django/docs/howto/custom-management-commands.txt

====================================
Writing custom django-admin commands
====================================

.. module:: django.core.management

Applications can register their own actions with ``manage.py``. For example,
you might want to add a ``manage.py`` action for a Django app that you're
distributing. In this document, we will be building a custom ``closepoll``
command for the ``polls`` application from the
:doc:`tutorial</intro/tutorial01>`.

To do this, just add a ``management/commands`` directory to the application.
Django will register a ``manage.py`` command for each Python module in that
directory whose name doesn't begin with an underscore. For example::

    polls/
        __init__.py
        models.py
        management/
            __init__.py
            commands/
                __init__.py
                _private.py
                closepoll.py
        tests.py
        views.py

On Python 2, be sure to include ``__init__.py`` files in both the
``management`` and ``management/commands`` directories as done above or your
command will not be detected.

In this example, the ``closepoll`` command will be made available to any project
that includes the ``polls`` application in :setting:`INSTALLED_APPS`.

The ``_private.py`` module will not be available as a management command.

The ``closepoll.py`` module has only one requirement -- it must define a class
``Command`` that extends :class:`BaseCommand` or one of its
:ref:`subclasses<ref-basecommand-subclasses>`.

.. admonition:: Standalone scripts

  Custom management commands are especially useful for running standalone
  scripts or for scripts that are periodically executed from the UNIX crontab
  or from Windows scheduled tasks control panel.

To implement the command, edit ``polls/management/commands/closepoll.py`` to
look like this::

    from django.core.management.base import BaseCommand, CommandError
    from polls.models import Poll

    class Command(BaseCommand):
        help = 'Closes the specified poll for voting'

        def add_arguments(self, parser):
            parser.add_argument('poll_id', nargs='+', type=int)

        def handle(self, *args, **options):
            for poll_id in options['poll_id']:
                try:
                    poll = Poll.objects.get(pk=poll_id)
                except Poll.DoesNotExist:
                    raise CommandError('Poll "%s" does not exist' % poll_id)

                poll.opened = False
                poll.save()

                self.stdout.write('Successfully closed poll "%s"' % poll_id)

Before Django 1.8, management commands were based on the :py:mod:`optparse`
module, and positional arguments were passed in ``*args`` while optional
arguments were passed in ``**options``. Now that management commands use
:py:mod:`argparse` for argument parsing, all arguments are passed in
``**options`` by default, unless you name your positional arguments to ``args``
(compatibility mode). You are encouraged to exclusively use ``**options`` for
new commands.

.. _management-commands-output:

.. note::
    When you are using management commands and wish to provide console
    output, you should write to ``self.stdout`` and ``self.stderr``,
    instead of printing to ``stdout`` and ``stderr`` directly. By
    using these proxies, it becomes much easier to test your custom
    command. Note also that you don't need to end messages with a newline
    character, it will be added automatically, unless you specify the ``ending``
    parameter::

        self.stdout.write("Unterminated line", ending='')

The new custom command can be called using ``python manage.py closepoll
<poll_id>``.

The ``handle()`` method takes one or more ``poll_ids`` and sets ``poll.opened``
to ``False`` for each one. If the user referenced any nonexistent polls, a
:class:`CommandError` is raised. The ``poll.opened`` attribute does not exist
in the :doc:`tutorial</intro/tutorial01>` and was added to
``polls.models.Poll`` for this example.

.. _custom-commands-options:

Accepting optional arguments
============================

The same ``closepoll`` could be easily modified to delete a given poll instead
of closing it by accepting additional command line options. These custom
options can be added in the :meth:`~BaseCommand.add_arguments` method like this::

    class Command(BaseCommand):
        def add_arguments(self, parser):
            # Positional arguments
            parser.add_argument('poll_id', nargs='+', type=int)

            # Named (optional) arguments
            parser.add_argument('--delete',
                action='store_true',
                dest='delete',
                default=False,
                help='Delete poll instead of closing it')

        def handle(self, *args, **options):
            # ...
            if options['delete']:
                poll.delete()
            # ...

.. versionchanged:: 1.8

    Previously, only the standard :py:mod:`optparse` library was supported and
    you would have to extend the command ``option_list`` variable with
    ``optparse.make_option()``.

The option (``delete`` in our example) is available in the options dict
parameter of the handle method. See the :py:mod:`argparse` Python documentation
for more about ``add_argument`` usage.

In addition to being able to add custom command line options, all
:doc:`management commands</ref/django-admin>` can accept some
default options such as :djadminopt:`--verbosity` and :djadminopt:`--traceback`.

.. _management-commands-and-locales:

Management commands and locales
===============================

By default, the :meth:`BaseCommand.execute` method deactivates translations
because some commands shipped with Django perform several tasks (for example,
user-facing content rendering and database population) that require a
project-neutral string language.

