VCR.py

This is a Python version of Ruby's VCR library.

What it does

Simplify and speed up testing HTTP by recording all HTTP interactions and saving them to "cassette" files, which are yaml files containing the contents of your requests and responses. Then when you run your tests again, they all just hit the text files instead of the internet. This speeds up your tests and lets you work offline.

If the server you are testing against ever changes its API, all you need to do is delete your existing cassette files, and run your tests again. All of the mocked responses will be updated with the new API.

Compatibility Notes

VCR.py supports Python 2.6 and 2.7, 3.3, 3.4, and pypy.

The following http libraries are supported:

• urllib2
• urllib3
• http.client (python3)
• requests (both 1.x and 2.x versions)
• httplib2
• boto

Usage

import vcr
import urllib2

with vcr.use_cassette('fixtures/vcr_cassettes/synopsis.yaml'):
    response = urllib2.urlopen('http://www.iana.org/domains/reserved').read()
    assert 'Example domains' in response

Run this test once, and VCR.py will record the HTTP request to fixtures/vcr_cassettes/synopsis.yml. Run it again, and VCR.py will replay the response from iana.org when the http request is made. This test is now fast (no real HTTP requests are made anymore), deterministic (the test will continue to pass, even if you are offline, or iana.org goes down for maintenance) and accurate (the response will contain the same headers and body you get from a real request).

You can also use VCR.py as a decorator. The same request above would look like this:

@vcr.use_cassette('fixtures/vcr_cassettes/synopsis.yaml')
def test_iana():
    response = urllib2.urlopen('http://www.iana.org/domains/reserved').read()
    assert 'Example domains' in response

All of the parameters and configuration works the same for the decorator version.

Configuration

If you don't like VCR's defaults, you can set options by instantiating a VCR class and setting the options on it.

import vcr

my_vcr = vcr.VCR(
    serializer = 'json',
    cassette_library_dir = 'fixtures/cassettes',
    record_mode = 'once',
    match_on = ['uri', 'method'],
)

with my_vcr.use_cassette('test.json'):
    # your http code here

Otherwise, you can override options each time you use a cassette.

with vcr.use_cassette('test.yml', serializer='json', record_mode='once'):
    # your http code here

Note: Per-cassette overrides take precedence over the global config.

Request matching

Request matching is configurable and allows you to change which requests VCR considers identical. The default behavior is ['method', 'scheme', 'host', 'port', 'path', 'query'] which means that requests with both the same URL and method (ie POST or GET) are considered identical.

This can be configured by changing the match_on setting.

The following options are available :

• method (for example, POST or GET)
• uri (the full URI.)
• host (the hostname of the server receiving the request)
• port (the port of the server receiving the request)
• path (the path of the request)
• query (the query string of the request)
• body (the entire request body)
• headers (the headers of the request)

Backwards compatible matchers:

• url (the uri alias)

If these options don't work for you, you can also register your own request matcher. This is described in the Advanced section of this README.

Record Modes

VCR supports 4 record modes (with the same behavior as Ruby's VCR):

once

• Replay previously recorded interactions.
• Record new interactions if there is no cassette file.
• Cause an error to be raised for new requests if there is a cassette file.

It is similar to the new_episodes record mode, but will prevent new, unexpected requests from being made (i.e. because the request URI changed).

once is the default record mode, used when you do not set one.

new_episodes

• Record new interactions.
• Replay previously recorded interactions. It is similar to the once record mode, but will always record new interactions, even if you have an existing recorded one that is similar, but not identical.

This was the default behavior in versions < 0.3.0

none

• Replay previously recorded interactions.
• Cause an error to be raised for any new requests. This is useful when your code makes potentially dangerous HTTP requests. The none record mode guarantees that no new HTTP requests will be made.

all

• Record new interactions.
• Never replay previously recorded interactions. This can be temporarily used to force VCR to re-record a cassette (i.e. to ensure the responses are not out of date) or can be used when you simply want to log all HTTP requests.

