Python version of Pact mocking, generation and verification.

Enables consumer driven contract testing, providing unit test mocking of provider services and DSL for the consumer project, and interaction playback and verification for the service provider project. Currently supports versions 1.1, 2 and 3 of the Pact specification.

For more information about what Pact is, and how it can help you test your code more efficiently, check out the Pact documentation.

Contains code originally from the pact-python project.

pactman is maintained by the ReeceTech team as part of their toolkit to keep their large (and growing) microservices architecture under control.

• How to use pactman
  • Installation
  • Writing a Pact
  • Writing the Consumer Test
  • Expecting Variable Content
  • Verifying Pacts Against a Service
• Development
  • Release History

The key difference is all functionality is implemented in Python, rather than shelling out or forking to the ruby implementation. This allows for a much nicer mocking user experience (it mocks urllib3 directly), is faster, less messy configuration (multiple providers means multiple ruby processes spawned on different ports).

Where pact-python required management of a background Ruby server, and manually starting and stopping it, pactman allows a much nicer usage like:

import requests
from pactman import Consumer, Provider

pact = Consumer('Consumer').has_pact_with(Provider('Provider'))

def test_interaction():
    pact.given("some data exists").upon_receiving("a request") \
        .with_request("get", "/", query={"foo": ["bar"]}).will_respond_with(200)
    with pact:
        requests.get(pact.uri, params={"foo": ["bar"]})

It also supports a broader set of the pact specification (versions 1.1 through to 3).

The pact verifier has been engineered from the start to talk to a pact broker (both to discover pacts and to return verification results).

There’s a few other quality of life improvements, but those are the big ones.

pactman requires Python 3.6 to run.

pip install pactman

Creating a complete contract is a two step process:

1. Create a unit test on the consumer side that declares the expectations it has of the provider
2. Create a provider state that allows the contract to pass when replayed against the provider

If we have a method that communicates with one of our external services, which we'll call Provider, and our product, Consumer is hitting an endpoint on Provider at /users/<user> to get information about a particular user.

If the Consumer's code to fetch a user looked like this:

import requests

def get_user(user_name):
    response = requests.get(f'http://service.example/users/{user_name}')
    return response.json()

Then Consumer's contract test is a regular unit test, but using pactman for mocking, and might look something like this:

import unittest
from pactman import Consumer, Provider

pact = Consumer('Consumer').has_pact_with(Provider('Provider'))

class GetUserInfoContract(unittest.TestCase):
  def test_get_user(self):
    expected = {
      'username': 'UserA',
      'id': 123,
      'groups': ['Editors']
    }

    pact.given(
        'UserA exists and is not an administrator'
    ).upon_receiving(
        'a request for UserA'
    ).with_request(
        'GET', '/users/UserA'
    ) .will_respond_with(200, body=expected)

    with pact:
      result = get_user('UserA')

    self.assertEqual(result, expected)

This does a few important things:

• Defines the Consumer and Provider objects that describe our product and our service under test
• Uses given to define the setup criteria for the Provider UserA exists and is not an administrator
• Defines what the request that is expected to be made by the consumer will contain
• Defines how the server is expected to respond

Using the Pact object as a context manager, we call our method under test which will then communicate with the Pact mock. The mock will respond with the items we defined, allowing us to assert that the method processed the response and returned the expected value.

If you want more control over when the mock is configured and the interactions verified, use the setup and verify methods, respectively:

Consumer('Consumer').has_pact_with(Provider('Provider')).given(
    'UserA exists and is not an administrator'
).upon_receiving(
    'a request for UserA'
).with_request(
    'GET', '/users/UserA'
) .will_respond_with(200, body=expected)

pact.setup()
try:
    # Some additional steps before running the code under test
    result = get_user('UserA')
    # Some additional steps before verifying all interactions have occurred
finally:
    pact.verify()

You may have noticed that the pact relationship is defined at the module level in our examples:

pact = Consumer('Consumer').has_pact_with(Provider('Provider'))

This is because it must only be done once per test suite. By default the pact file is cleared out when that relationship is defined, so if you define it more than once per test suite you'll end up only storing the last pact declared per relationship. For more on this subject, see writing multiple pacts.

When defining the expected HTTP request that your code is expected to make you can specify the method, path, body, headers, and query:

pact.with_request(
    method='GET',
    path='/api/v1/my-resources/',
    query={'search': 'example'}
)

query is used to specify URL query parameters, so the above example expects a request made to /api/v1/my-resources/?search=example.

pact.with_request(
    method='POST',
    path='/api/v1/my-resources/123',
    body={'user_ids': [1, 2, 3]},
    headers={'Content-Type': 'application/json'},
)

You can define exact values for your expected request like the examples above, or you can use the matchers defined later to assist in handling values that are variable.

