.com", we wish to supply the correct hostname to check the certificate against manually, to work around this problem. In order to do that, we will supply our own policy that creates a different TLS configuration depending on the hostname, like so:

import sys

from zope.interface import implementer

from twisted.internet.task import react
from twisted.internet.ssl import optionsForClientTLS

from twisted.web.iweb import IPolicyForHTTPS
from twisted.web.client import Agent, ResponseFailed, BrowserLikePolicyForHTTPS

@implementer(IPolicyForHTTPS)
class OneHostnameWorkaroundPolicy(object):
    def __init__(self):
        self._normalPolicy = BrowserLikePolicyForHTTPS()
    def creatorForNetloc(self, hostname, port):
        if hostname == b"wrong.host.badssl.com":
            hostname = b"badssl.com"
        return self._normalPolicy.creatorForNetloc(hostname, port)

@react
def main(reactor):
    agent = Agent(reactor, OneHostnameWorkaroundPolicy())
    requested = agent.request(b"GET", sys.argv[1].encode("ascii"))
    def gotResponse(response):
        print(response.code)
    def noResponse(failure):
        failure.trap(ResponseFailed)
        print(failure.value.reasons[0].getTraceback())
    return requested.addCallbacks(gotResponse, noResponse)

Now, invoking python example.py https://wrong.host.badssl.com/ will happily give us a 200 status code; however, running it with https://expired.badssl.com/ or https://self-signed.badssl.com/ or any of the other error hostnames should still give an error.

Note

The technique presented above does not ignore the mismatching hostname, but rather, it provides the erroneous hostname specifically so that the misconfiguration is expected. Twisted does not document any facilities for disabling verification, since that makes TLS useless; instead, we strongly encourage you to figure out what properties you need from the connection and verify those.

Using this TLS policy mechanism, you can customize Agent to use any feature of TLS that Twisted has support for, including the examples given above; client certificates, alternate trust roots, and so on. For a more detailed explanation of what options exist for client TLS configuration in Twisted, check out the documentation for the optionsForClientTLS API and the Using SSL in Twisted chapter of this documentation.

HTTP Persistent Connection

HTTP persistent connections use the same TCP connection to send and receive multiple HTTP requests/responses. This reduces latency and TCP connection establishment overhead.

The constructor of twisted.web.client.Agent takes an optional parameter pool, which should be an instance of HTTPConnectionPool , which will be used to manage the connections. If the pool is created with the parameter persistent set to True (the default), it will not close connections when the request is done, and instead hold them in its cache to be re-used.

Here’s an example which sends requests over a persistent connection:

from twisted.internet import reactor
from twisted.internet.defer import Deferred, DeferredList
from twisted.internet.protocol import Protocol
from twisted.web.client import Agent, HTTPConnectionPool

class IgnoreBody(Protocol):
    def __init__(self, deferred):
        self.deferred = deferred

    def dataReceived(self, bytes):
        pass

    def connectionLost(self, reason):
        self.deferred.callback(None)


def cbRequest(response):
    print('Response code:', response.code)
    finished = Deferred()
    response.deliverBody(IgnoreBody(finished))
    return finished

pool = HTTPConnectionPool(reactor)
agent = Agent(reactor, pool=pool)

def requestGet(url):
    d = agent.request('GET', url)
    d.addCallback(cbRequest)
    return d

# Two requests to the same host:
d = requestGet('http://localhost:8080/foo').addCallback(
    lambda ign: requestGet("http://localhost:8080/bar"))
def cbShutdown(ignored):
    reactor.stop()
d.addCallback(cbShutdown)

reactor.run()

Here, the two requests are to the same host, one after the each other. In most cases, the same connection will be used for the second request, instead of two different connections when using a non-persistent pool.