Advanced Features

If you want, VCR.py can return information about the cassette it is using to record your requests and responses. This will let you record your requests and responses and make assertions on them, to make sure that your code under test is generating the expected requests and responses. This feature is not present in Ruby's VCR, but I think it is a nice addition. Here's an example:

import vcr
import urllib2

with vcr.use_cassette('fixtures/vcr_cassettes/synopsis.yaml') as cass:
    response = urllib2.urlopen('http://www.zombo.com/').read()
    # cass should have 1 request inside it
    assert len(cass) == 1 
    # the request uri should have been http://www.zombo.com/
    assert cass.requests[0].uri == 'http://www.zombo.com/'

The Cassette object exposes the following properties which I consider part of the API. The fields are as follows:

• requests: A list of vcr.Request objects corresponding to the http requests that were made during the recording of the cassette. The requests appear in the order that they were originally processed.
• responses: A list of the responses made.
• play_count: The number of times this cassette has played back a response.
• all_played: A boolean indicating whether all the responses have been played back.
• responses_of(request): Access the responses that match a given request

The Request object has the following properties:

• uri: The full uri of the request. Example: "https://google.com/?q=vcrpy"
• scheme: The scheme used to make the request (http or https)
• host: The host of the request, for example "www.google.com"
• port: The port the request was made on
• path: The path of the request. For example "/" or "/home.html"
• query: The parsed query string of the request. Sorted list of name, value pairs.
• method : The method used to make the request, for example "GET" or "POST"
• body: The body of the request, usually empty except for POST / PUT / etc

Backwards compatible properties:

• url: The uri alias
• protocol: The scheme alias

Register your own serializer

Don't like JSON or YAML? That's OK, VCR.py can serialize to any format you would like. Create your own module or class instance with 2 methods:

• def deserialize(cassette_string)
• def serialize(cassette_dict)

Finally, register your class with VCR to use your new serializer.

import vcr

class BogoSerializer(object):
    """
    Must implement serialize() and deserialize() methods
    """
    pass

my_vcr = vcr.VCR()
my_vcr.register_serializer('bogo', BogoSerializer())

with my_vcr.use_cassette('test.bogo', serializer='bogo'):
    # your http here

# After you register, you can set the default serializer to your new serializer

my_vcr.serializer = 'bogo'

with my_vcr.use_cassette('test.bogo'):
    # your http here

Register your own request matcher

Create your own method with the following signature

def my_matcher(r1, r2):

Your method receives the two requests and must return True if they match, False if they don't.

Finally, register your method with VCR to use your new request matcher.

import vcr

def jurassic_matcher(r1, r2):
    return r1.uri == r2.uri and 'JURASSIC PARK' in r1.body

my_vcr = vcr.VCR()
my_vcr.register_matcher('jurassic', jurassic_matcher)

with my_vcr.use_cassette('test.yml', match_on=['jurassic']):
    # your http here

# After you register, you can set the default match_on to use your new matcher

my_vcr.match_on = ['jurassic']

with my_vcr.use_cassette('test.yml'):
    # your http here

Filter sensitive data from the request

If you are checking your cassettes into source control, and are using some form of authentication in your tests, you can filter out that information so it won't appear in your cassette files. There are a few ways to do this:

Filter information from HTTP Headers

Use the filter_headers configuration option with a list of headers to filter.

with my_vcr.use_cassette('test.yml', filter_headers=['authorization']):
    # sensitive HTTP request goes here

Filter information from HTTP querystring

Use the filter_query_parameters configuration option with a list of query parameters to filter.

with my_vcr.use_cassette('test.yml', filter_query_parameters=['api_key']):
    requests.get('http://api.com/getdata?api_key=secretstring')

Filter information from HTTP post data

Use the filter_post_data_parameters configuration option with a list of post data parameters to filter.

with my_vcr.use_cassette('test.yml', filter_post_data_parameters=['client_secret']):
    requests.post('http://api.com/postdata', data={'api_key': 'secretstring'})