The has_pact_with(provider...) call has quite a few options documented in its API, but a couple are worth mentioning in particular:

version declares the pact specification version that the provider supports. This defaults to "2.0.0", but "3.0.0" is also acceptable if your provider supports Pact specification version 3:

from pactman import Consumer, Provider
pact = Consumer('Consumer').has_pact_with(Provider('Provider'), version='3.0.0')

file_write_mode defaults to "overwrite" and should be that or "merge". Overwrite ensures that any existing pact file will be removed when has_pact_with() is invoked. Merge will retain the pact file and add new pacts to that file. See writing multiple pacts. If you absolutely do not want pact files to be written, use "never".

use_mocking_server defaults to False and controls the mocking method used by pactman. The default is to patch urllib3, which is the library underpinning requests and is also used by some other projects. If you are using a different library to make your HTTP requests which does not use urllib3 underneath then you will need to set the use_mocking_server argument to True. This causes pactman to run an actual HTTP server to mock the requests (the server is listening on pact.uri - use that to redirect your HTTP requests to the mock server.) You may also set the PACT_USE_MOCKING_SERVER environment variable to "yes" to force your entire suite to use the server approach. You should declare the pact particpants (consumer and provider) outside of your tests and will need to start and stop the mocking service outside of your tests too. The code below shows what using the server might look like:

import atexit
from pactman import Consumer, Provider
pact = Consumer('Consumer').has_pact_with(Provider('Provider'), use_mocking_server=True)
pact.start_mocking()
atexit.register(pact.stop_mocking)

You'd then use pact to declare pacts between those participants.

During a test run you're likely to need to write multiple pact interactions for a consumer/provider relationship. pactman will manage the pact file as follows:

• When has_pact_with() is invoked it will by default remove any existing pact JSON file for the stated consumer & provider.
• You may invoke Consumer('Consumer').has_pact_with(Provider('Provider')) once at the start of your tests. This could be done as a pytest module or session fixture, or through some other mechanism and store it in a variable. By convention this is called pact in all of our examples.
• If that is not suitable, you may manually indicate to has_pact_with() that it should either retain (file_write_mode="merge") or remove (file_write_mode="overwrite") the existing pact file.

You use given() to indicate to the provider that they should have some state in order to be able to satisfy the interaction. You should agree upon the state and its specification in discussion with the provider.

If you are defining a version 3 pact you may define provider states more richly, for example:

(pact
    .given("this is a simple state as in v2")
    .and_given("also the user must exist", username="alex")
)

Now you may specify additional parameters to accompany your provider state text. These are passed as keyword arguments, and they're optional. You may also provider additional provider states using the and_given() call, which may be invoked many times if necessary. It and given() have the same calling convention: a provider state name and any optional parameters.

The default validity testing of equal values works great if that user information is always static, but what happens if the user has a last updated field that is set to the current time every time the object is modified? To handle variable data and make your tests more robust, there are several helpful matchers:

Available in version 3.0.0+ pacts

Asserts that the value should contain the given substring, for example::

from pactman import Includes, Like
Like({
    'id': 123, # match integer, value varies
    'content': Includes('spam', 'Sample spamming content')  # content must contain the string "spam"
})

The matcher and sample_data are used differently by consumer and provider depending upon whether they're used in the with_request() or will_respond_with() sections of the pact. Using the above example:

When you run the tests for the consumer, the mock will verify that the data the consumer uses in its request contains the matcher string, raising an AssertionError if invalid. When the contract is verified by the provider, the sample_data will be used in the request to the real provider service, in this case 'Sample spamming content'.

When you run the tests for the consumer, the mock will return the data you provided as sample_data, in this case 'Sample spamming content'. When the contract is verified on the provider, the data returned from the real provider service will be verified to ensure it contains the matcher string.

Asserts the value should match the given regular expression. You could use this to expect a timestamp with a particular format in the request or response where you know you need a particular format, but are unconcerned about the exact date:

from pactman import Term

(pact
 .given('UserA exists and is not an administrator')
 .upon_receiving('a request for UserA')
 .with_request(
   'post',
   '/users/UserA/info',
   body={'commencement_date': Term('\d+-\d+-\d', '1972-01-01')})
 .will_respond_with(200, body={
    'username': 'UserA',
    'last_modified': Term('\d+-\d+-\d+T\d+:\d+:\d+', '2016-12-15T20:16:01')
 }))

The matcher and sample_data are used differently by consumer and provider depending upon whether they're used in the with_request() or will_respond_with() sections of the pact. Using the above example:

When you run the tests for the consumer, the mock will verify that the commencement_date the consumer uses in its request matches the matcher, raising an AssertionError if invalid. When the contract is verified by the provider, the sample_data will be used in the request to the real provider service, in this case 1972-01-01.