.. versionchanged:: 1.8

    In previous versions, Django forced the "en-us" locale instead of
    deactivating translations.

If, for some reason, your custom management command needs to use a fixed locale,
you should manually activate and deactivate it in your
:meth:`~BaseCommand.handle` method using the functions provided by the I18N
support code::

    from django.core.management.base import BaseCommand, CommandError
    from django.utils import translation

    class Command(BaseCommand):
        ...
        can_import_settings = True

        def handle(self, *args, **options):

            # Activate a fixed locale, e.g. Russian
            translation.activate('ru')

            # Or you can activate the LANGUAGE_CODE # chosen in the settings:
            from django.conf import settings
            translation.activate(settings.LANGUAGE_CODE)

            # Your command logic here
            ...

            translation.deactivate()

Another need might be that your command simply should use the locale set in
settings and Django should be kept from deactivating it. You can achieve
it by using the :data:`BaseCommand.leave_locale_alone` option.

When working on the scenarios described above though, take into account that
system management commands typically have to be very careful about running in
non-uniform locales, so you might need to:

* Make sure the :setting:`USE_I18N` setting is always ``True`` when running
  the command (this is a good example of the potential problems stemming
  from a dynamic runtime environment that Django commands avoid offhand by
  deactivating translations).

* Review the code of your command and the code it calls for behavioral
  differences when locales are changed and evaluate its impact on
  predictable behavior of your command.

Testing
=======

Information on how to test custom management commands can be found in the
:ref:`testing docs <topics-testing-management-commands>`.

Command objects
===============

.. class:: BaseCommand

The base class from which all management commands ultimately derive.

Use this class if you want access to all of the mechanisms which
parse the command-line arguments and work out what code to call in
response; if you don't need to change any of that behavior,
consider using one of its :ref:`subclasses<ref-basecommand-subclasses>`.

Subclassing the :class:`BaseCommand` class requires that you implement the
:meth:`~BaseCommand.handle` method.

Attributes
----------

All attributes can be set in your derived class and can be used in
:class:`BaseCommand`’s :ref:`subclasses<ref-basecommand-subclasses>`.

.. attribute:: BaseCommand.args

  A string listing the arguments accepted by the command,
  suitable for use in help messages; e.g., a command which takes
  a list of application names might set this to '<app_label
  app_label ...>'.

  .. deprecated:: 1.8

      This should be done now in the :meth:`~BaseCommand.add_arguments()`
      method, by calling the ``parser.add_argument()`` method. See the
      ``closepoll`` example above.

.. attribute:: BaseCommand.can_import_settings

  A boolean indicating whether the command needs to be able to
  import Django settings; if ``True``, ``execute()`` will verify
  that this is possible before proceeding. Default value is
  ``True``.

.. attribute:: BaseCommand.help

  A short description of the command, which will be printed in the
  help message when the user runs the command
  ``python manage.py help <command>``.

.. attribute:: BaseCommand.missing_args_message

.. versionadded:: 1.8

  If your command defines mandatory positional arguments, you can customize
  the message error returned in the case of missing arguments. The default is
  output by :py:mod:`argparse` ("too few arguments").

.. attribute:: BaseCommand.option_list

  This is the list of ``optparse`` options which will be fed
  into the command's ``OptionParser`` for parsing arguments.

  .. deprecated:: 1.8

      You should now override the :meth:`~BaseCommand.add_arguments` method to
      add custom arguments accepted by your command.
      See :ref:`the example above <custom-commands-options>`.

.. attribute:: BaseCommand.output_transaction

  A boolean indicating whether the command outputs SQL
  statements; if ``True``, the output will automatically be
  wrapped with ``BEGIN;`` and ``COMMIT;``. Default value is
  ``False``.

.. attribute:: BaseCommand.requires_system_checks

.. versionadded:: 1.7

  A boolean; if ``True``, the entire Django project will be checked for
  potential problems prior to executing the command. If
  ``requires_system_checks`` is missing, the value of
  ``requires_model_validation`` is used. If the latter flag is missing
  as well, the default value (``True``) is used. Defining both
  ``requires_system_checks`` and ``requires_model_validation`` will result
  in an error.

.. attribute:: BaseCommand.requires_model_validation

.. deprecated:: 1.7
    Replaced by ``requires_system_checks``

  A boolean; if ``True``, validation of installed models will be
  performed prior to executing the command. Default value is
  ``True``. To validate an individual application's models
  rather than all applications' models, call
  :meth:`~BaseCommand.validate` from :meth:`~BaseCommand.handle`.

.. attribute:: BaseCommand.leave_locale_alone

  A boolean indicating whether the locale set in settings should be preserved
  during the execution of the command instead of being forcibly set to 'en-us'.