Custom Request filtering

If none of these covers your request filtering needs, you can register a callback that will manipulate the HTTP request before adding it to the cassette. Use the before_record configuration option to so this. Here is an example that will never record requests to the /login endpoint.

def before_record_cb(request):
    if request.path != '/login':
        return request

my_vcr = vcr.VCR(
    before_record = before_record_cb,
)
with my_vcr.use_cassette('test.yml'):
    # your http code here

You can also mutate the response using this callback. For example, you could remove all query parameters from any requests to the '/login' path.

def scrub_login_request(request):
    if request.path == '/login':
        request.uri, _ =  urllib.splitquery(response.uri)
    return request

my_vcr = vcr.VCR(
    before_record=scrub_login_request,
)
with my_vcr.use_cassette('test.yml'):
    # your http code here

Custom Response Filtering

VCR.py also suports response filtering with the before_record_response keyword argument. It's usage is similar to that of before_record:

def scrub_string(string, replacement=''):
    def before_record_reponse(response):
        return response['body']['string'] = response['body']['string'].replace(string, replacement)
    return scrub_string

my_vcr = vcr.VCR(
    before_record=scrub_string(settings.USERNAME, 'username'),
)
with my_vcr.use_cassette('test.yml'):
     # your http code here    

Ignore requests

If you would like to completely ignore certain requests, you can do it in a few ways:

• Set the ignore_localhost option equal to True. This will not record any requests sent to (or responses from) localhost, 127.0.0.1, or 0.0.0.0.
• Set the ignore_hosts configuration option to a list of hosts to ignore
• Add a before_record callback that returns None for requests you want to ignore

Requests that are ignored by VCR will not be saved in a cassette, nor played back from a cassette. VCR will completely ignore those requests as if it didn't notice them at all, and they will continue to hit the server as if VCR were not there.

Custom Patches

If you use a custom HTTPConnection class, or otherwise make http requests in a way that requires additional patching, you can use the custom_patches keyword argument of the VCR and Cassette objects to patch those objects whenever a cassette's context is entered. To patch a custom version of HTTPConnection you can do something like this:

import where_the_custom_https_connection_lives
from vcr.stubs import VCRHTTPSConnection
my_vcr = config.VCR(custom_patches=((where_the_custom_https_connection_lives, 'CustomHTTPSConnection', VCRHTTPSConnection),))

@my_vcr.use_cassette(...)

Automatic Cassette Naming

VCR.py now allows the omission of the path argument to the use_cassette function. Both of the following are now legal/should work

@my_vcr.use_cassette
def my_test_function():
    ...

@my_vcr.use_cassette()
def my_test_function():
    ...

In both cases, VCR.py will use a path that is generated from the provided test function's name. If no cassette_library_dir has been set, the cassette will be in a file with the name of the test function in directory of the file in which the test function is declared. If a cassette_library_dir is set, has been set, the cassette will appear in that directory in a file with the name of the decorated function.

It is possible to control the path produced by the automatic naming machinery by customizing the path_transformer and func_path_generator vcr variables. To add an extension to all cassette names, use VCR.ensure_suffix as follows:

my_vcr = VCR(path_transformer=VCR.ensure_suffix('.yaml'))

@my_vcr.use_cassette
def my_test_function():

Installation

VCR.py is a package on PyPI, so you can pip install vcrpy (first you may need to brew install libyaml [Homebrew])

Ruby VCR compatibility

VCR.py does not aim to match the format of the Ruby VCR YAML files. Cassettes generated by Ruby's VCR are not compatible with VCR.py.

Running VCR's test suite

The tests are all run automatically on Travis CI, but you can also run them yourself using py.test and Tox. Tox will automatically run them in all environments VCR.py supports. The test suite is pretty big and slow, but you can tell tox to only run specific tests like this:

tox -e py27requests -- -v -k "'test_status_code or test_gzip'"

This will run only tests that look like test_status_code or test_gzip in the test suite, and only in the python 2.7 environment that has requests installed.

