Files
@ 877bcf22bf71
Branch filter:
Location: kallithea/docs/overview.rst
877bcf22bf71
6.2 KiB
text/prs.fallenstein.rst
pytest migration: introduce TestControllerPytest
In order to allow tests to benefit from pytest specific functionality, like
fixtures, they can no longer derive from unittest.TestCase. What's more,
while they can derive from any user-defined class, none of the classes
involved (the test class itself nor any of the base classes) can have an
__init__ method.
Converting all tests from unittest-style to pytest-style in one commit is
not realistic. Hence, a more gradual approach is needed.
Most existing test classes derive from TestController, which in turn derives
from BaseTestCase, which derives from unittest.TestCase. Some test classes
derive directly from BaseTestCase.
Supporting both unittest-style and pytest-style from TestController directly
is not possible: pytest-style _cannot_ and unittest-style _must_ derive from
unittest.TestCase. Thus, in any case, an extra level in the class hierarchy
is needed (TestController deriving from Foo and from unittest.TestCase;
pytest-style test classes would then directly derive from Foo).
The requirement that pytest-style test classes cannot have an __init__
method anywhere in the class hierarchy imposes another restriction that
makes it difficult to support both unittest-style and pytest-style test
classes with one class. Any init code needs to be placed in another method
than __init__ and be called explicitly when the test class is initialized.
For unittest-style test classes this would naturally be done with a
setupClass method, but several test classes already use that. Thus, there
would need to be explicit 'super' calls from the test classes. This is
technically possible but not very nice.
A more transparent approach (from the existing test classes point of view),
implemented by this patch, works as follows:
- the implementation of the existing TestController class is now put under
a new class BaseTestController. To accomodate pytest, the __init__ method
is renamed init.
- contrary to the original TestController, BaseTestController does not
derive from BaseTestCase (and neither from unittest.TestCase). Instead,
the 'new' TestController derives both from BaseTestCase, which is
untouched, and from BaseTestController.
- TestController has an __init__ method that calls the base classes'
__init__ methods and the renamed 'init' method of BaseTestController.
- a new class TestControllerPytest is introduced that derives from
BaseTestController but not from BaseTestCase. It uses a pytest fixture to
automatically call the setup functionality previously provided by
BaseTestCase and also calls 'init' on BaseTestController. This means a
little code duplication but is hard to avoid.
The app setup fixture is scoped on the test method, which means that the app
is recreated for every test (unlike for the unittest-style tests where the
app is created per test class). This has the advantage of detecting current
inter-test dependencies and thus improve the health of our test suite. This
in turn is one step closer to allowing parallel test execution.
The unittest-style assert methods (assertEqual, assertIn, ...) do not exist
for pytest-style tests. To avoid having to change all existing test cases
upfront, provide transitional implementations of these methods. The
conversion of the unittest asserts to the pytest/python asserts can happen
gradually over time.
In order to allow tests to benefit from pytest specific functionality, like
fixtures, they can no longer derive from unittest.TestCase. What's more,
while they can derive from any user-defined class, none of the classes
involved (the test class itself nor any of the base classes) can have an
__init__ method.
Converting all tests from unittest-style to pytest-style in one commit is
not realistic. Hence, a more gradual approach is needed.
Most existing test classes derive from TestController, which in turn derives
from BaseTestCase, which derives from unittest.TestCase. Some test classes
derive directly from BaseTestCase.
Supporting both unittest-style and pytest-style from TestController directly
is not possible: pytest-style _cannot_ and unittest-style _must_ derive from
unittest.TestCase. Thus, in any case, an extra level in the class hierarchy
is needed (TestController deriving from Foo and from unittest.TestCase;
pytest-style test classes would then directly derive from Foo).
The requirement that pytest-style test classes cannot have an __init__
method anywhere in the class hierarchy imposes another restriction that
makes it difficult to support both unittest-style and pytest-style test
classes with one class. Any init code needs to be placed in another method
than __init__ and be called explicitly when the test class is initialized.
For unittest-style test classes this would naturally be done with a
setupClass method, but several test classes already use that. Thus, there
would need to be explicit 'super' calls from the test classes. This is
technically possible but not very nice.
A more transparent approach (from the existing test classes point of view),
implemented by this patch, works as follows:
- the implementation of the existing TestController class is now put under
a new class BaseTestController. To accomodate pytest, the __init__ method
is renamed init.
- contrary to the original TestController, BaseTestController does not
derive from BaseTestCase (and neither from unittest.TestCase). Instead,
the 'new' TestController derives both from BaseTestCase, which is
untouched, and from BaseTestController.
- TestController has an __init__ method that calls the base classes'
__init__ methods and the renamed 'init' method of BaseTestController.
- a new class TestControllerPytest is introduced that derives from
BaseTestController but not from BaseTestCase. It uses a pytest fixture to
automatically call the setup functionality previously provided by
BaseTestCase and also calls 'init' on BaseTestController. This means a
little code duplication but is hard to avoid.
The app setup fixture is scoped on the test method, which means that the app
is recreated for every test (unlike for the unittest-style tests where the
app is created per test class). This has the advantage of detecting current
inter-test dependencies and thus improve the health of our test suite. This
in turn is one step closer to allowing parallel test execution.
The unittest-style assert methods (assertEqual, assertIn, ...) do not exist
for pytest-style tests. To avoid having to change all existing test cases
upfront, provide transitional implementations of these methods. The
conversion of the unittest asserts to the pytest/python asserts can happen
gradually over time.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 | .. _overview:
=====================
Installation overview
=====================
Some overview and some details that can help understanding the options when
installing Kallithea.