Multiple Connections to the Same Server

twisted.web.client.HTTPConnectionPool instances have an attribute called maxPersistentPerHost which limits the number of cached persistent connections to the same server. The default value is 2. This is effective only when the persistent option is True. You can change the value like bellow:

from twisted.web.client import HTTPConnectionPool

pool = HTTPConnectionPool(reactor, persistent=True)
pool.maxPersistentPerHost = 1

With the default value of 2, the pool keeps around two connections to the same host at most. Eventually the cached persistent connections will be closed, by default after 240 seconds; you can change this timeout value with the cachedConnectionTimeout attribute of the pool. To force all connections to close use the closeCachedConnections method.

Automatic Retries

If a request fails without getting a response, and the request is something that hopefully can be retried without having any side-effects (e.g. a request with method GET), it will be retried automatically when sending a request over a previously-cached persistent connection. You can disable this behavior by setting retryAutomatically to False . Note that each request will only be retried once.

Following redirects

By itself, Agent doesn’t follow HTTP redirects (responses with 301, 302, 303, 307 status codes and a location header field). You need to use the twisted.web.client.RedirectAgent class to do so. It implements a rather strict behavior of the RFC, meaning it will redirect 301 and 302 as 307, only on GET and HEAD requests.

The following example shows how to have a redirect-enabled agent.

from twisted.python.log import err
from twisted.web.client import Agent, RedirectAgent
from twisted.internet import reactor

def display(response):
    print("Received response")
    print(response)

def main():
    agent = RedirectAgent(Agent(reactor))
    d = agent.request("GET", "http://example.com/")
    d.addCallbacks(display, err)
    d.addCallback(lambda ignored: reactor.stop())
    reactor.run()

if __name__ == "__main__":
    main()

In contrast, twisted.web.client.BrowserLikeRedirectAgent implements more lenient behaviour that closely emulates what web browsers do; in other words 301 and 302 POST redirects are treated like 303, meaning the method is changed to GET before making the redirect request.

As mentioned previously, Response contains a reference to both the request that it is a response to, and the previously received response , accessible by previousResponse . In most cases there will not be a previous response, but in the case of RedirectAgent the response history can be obtained by following the previous responses from response to response.

Using a HTTP proxy

To be able to use HTTP proxies with an agent, you can use the twisted.web.client.ProxyAgent class. It supports the same interface as Agent, but takes the endpoint of the proxy as initializer argument. This is specifically intended for talking to servers that implement the proxying variation of the HTTP protocol; for other types of proxies you will want Agent.usingEndpointFactory (see documentation below).

Here’s an example demonstrating the use of an HTTP proxy running on localhost:8000.

from twisted.python.log import err
from twisted.web.client import ProxyAgent
from twisted.internet import reactor
from twisted.internet.endpoints import TCP4ClientEndpoint

def display(response):
    print("Received response")
    print(response)

def main():
    endpoint = TCP4ClientEndpoint(reactor, "localhost", 8000)
    agent = ProxyAgent(endpoint)
    d = agent.request("GET", "https://example.com/")
    d.addCallbacks(display, err)
    d.addCallback(lambda ignored: reactor.stop())
    reactor.run()

if __name__ == "__main__":
    main()

Please refer to the endpoints documentation for more information about how they work and the twisted.internet.endpoints API documentation to learn what other kinds of endpoints exist.

Handling HTTP cookies

An existing agent instance can be wrapped with twisted.web.client.CookieAgent to automatically store, send and track HTTP cookies. A CookieJar instance, from the Python standard library module cookielib , is used to store the cookie information. An example of using CookieAgent to perform a request and display the collected cookies might look like this:

cookies.py

from twisted.internet import reactor
from twisted.python import compat, log
from twisted.web.client import Agent, CookieAgent


def displayCookies(response, cookieJar):
    print("Received response")
    print(response)
    print("Cookies:", len(cookieJar))
    for cookie in cookieJar:
        print(cookie)


def main():
    cookieJar = compat.cookielib.CookieJar()
    agent = CookieAgent(Agent(reactor), cookieJar)

    d = agent.request(b"GET", b"http://httpbin.org/cookies/set?some=data")
    d.addCallback(displayCookies, cookieJar)
    d.addErrback(log.err)
    d.addCallback(lambda ignored: reactor.stop())
    reactor.run()


if __name__ == "__main__":
    main()

