Running ``pip install -e .`` 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-py3.8.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.)

.. note::

   Kallithea includes front-end code that needs to be processed to prepare

   static files that can be served at run time and used on the client side. The

   tool npm_ is used to download external dependencies and orchestrate the

   processing. The ``npm`` binary must thus be available at install time but is

   not used at run time.

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 Gearbox_ tool as command line interface. Gearbox provides

  ``gearbox 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.

  Gearbox comes with its own built-in web server for development but Kallithea

  defaults to using Waitress_. Gunicorn_ and Gevent_ are also options. These

  web servers have different limited feature sets.

  The web server used by ``gearbox serve`` is configured in the ``.ini`` file.

  Create it with ``config-create`` using for example ``http_server=waitress``

  to get a configuration starting point for your choice of web server.

  (Gearbox will do like ``paste`` and use the WSGI application entry point

  ``kallithea.config.application:make_app`` as specified in ``setup.py``.)

- `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. Use

  ``config-create`` with ``http_server=uwsgi`` to get a ``.ini`` file with

  uWSGI configuration.

- 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 ``gearbox serve`` 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.

.. note::

   Kallithea, the libraries it uses, and Python itself do in several places use

   simple caching in memory. Caches and memory are not always released in a way

   that is suitable for long-running processes. They might appear to be leaking

   memory. The worker processes should thus regularly be restarted - for

   example after 1000 requests and/or one hour. This can usually be done by the

   web server or the tool used for running it as a system service.

.. _Python: http://www.python.org/

.. _Gunicorn: http://gunicorn.org/

.. _Gevent: http://www.gevent.org/

.. _Waitress: https://docs.pylonsproject.org/projects/waitress/

.. _Gearbox: https://turbogears.readthedocs.io/en/latest/turbogears/gearbox.html

.. _PyPI: https://pypi.python.org/pypi

.. _Apache httpd: http://httpd.apache.org/

.. _isapi-wsgi: https://github.com/hexdump42/isapi-wsgi

.. _uWSGI: https://uwsgi-docs.readthedocs.io/

.. _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

.. _HAProxy: http://www.haproxy.org/

.. _Varnish: https://www.varnish-cache.org/

.. _npm: https://www.npmjs.com/