When you run the tests for the consumer, the mock will return the last_modified you provided as sample_data, in this case 2016-12-15T20:16:01. When the contract is verified on the provider, the regex will be used to search the response from the real provider service and the test will be considered successful if the regex finds a match in the response.

Asserts the element's type matches the sample_data. For example:

from pactman import Like
Like(123)  # Matches if the value is an integer
Like('hello world')  # Matches if the value is a string
Like(3.14)  # Matches if the value is a float

When you run the tests for the consumer, the mock will verify that values are of the correct type, raising an AssertionError if invalid. When the contract is verified by the provider, the sample_data will be used in the request to the real provider service.

When you run the tests for the consumer, the mock will return the sample_data. When the contract is verified on the provider, the values generated by the provider service will be checked to match the type of sample_data.

When a dictionary is used as an argument for Like, all the child objects (and their child objects etc.) will be matched according to their types, unless you use a more specific matcher like a Term.

from pactman import Like, Term
Like({
    'username': Term('[a-zA-Z]+', 'username'),
    'id': 123, # integer
    'confirmed': False, # boolean
    'address': { # dictionary
        'street': '200 Bourke St' # string
    }
})

Asserts the value is an array type that consists of elements like sample_data. It can be used to assert simple arrays:

from pactman import EachLike
EachLike(1)  # All items are integers
EachLike('hello')  # All items are strings

Or other matchers can be nested inside to assert more complex objects:

from pactman import EachLike, Term
EachLike({
    'username': Term('[a-zA-Z]+', 'username'),
    'id': 123,
    'groups': EachLike('administrators')
})

Note, you do not need to specify everything that will be returned from the Provider in a JSON response, any extra data that is received will be ignored and the tests will still pass.

For more information see Matching

Available in version 3.0.0+ pacts

If you have a sub-term of a Like which needs to match an exact value like the default validity test then you can use Equals, for example::

from pactman import Equals, Like
Like({
    'id': 123, # match integer, value varies
    'username': Equals('alex')  # username must always be "alex"
})

The body payload is assumed to be JSON data. In the absence of a Content-Type header we assume Content-Type: application/json; charset=UTF-8 (JSON text is Unicode and the default encoding is UTF-8).

During verification non-JSON payloads are compared for equality.

During mocking, the HTTP response will be handled as:

1. If there's no Content-Type header, assume JSON: serialise with json.dumps(), encode to UTF-8 and add the header Content-Type: application/json; charset=UTF-8.
2. If there's a Content-Type header and it says application/json then serialise with json.dumps() and use the charset in the header, defaulting to UTF-8.
3. Otherwise pass through the Content-Type header and body as-is. Binary data is not supported.

You have two options for verifying pacts against a service you created:

1. Use the pactman-verifier command-line program which replays the pact assertions against a running instance of your service, or
2. Use the pytest support built into pactman to replay the pacts as test cases, allowing use of other testing mechanisms such as mocking and transaction control.

Run pactman-verifier -h to see the options available. To run all pacts registered to a provider in a Pact Broker:

pactman-verifier -b http://pact-broker.example/ <provider name> <provider url> <provider setup url>

You can pass in a local pact file with -l, this will verify the service against the local file instead of the broker:

pactman-verifier -l /tmp/localpact.json <provider name> <provider url> <provider setup url>

You can use --custom-provider-header to pass in headers to be passed to provider state setup and verify calls. it can be used multiple times

pactman-verifier -b <broker url> --custom-provider-header "someheader:value" --custom-provider-header
"this:that" <provider name> <provider url> <provider state url>

An additional header may also be supplied in the PROVIDER_EXTRA_HEADER environment variable, though the command line argument(s) would override this.

In many cases, your contracts will need very specific data to exist on the provider to pass successfully. If you are fetching a user profile, that user needs to exist, if querying a list of records, one or more records needs to exist. To support decoupling the testing of the consumer and provider, Pact offers the idea of provider states to communicate from the consumer what data should exist on the provider.

When setting up the testing of a provider you will also need to setup the management of these provider states. The Pact verifier does this by making additional HTTP requests to the <provider setup url> you provide. This URL could be on the provider application or a separate one. Some strategies for managing state include:

• Having endpoints in your application that are not active in production that create and delete your datastore state
• A separate application that has access to the same datastore to create and delete, like a separate App Engine module or Docker container pointing to the same datastore
• A standalone application that can start and stop the other server with different datastore states

For more information about provider states, refer to the Pact documentation on Provider States.

