salt/doc/ref/states/writing.rst

.. _state-modules:

=============
State Modules
=============

State Modules are the components that map to actual enforcement and management
of Salt states.

.. _writing-state-modules:

States are Easy to Write!
=========================

State Modules should be easy to write and straightforward. The information
passed to the SLS data structures will map directly to the states modules.

Mapping the information from the SLS data is simple, this example should
illustrate:

.. code-block:: yaml

    /etc/salt/master: # maps to "name", unless a "name" argument is specified below
      file.managed: # maps to <filename>.<function> - e.g. "managed" in https://github.com/saltstack/salt/tree/|repo_primary_branch|/salt/states/file.py
        - user: root # one of many options passed to the manage function
        - group: root
        - mode: 644
        - source: salt://salt/master

Therefore this SLS data can be directly linked to a module, function, and
arguments passed to that function.

This does issue the burden, that function names, state names and function
arguments should be very human readable inside state modules, since they
directly define the user interface.

.. admonition:: Keyword Arguments

    Salt passes a number of keyword arguments to states when rendering them,
    including the environment, a unique identifier for the state, and more.
    Additionally, keep in mind that the requisites for a state are part of the
    keyword arguments. Therefore, if you need to iterate through the keyword
    arguments in a state, these must be considered and handled appropriately.
    One such example is in the :mod:`pkgrepo.managed
    <salt.states.pkgrepo.managed>` state, which needs to be able to handle
    arbitrary keyword arguments and pass them to module execution functions.
    An example of how these keyword arguments can be handled can be found
    here_.

    .. _here: https://github.com/saltstack/salt/blob/v0.16.2/salt/states/pkgrepo.py#L163-183


Best Practices
==============

A well-written state function will follow these steps:

.. note::
    This is an extremely simplified example. Feel free to browse the `source
    code`_ for Salt's state modules to see other examples.

    .. _`source code`: https://github.com/saltstack/salt/tree/|repo_primary_branch|/salt/states

1. Set up the return dictionary and perform any necessary input validation
   (type checking, looking for use of mutually-exclusive arguments, etc.).

   .. code-block:: python

       ret = {'name': name,
              'result': False,
              'changes': {},
              'comment': ''}

       if foo and bar:
           ret['comment'] = 'Only one of foo and bar is permitted'
           return ret

2. Check if changes need to be made. This is best done with an
   information-gathering function in an accompanying :ref:`execution module
   <writing-execution-modules>`. The state should be able to use the return
   from this function to tell whether or not the minion is already in the
   desired state.

   .. code-block:: python

       result = __salt__['modname.check'](name)

3. If step 2 found that the minion is already in the desired state, then exit
   immediately with a ``True`` result and without making any changes.

   .. code-block:: python

       if result:
           ret['result'] = True
           ret['comment'] = '{0} is already installed'.format(name)
           return ret

4. If step 2 found that changes *do* need to be made, then check to see if the
   state was being run in test mode (i.e. with ``test=True``). If so, then exit
   with a ``None`` result, a relevant comment, and (if possible) a ``changes``
   entry describing what changes would be made.

   .. code-block:: python

       if __opts__['test']:
           ret['result'] = None
           ret['comment'] = '{0} would be installed'.format(name)
           ret['changes'] = result
           return ret

5. Make the desired changes. This should again be done using a function from an
   accompanying execution module. If the result of that function is enough to
   tell you whether or not an error occurred, then you can exit with a
   ``False`` result and a relevant comment to explain what happened.

   .. code-block:: python

       result = __salt__['modname.install'](name)

6. Perform the same check from step 2 again to confirm whether or not the
   minion is in the desired state. Just as in step 2, this function should be
   able to tell you by its return data whether or not changes need to be made.

   .. code-block:: python

       ret['changes'] = __salt__['modname.check'](name)

   As you can see here, we are setting the ``changes`` key in the return
   dictionary to the result of the ``modname.check`` function (just as we did
   in step 4). The assumption here is that the information-gathering function
   will return a dictionary explaining what changes need to be made. This may
   or may not fit your use case.

7. Set the return data and return!

   .. code-block:: python

       if ret['changes']:
           ret['comment'] = '{0} failed to install'.format(name)
       else:
           ret['result'] = True
           ret['comment'] = '{0} was installed'.format(name)

       return ret

Using Custom State Modules
==========================

Before the state module can be used, it must be distributed to minions. This
can be done by placing them into ``salt://_states/``. They can then be
distributed manually to minions by running :mod:`saltutil.sync_states
<salt.modules.saltutil.sync_states>` or :mod:`saltutil.sync_all
<salt.modules.saltutil.sync_all>`. Alternatively, when running a
:ref:`highstate <running-highstate>` custom types will automatically be synced.

NOTE: Writing state modules with hyphens in the filename will cause issues
with !pyobjects routines.  Best practice to stick to underscores.