Also, in order for the boto tests to run, you will need an AWS key. Refer to the boto documentation for how to set this up. I have marked the boto tests as optional in Travis so you don't have to worry about them failing if you submit a pull request.

Logging

VCR.py has a few log messages you can turn on to help you figure out if HTTP requests are hitting a real server or not. You can turn them on like this:

import vcr
import requests
import logging

logging.basicConfig() # you need to initialize logging, otherwise you will not see anything from vcrpy
vcr_log = logging.getLogger("vcr")
vcr_log.setLevel(logging.INFO)

with vcr.use_cassette('headers.yml'):
        requests.get('http://httpbin.org/headers')

The first time you run this, you will see:

INFO:vcr.stubs:<Request (GET) http://httpbin.org/headers> not in cassette, sending to real server

The second time, you will see:

INFO:vcr.stubs:Playing response for <Request (GET) http://httpbin.org/headers> from cassette

If you set the loglevel to DEBUG, you will also get information about which matchers didn't match. This can help you with debugging custom matchers.

Upgrade

New Cassette Format

The cassette format has changed in VCR.py 1.x, the VCR.py 0.x cassettes cannot be used with VCR.py 1.x. The easiest way to upgrade is to simply delete your cassettes and re-record all of them. VCR.py also provides a migration script that attempts to upgrade your 0.x cassettes to the new 1.x format. To use it, run the following command:

python -m vcr.migration PATH

The PATH can be either a path to the directory with cassettes or the path to a single cassette.

Note: Back up your cassettes files before migration. The migration should only modify cassettes using the old 0.x format.

New serializer / deserializer API

If you made a custom serializer, you will need to update it to match the new API in version 1.0.x

• Serializers now take dicts and return strings.
• Deserializers take strings and return dicts (instead of requests, responses pair)