  Default value is ``False``.

  Make sure you know what you are doing if you decide to change the value of
  this option in your custom command if it creates database content that
  is locale-sensitive and such content shouldn't contain any translations (like
  it happens e.g. with django.contrib.auth permissions) as making the locale
  differ from the de facto default 'en-us' might cause unintended effects. See
  the `Management commands and locales`_ section above for further details.

  This option can't be ``False`` when the
  :data:`~BaseCommand.can_import_settings` option is set to ``False`` too
  because attempting to set the locale needs access to settings. This condition
  will generate a :class:`CommandError`.

Methods
-------

:class:`BaseCommand` has a few methods that can be overridden but only
the :meth:`~BaseCommand.handle` method must be implemented.

.. admonition:: Implementing a constructor in a subclass

    If you implement ``__init__`` in your subclass of :class:`BaseCommand`,
    you must call :class:`BaseCommand`’s ``__init__``::

        class Command(BaseCommand):
            def __init__(self, *args, **kwargs):
                super(Command, self).__init__(*args, **kwargs)
                # ...

.. method:: BaseCommand.add_arguments(parser)

.. versionadded:: 1.8

    Entry point to add parser arguments to handle command line arguments passed
    to the command. Custom commands should override this method to add both
    positional and optional arguments accepted by the command. Calling
    ``super()`` is not needed when directly subclassing ``BaseCommand``.

.. method:: BaseCommand.get_version()

    Returns the Django version, which should be correct for all built-in Django
    commands. User-supplied commands can override this method to return their
    own version.

.. method:: BaseCommand.execute(*args, **options)

    Tries to execute this command, performing system checks if needed (as
    controlled by the :attr:`requires_system_checks` attribute). If the command
    raises a :class:`CommandError`, it's intercepted and printed to stderr.

.. admonition:: Calling a management command in your code

    ``execute()`` should not be called directly from your code to execute a
    command. Use :ref:`call_command <call-command>` instead.

.. method:: BaseCommand.handle(*args, **options)

    The actual logic of the command. Subclasses must implement this method.

.. method:: BaseCommand.check(app_configs=None, tags=None, display_num_errors=False)

.. versionadded:: 1.7

    Uses the system check framework to inspect the entire Django project for
    potential problems. Serious problems are raised as a :class:`CommandError`;
    warnings are output to stderr; minor notifications are output to stdout.

    If ``app_configs`` and ``tags`` are both ``None``, all system checks are
    performed. ``tags`` can be a list of check tags, like ``compatibility`` or
    ``models``.

.. method:: BaseCommand.validate(app=None, display_num_errors=False)

.. deprecated:: 1.7
    Replaced with the :djadmin:`check` command

    If ``app`` is None, then all installed apps are checked for errors.

.. _ref-basecommand-subclasses:

BaseCommand subclasses
----------------------

.. class:: AppCommand

A management command which takes one or more installed application labels as
arguments, and does something with each of them.

Rather than implementing :meth:`~BaseCommand.handle`, subclasses must
implement :meth:`~AppCommand.handle_app_config`, which will be called once for
each application.

.. method:: AppCommand.handle_app_config(app_config, **options)

    Perform the command's actions for ``app_config``, which will be an
    :class:`~django.apps.AppConfig` instance corresponding to an application
    label given on the command line.

.. class:: LabelCommand

A management command which takes one or more arbitrary arguments
(labels) on the command line, and does something with each of
them.

Rather than implementing :meth:`~BaseCommand.handle`, subclasses must implement
:meth:`~LabelCommand.handle_label`, which will be called once for each label.

.. method:: LabelCommand.handle_label(label, **options)

    Perform the command's actions for ``label``, which will be the
    string as given on the command line.

.. class:: NoArgsCommand

.. deprecated:: 1.8

    Use :class:`BaseCommand` instead, which takes no arguments by default.

A command which takes no arguments on the command line.

Rather than implementing :meth:`~BaseCommand.handle`, subclasses must implement
:meth:`~NoArgsCommand.handle_noargs`; :meth:`~BaseCommand.handle` itself is
overridden to ensure no arguments are passed to the command.

.. method:: NoArgsCommand.handle_noargs(**options)

    Perform this command's actions

.. _ref-command-exceptions:

Command exceptions
------------------

.. class:: CommandError

Exception class indicating a problem while executing a management
command.

If this exception is raised during the execution of a management
command from a command line console, it will be caught and turned into a
nicely-printed error message to the appropriate output stream (i.e., stderr);
as a result, raising this exception (with a sensible description of the
error) is the preferred way to indicate that something has gone
wrong in the execution of a command.

If a management command is called from code through
:ref:`call_command <call-command>`, it's up to you to catch the exception
when needed.