Automatic Content Encoding Negotiation

twisted.web.client.ContentDecoderAgent adds support for sending Accept-Encoding request headers and interpreting Content-Encoding response headers. These headers allow the server to encode the response body somehow, typically with some compression scheme to save on transfer costs. ContentDecoderAgent provides this functionality as a wrapper around an existing agent instance. Together with one or more decoder objects (such as twisted.web.client.GzipDecoder ), this wrapper automatically negotiates an encoding to use and decodes the response body accordingly. To application code using such an agent, there is no visible difference in the data delivered.

gzipdecoder.py

from twisted.internet import reactor
from twisted.internet.defer import Deferred
from twisted.internet.protocol import Protocol
from twisted.python import log
from twisted.web.client import Agent, ContentDecoderAgent, GzipDecoder


class BeginningPrinter(Protocol):
    def __init__(self, finished):
        self.finished = finished
        self.remaining = 1024 * 10

    def dataReceived(self, bytes):
        if self.remaining:
            display = bytes[: self.remaining]
            print("Some data received:")
            print(display)
            self.remaining -= len(display)

    def connectionLost(self, reason):
        print("Finished receiving body:", reason.type, reason.value)
        self.finished.callback(None)


def printBody(response):
    finished = Deferred()
    response.deliverBody(BeginningPrinter(finished))
    return finished


def main():
    agent = ContentDecoderAgent(Agent(reactor), [(b"gzip", GzipDecoder)])

    d = agent.request(b"GET", b"http://httpbin.org/gzip")
    d.addCallback(printBody)
    d.addErrback(log.err)
    d.addCallback(lambda ignored: reactor.stop())
    reactor.run()


if __name__ == "__main__":
    main()

Implementing support for new content encodings is as simple as writing a new class like GzipDecoder that can decode a response using the new encoding. As there are not many content encodings in widespread use, gzip is the only encoding supported by Twisted itself.

Connecting To Non-standard Destinations

Typically you want your HTTP client to open a TCP connection directly to the web server. Sometimes however it’s useful to be able to connect some other way, e.g. making an HTTP request over a SOCKS proxy connection or connecting to a server listening on a UNIX socket. For this reason, there is an alternate constructor called Agent.usingEndpointFactory that takes an endpointFactory argument. This argument must provide the twisted.web.iweb.IAgentEndpointFactory interface. Note that when talking to a HTTP proxy, i.e. a server that implements the proxying-specific variant of HTTP you should use ProxyAgent - see documentation above.

endpointconstructor.py

# Copyright (c) Twisted Matrix Laboratories.
# See LICENSE for details.

"""
Send a HTTP request to Docker over a Unix socket.

Will probably need to be run as root.

Usage:
    $ sudo python endpointconstructor.py [<docker API path>]
"""


from sys import argv

from zope.interface import implementer

from twisted.internet.endpoints import UNIXClientEndpoint
from twisted.internet.task import react
from twisted.web.client import Agent, readBody
from twisted.web.iweb import IAgentEndpointFactory


@implementer(IAgentEndpointFactory)
class DockerEndpointFactory:
    """
    Connect to Docker's Unix socket.
    """

    def __init__(self, reactor):
        self.reactor = reactor

    def endpointForURI(self, uri):
        return UNIXClientEndpoint(self.reactor, b"/var/run/docker.sock")


def main(reactor, path=b"/containers/json?all=1"):
    agent = Agent.usingEndpointFactory(reactor, DockerEndpointFactory(reactor))
    d = agent.request(b"GET", b"unix://localhost" + path)
    d.addCallback(readBody)
    d.addCallback(print)
    return d


react(main, argv[1:])

Conclusion

You should now understand the basics of the Twisted Web HTTP client. In particular, you should understand:

  • How to issue requests with arbitrary methods, headers, and bodies.

  • How to access the response version, code, phrase, headers, and body.

  • How to store, send, and track cookies.

  • How to control the streaming of the response body.

  • How to enable the HTTP persistent connection, and control the number of connections.