Changelog

• 1.4.2 Fix a bug caused by requests 2.7 and chunked transfer encoding
• 1.4.1 Include README, tests, LICENSE in package. Thanks @ralphbean.
• 1.4.0 Filter post data parameters (thanks @eadmundo), support for posting files through requests, inject_cassette kwarg to access cassette from use_cassette decorated function, with_current_defaults actually works (thanks @samstav).
• 1.3.0 Fix/add support for urllib3 (thanks @aisch), fix default port for https (thanks @abhinav).
• 1.2.0 Add custom_patches argument to VCR/Cassette objects to allow users to stub custom classes when cassettes become active.
• 1.1.4 Add force reset around calls to actual connection from stubs, to ensure compatibility with the version of httplib/urlib2 in python 2.7.9.
• 1.1.3 Fix python3 headers field (thanks @rtaboada), fix boto test (thanks @telaviv), fix new_episodes record mode (thanks @jashugan), fix Windows connectionpool stub bug (thanks @gazpachoking), add support for requests 2.5
• 1.1.2 Add urllib==1.7.1 support. Make json serialize error handling correct Improve logging of match failures.
• 1.1.1 Use function signature preserving wrapt.decorator to write the decorator version of use_cassette in order to ensure compatibility with py.test fixtures and python 2. Move all request filtering into the before_record_callable.
• 1.1.0 Add before_record_response. Fix several bugs related to the context management of cassettes.
• 1.0.3: Fix an issue with requests 2.4 and make sure case sensitivity is consistent across python versions
• 1.0.2: Fix an issue with requests 2.3
• 1.0.1: Fix a bug with the new ignore requests feature and the once record mode
• 1.0.0: BACKWARDS INCOMPATIBLE: Please see the 'upgrade' section in the README. Take a look at the matcher section as well, you might want to update your match_on settings. Add support for filtering sensitive data from requests, matching query strings after the order changes and improving the built-in matchers, (thanks to @mshytikov), support for ignoring requests to certain hosts, bump supported Python3 version to 3.4, fix some bugs with Boto support (thanks @marusich), fix error with URL field capitalization in README (thanks @simon-weber), added some log messages to help with debugging, added all_played property on cassette (thanks @mshytikov)
• 0.7.0: VCR.py now supports Python 3! (thanks @asundg) Also I refactored the stub connections quite a bit to add support for the putrequest and putheader calls. This version also adds support for httplib2 (thanks @nilp0inter). I have added a couple tests for boto since it is an http client in its own right. Finally, this version includes a fix for a bug where requests wasn't being patched properly (thanks @msabramo).
• 0.6.0: Store response headers as a list since a HTTP response can have the same header twice (happens with set-cookie sometimes). This has the added benefit of preserving the order of headers. Thanks @smallcode for the bug report leading to this change. I have made an effort to ensure backwards compatibility with the old cassettes' header storage mechanism, but if you want to upgrade to the new header storage, you should delete your cassettes and re-record them. Also this release adds better error messages (thanks @msabramo) and adds support for using VCR as a decorator (thanks @smallcode for the motivation)
• 0.5.0: Change the response_of method to responses_of since cassettes can now contain more than one response for a request. Since this changes the API, I'm bumping the version. Also includes 2 bugfixes: a better error message when attempting to overwrite a cassette file, and a fix for a bug with requests sessions (thanks @msabramo)
• 0.4.0: Change default request recording behavior for multiple requests. If you make the same request multiple times to the same URL, the response might be different each time (maybe the response has a timestamp in it or something), so this will make the same request multiple times and save them all. Then, when you are replaying the cassette, the responses will be played back in the same order in which they were received. If you were making multiple requests to the same URL in a cassette before version 0.4.0, you might need to regenerate your cassette files. Also, removes support for the cassette.play_count counter API, since individual requests aren't unique anymore. A cassette might contain the same request several times. Also removes secure overwrite feature since that was breaking overwriting files in Windows, and fixes a bug preventing request's automatic body decompression from working.
• 0.3.5: Fix compatibility with requests 2.x
• 0.3.4: Bugfix: close file before renaming it. This fixes an issue on Windows. Thanks @smallcode for the fix.
• 0.3.3: Bugfix for error message when an unreigstered custom matcher was used
• 0.3.2: Fix issue with new config syntax and the match_on parameter. Thanks, @chromy!
• 0.3.1: Fix issue causing full paths to be sent on the HTTP request line.
• 0.3.0: Backwards incompatible release - Added support for record modes, and changed the default recording behavior to the "once" record mode. Please see the documentation on record modes for more. Added support for custom request matching, and changed the default request matching behavior to match only on the URL and method. Also, improved the httplib mocking to add support for the HTTPConnection.send() method. This means that requests won't actually be sent until the response is read, since I need to record the entire request in order to match up the appropriate response. I don't think this should cause any issues unless you are sending requests without ever loading the response (which none of the standard httplib wrappers do, as far as I know. Thanks to @fatuhoku for some of the ideas and the motivation behind this release.
• 0.2.1: Fixed missing modules in setup.py
• 0.2.0: Added configuration API, which lets you configure some settings on VCR (see the README). Also, VCR no longer saves cassettes if they haven't changed at all and supports JSON as well as YAML (thanks @sirpengi). Added amazing new skeumorphic logo, thanks @hairarrow.
• 0.1.0: backwards incompatible release - delete your old cassette files: This release adds the ability to access the cassette to make assertions on it, as well as a major code refactor thanks to @dlecocq. It also fixes a couple longstanding bugs with redirects and HTTPS. [#3 and #4]
• 0.0.4: If you have libyaml installed, vcrpy will use the c bindings instead. Speed up your tests! Thanks @dlecocq
• 0.0.3: Add support for requests 1.2.3. Support for older versions of requests dropped (thanks @vitormazzi and @bryanhelmig)
• 0.0.2: Add support for requests / urllib3
• 0.0.1: Initial Release

License

This library uses the MIT license. See LICENSE.txt for more details