To verify pacts for a provider you would write a new pytest test module in the provider's test suite. If you don't want it to be exercised in your usual unit test run you can call it verify_pacts.py.

Your test code needs to use the pact_verifier fixture provided by pactman, invoking its verify() method with the URL to the running instance of your service (pytest-django provides a handy live_server fixture which works well here) and a callback to set up provider states (described below).

You'll need to include some extra command-line arguments to pytest (also described below) to indicate where the pacts should come from, and whether verification results should be posted to a pact broker.

An example for a Django project might contain:

from django.contrib.auth.models import User
from pactman.verifier.verify import ProviderStateMissing

def provider_state(name, **params):
    if name == 'the user "pat" exists':
        User.objects.create(username='pat', fullname=params['fullname'])
    else:
        raise ProviderStateMissing(name)

def test_pacts(live_server, pact_verifier):
    pact_verifier.verify(live_server.url, provider_state)

The pact_verifier.verify call may also take a third argument to supply additional HTTP headers to send to the server during verification - specify them as a dictionary.

The test function may do any level of mocking and data setup using standard pytest fixtures - so mocking downstream APIs or other interactions within the provider may be done with standard monkeypatching.

The provider_state function passed to pact_verifier.verify will be passed the providerState and providerStates for all pacts being verified.

• For pacts with providerState the name argument will be the providerState value, and params will be empty.
• For pacts with providerStates the function will be invoked once per entry in providerStates array with the name argument taken from the array entry name parameter, and params from the params parameter.

Once you have written the pytest code, you need to invoke pytest with additional arguments:

--pact-broker-url=<URL> provides the base URL of the Pact broker to retrieve pacts from for the provider. You must also provide --pact-provider-name=<ProviderName> to identify which provider to retrieve pacts for from the broker.

The broker URL and provider name may alternatively be provided through the environment variables PACT_BROKER_URL and PACT_PROVIDER_NAME.

You may provide --pact-verify-consumer=<ConsumerName> to limit the pacts verified to just that consumer. As with the command-line verifier, you may provide basic auth details in the broker URL, or through the PACT_BROKER_AUTH environment variable. If your broker requires a bearer token you may provide it with --pact-broker-token=<TOKEN> or the PACT_BROKER_TOKEN environment variable.

--pact-files=<file pattern> verifies some on-disk pact JSON files identified by the wildcard pattern (unix glob pattern matching, use ** to match multiple directories).

If you pulled the pacts from a broker and wish to publish verification results, use --pact-publish-results to turn on publishing the results. This option also requires you to specify --pact-provider-version=<version>.

So, for example:

# verify some local pacts in /tmp/pacts
$ pytest --pact-files=/tmp/pacts/*.json tests/verify_pacts.py

# verify some pacts in a broker for the provider MyService
$ pytest --pact-broker-url=http://pact-broker.example/ --pact-provider-name=MyService tests/verify_pacts.py

If you need to see the traceback that caused a pact failure you can use the verbosity flag to pytest (pytest -v).

See the "pact" section in the pytest command-line help (pytest -h) for all command-line options.

You may also specify the broker URL in the environment variable PACT_BROKER_URL.

If HTTP Basic Auth is required for the broker, that may be provided in the URL:

pactman-verifier -b http://user:password@pact-broker.example/ ...
pytest --pact-broker-url=http://user:password@pact-broker.example/ ...

or set in the PACT_BROKER_AUTH environment variable as user:password.

If your broker needs a bearer token then you may provide that on the command line or set it in the environment variable PACT_BROKER_TOKEN.

If your consumer pacts have tags (called "consumer version tags" because they attach to specific versions) then you may specify the tag(s) to fetch pacts for on the command line. Multiple tags may be specified, and all pacts matching any tags specified will be verified. For example, to ensure you're verifying your Provider against the production pact versions from your Consumers, use:

pactman-verifier --consumer-version-tag=production -b http://pact-broker.example/ ...
pytest --pact-verify-consumer-tag=production --pact-broker-url=http://pact-broker.example/ ...