Python environment
------------------
**Kallithea** is written entirely in Python_ and requires Python version
2.6 or higher. Python 3.x is currently not supported.
Given a Python installation, there are different ways of providing the
environment for running Python applications. Each of them pretty much
corresponds to a ``site-packages`` directory somewhere where packages can be
installed.
Kallithea itself can be run from source or be installed, but even when running
from source, there are some dependencies that must be installed in the Python
environment used for running Kallithea.
- Packages *could* be installed in Python's ``site-packages`` directory ... but
that would require running pip_ as root and it would be hard to uninstall or
upgrade and is probably not a good idea unless using a package manager.
- Packages could also be installed in ``~/.local`` ... but that is probably
only a good idea if using a dedicated user per application or instance.
- Finally, it can be installed in a virtualenv_. That is a very lightweight
"container" where each Kallithea instance can get its own dedicated and
self-contained virtual environment.
We recommend using virtualenv for installing Kallithea.
Installation methods
--------------------
Kallithea must be installed on a server. Kallithea is installed in a Python
environment so it can use packages that are installed there and make itself
available for other packages.
Two different cases will pretty much cover the options for how it can be
installed.
- The Kallithea source repository can be cloned and used -- it is kept stable and
can be used in production. The Kallithea maintainers use the development
branch in production. The advantage of installation from source and regularly
updating it is that you take advantage of the most recent improvements. Using
it directly from a DVCS also means that it is easy to track local customizations.
Running ``setup.py develop`` in the source will use pip to install the
necessary dependencies in the Python environment and create a
``.../site-packages/Kallithea.egg-link`` file there that points at the Kallithea
source.
- Kallithea can also be installed from ready-made packages using a package manager.
The official released versions are available on PyPI_ and can be downloaded and
installed with all dependencies using ``pip install kallithea``.
With this method, Kallithea is installed in the Python environment as any
other package, usually as a ``.../site-packages/Kallithea-X-py2.7.egg/``
directory with Python files and everything else that is needed.
(``pip install kallithea`` from a source tree will do pretty much the same
but build the Kallithea package itself locally instead of downloading it.)
Web server
----------
Kallithea is (primarily) a WSGI_ application that must be run from a web
server that serves WSGI applications over HTTP.
Kallithea itself is not serving HTTP (or HTTPS); that is the web server's
responsibility. Kallithea does however need to know its own user facing URL
(protocol, address, port and path) for each HTTP request. Kallithea will
usually use its own HTML/cookie based authentication but can also be configured
to use web server authentication.
There are several web server options:
- Kallithea uses the Paste_ tool as command line interface. Paste provides
``paster serve`` as a convenient way to launch a Python WSGI / web server
from the command line. That is perfect for development and evaluation.
Actual use in production might have different requirements and need extra
work to make it manageable as a scalable system service.
Paste comes with its own built-in web server but Kallithea defaults to use
Waitress_. Gunicorn_ is also an option. These web servers have different
limited feature sets.
The web server used by ``paster`` is configured in the ``.ini`` file passed
to it. The entry point for the WSGI application is configured
in ``setup.py`` as ``kallithea.config.middleware:make_app``.
- `Apache httpd`_ can serve WSGI applications directly using mod_wsgi_ and a
simple Python file with the necessary configuration. This is a good option if
Apache is an option.
- uWSGI_ is also a full web server with built-in WSGI module.
- IIS_ can also server WSGI applications directly using isapi-wsgi_.
- A `reverse HTTP proxy <https://en.wikipedia.org/wiki/Reverse_proxy>`_
can be put in front of another web server which has WSGI support.
Such a layered setup can be complex but might in some cases be the right
option, for example to standardize on one internet-facing web server, to add
encryption or special authentication or for other security reasons, to
provide caching of static files, or to provide load balancing or fail-over.
Nginx_, Varnish_ and HAProxy_ are often used for this purpose, often in front
of a ``paster`` server that somehow is wrapped as a service.
The best option depends on what you are familiar with and the requirements for
performance and stability. Also, keep in mind that Kallithea mainly is serving
dynamically generated pages from a relatively slow Python process. Kallithea is
also often used inside organizations with a limited amount of users and thus no
continuous hammering from the internet.
.. _Python: http://www.python.org/
.. _Gunicorn: http://gunicorn.org/
.. _Waitress: http://waitress.readthedocs.org/en/latest/
.. _virtualenv: http://pypi.python.org/pypi/virtualenv
.. _Paste: http://pythonpaste.org/
.. _PyPI: https://pypi.python.org/pypi
.. _Apache httpd: http://httpd.apache.org/
.. _mod_wsgi: https://code.google.com/p/modwsgi/
.. _isapi-wsgi: https://github.com/hexdump42/isapi-wsgi
.. _uWSGI: https://uwsgi-docs.readthedocs.org/en/latest/
.. _nginx: http://nginx.org/en/
.. _iis: http://en.wikipedia.org/wiki/Internet_Information_Services
.. _pip: http://en.wikipedia.org/wiki/Pip_%28package_manager%29
.. _WSGI: http://en.wikipedia.org/wiki/Web_Server_Gateway_Interface
.. _pylons: http://www.pylonsproject.org/
.. _HAProxy: http://www.haproxy.org/
.. _Varnish: https://www.varnish-cache.org/
|