Please read CONTRIBUTING.md

3.0.0 (FUTURE, DEPRECATION WARNINGS)

• remove DEPRECATED --pact-consumer-name command-line option

2.31.0

• Fix bug where a Pact specification version of 4.0 in the broker metadata could not be parsed as a semantic version

2.30.0

• DELETE requests may now have query strings, thanks @MazeDeveloper
• Nicer feedback if no pact source is specified on command line, thanks @artamonovkirill
• Add PACT_PROVIDER_NAME to environment vars, thanks @artamonovkirill

2.29.0

• Added support for ** recursive globbing with --pact-files, thanks @maksimt

2.28.0

• Fixed edge case where fail() was not being invoked in an exact match causing the pytest reporter to not know there'd been a failure
• Address deprecation of semver.parse in semver
• Dropped Python 3.6 testing, added Python 3.8 testing

2.27.0

• Fix typo in pytest plugin preventing --pact-verify-consumer-tag from working
• Added PATCH support

2.26.0

• Allow pytest verification to specify extra_provider_headers

2.25.0

• Add option to allow pytest to succeed even if a pact verification fails

2.24.0

• Better integration of pact failure information in pytest

2.23.0

• Enable setting of authentication credentials when connecting to the pact broker
• Allow filtering of pacts fetched from broker to be filtered by consumer version tag
• Improve the naming and organisation of the pytest command line options

2.22.0

• Better implementation of change in 2.21.0

2.21.0

• Handle warning level messages in command line output handler

2.20.0

• Fix pytest mode to correctly detect array element rule failure as a pytest failure
• Allow restricting pytest verification runs to a single consumer using --pact-consumer-name

2.19.0

• Correct teardown of pact context manager where the pact is used in multiple interactions (with interaction1, interaction2 instead of with pact).

2.18.0

• Correct bug in cleanup that resulted in urllib mocking breaking.

2.17.0

• Handle absence of any provider state (!) in pytest setup.

2.16.0

• Delay shenanigans around checking pacts directory until pacts are actually written to allow module-level pact definition without side effects.

2.15.0

• Fix structure of serialisation for header matching rules.
• Add "never" to the file_write_mode options.
• Handle x-www-form-urlencoded POST request bodies.

2.14.0

• Improve verbose messages to clarify what they're saying.

2.13.0

• Add ability to supply additional headers to provider during verification (thanks @ryallsa)

2.12.1

• Fix pact-python Term compatibility

2.12.0

• Add Equals and Includes matchers for pact v3+
• Make verification fail if missing header specified in interaction
• Significantly improved support for pytest provider verification of pacts
• Turned pact state call failures into warnings rather than errors

2.11.0

• Ensure query param values are lists

2.10.0

• Allow has_pact_with() to accept file_write_mode
• Fix bug introduced in 2.9.0 where generating multiple pacts would result in a single pact being recorded

2.9.0

• Fix with_request when called with a dict query (thanks Cong)
• Make start_mocking() and stop_mocking() optional with non-server mocking
• Add shortcut so python -m pactman.verifier.command_line is just python -m pactman (mostly used in testing before release)
• Handle the None provider state
• Ensure pact spec versions are consistent across all mocks used to generate a pact file

2.8.0

• Close up some edge cases in body content during mocking, and document in README

2.7.0

• Added and_given() as a method of defining additonal provider states for v3+ pacts
• Added more tests for pact generation (serialisation) which fixed a few edge case bugs
• Fix handling of lower-case HTTP methods in verifier (thanks Cong!)

2.6.1

• Fix issue where mocked urlopen didn't handle the correct number of positional arguments

2.6.0

• Fix several issues cause by a failure to detect failure in several test cases (header, path and array element rules may not have been applied)
• Fix rules applying to a single non-first element in an array
• Fix generation of consumer / provider name in <v3 pacts

2.5.0

• Fix some bugs around empty array verification

2.4.0

• Create the pact destination dir if it's missing and its parent exists

2.3.0

• Fix some issues around mocking request queries and the mock's verification of same
• Fix header regex matching in mock verification
• Actually use the version passed in to has_pact_with()
• Fix some pact v3 generation issues (thanks pan Jacek)

2.2.0

• Reinstate lost result output.

2.1.0

• Corrected the definition of request payload when there is no body in the request

2.0.0

• Correctly determine pact verification result when publishing to broker.

1.2.0

• Corrected use of format_path in command line error handling.
• Tweaked README for clarity.

1.1.0

• Renamed the pact-verifier command to pactman-verifier to avoid confusion with other pre-existing packages that provide a command-line incompatible pact-verifier command.
• Support verification of HEAD requests (oops).

1.0.8

• Corrected project URL in project metadata (thanks Jonathan Moss)
• Fix verbose output

1.0.7

• Added some Trove classifiers to aid potential users.

1.0.6

• Corrected mis-named command-line option.

1.0.5

• Corrected some packaging issues

1.0.4

• Initial release of pactman, including ReeceTech's pact-verifier version 3.17 and pact-python version 0.17.0