Any custom states which have been synced to a minion, that are named the same
as one of Salt's default set of states, will take the place of the default
state with the same name. Note that a state module's name defaults to one based
on its filename (i.e. ``foo.py`` becomes state module ``foo``), but that its
name can be overridden by using a :ref:`__virtual__ function
<virtual-modules>`.

Cross Calling Execution Modules from States
===========================================

As with Execution Modules, State Modules can also make use of the ``__salt__``
and ``__grains__`` data. See :ref:`cross calling execution modules
<cross-calling-execution-modules>`.

It is important to note that the real work of state management should not be
done in the state module unless it is needed. A good example is the pkg state
module. This module does not do any package management work, it just calls the
pkg execution module. This makes the pkg state module completely generic, which
is why there is only one pkg state module and many backend pkg execution
modules.

On the other hand some modules will require that the logic be placed in the
state module, a good example of this is the file module. But in the vast
majority of cases this is not the best approach, and writing specific
execution modules to do the backend work will be the optimal solution.

.. _cross-calling-state-modules:

Cross Calling State Modules
===========================

All of the Salt state modules are available to each other and state modules can call
functions available in other state modules.

The variable ``__states__`` is packed into the modules after they are loaded into
the Salt minion.

The ``__states__`` variable is a :ref:`Python dictionary <python:typesmapping>`
containing all of the state modules. Dictionary keys are strings representing
the names of the modules and the values are the functions themselves.

Salt state modules can be cross-called by accessing the value in the
``__states__`` dict:

.. code-block:: python

    ret = __states__['file.managed'](name='/tmp/myfile', source='salt://myfile')

This code will call the `managed` function in the :mod:`file
<salt.states.file>` state module and pass the arguments ``name`` and ``source``
to it.

.. _state-return-data:

Return Data
===========

A State Module must return a dict containing the following keys/values:

- **name:** The same value passed to the state as "name".
- **changes:** A dict describing the changes made. Each thing changed should
  be a key, with its value being another dict with keys called "old" and "new"
  containing the old/new values. For example, the pkg state's **changes** dict
  has one key for each package changed, with the "old" and "new" keys in its
  sub-dict containing the old and new versions of the package. For example,
  the final changes dictionary for this scenario would look something like this:

  .. code-block:: python

    ret['changes'].update({'my_pkg_name': {'old': '',
                                           'new': 'my_pkg_name-1.0'}})


- **result:** A tristate value.  ``True`` if the action was successful,
  ``False`` if it was not, or ``None`` if the state was run in test mode,
  ``test=True``, and changes would have been made if the state was not run in
  test mode.

  +--------------------+-----------+------------------------+
  |                    | live mode | test mode              |
  +====================+===========+========================+
  | no changes         | ``True``  | ``True``               |
  +--------------------+-----------+------------------------+
  | successful changes | ``True``  | ``None``               |
  +--------------------+-----------+------------------------+
  | failed changes     | ``False`` | ``False`` or ``None``  |
  +--------------------+-----------+------------------------+

  .. note::

      Test mode does not predict if the changes will be successful or not,
      and hence the result for pending changes is usually ``None``.

      However, if a state is going to fail and this can be determined
      in test mode without applying the change, ``False`` can be returned.

- **comment:** A list of strings or a single string summarizing the result.
  Note that support for lists of strings is available as of Salt 2018.3.0.
  Lists of strings will be joined with newlines to form the final comment;
  this is useful to allow multiple comments from subparts of a state.
  Prefer to keep line lengths short (use multiple lines as needed),
  and end with punctuation (e.g. a period) to delimit multiple comments.

.. note::

    States should not return data which cannot be serialized such as frozensets.

Test State
==========

All states should check for and support ``test`` being passed in the options.
This will return data about what changes would occur if the state were actually
run. An example of such a check could look like this:

.. code-block:: python

    # Return comment of changes if test.
    if __opts__['test']:
        ret['result'] = None
        ret['comment'] = 'State Foo will execute with param {0}'.format(bar)
        return ret

Make sure to test and return before performing any real actions on the minion.

.. note::

    Be sure to refer to the ``result`` table listed above and displaying any
    possible changes when writing support for ``test``. Looking for changes in
    a state is essential to ``test=true`` functionality. If a state is predicted
    to have no changes when ``test=true`` (or ``test: true`` in a config file)
    is used, then the result of the final state **should not** be ``None``.

Watcher Function
================

If the state being written should support the watch requisite then a watcher
function needs to be declared. The watcher function is called whenever the
watch requisite is invoked and should be generic to the behavior of the state
itself.

The watcher function should accept all of the options that the normal state
functions accept (as they will be passed into the watcher function).

A watcher function typically is used to execute state specific reactive
behavior, for instance, the watcher for the service module restarts the
named service and makes it useful for the watcher to make the service
react to changes in the environment.

The watcher function also needs to return the same data that a normal state
function returns.


Mod_init Interface
==================

Some states need to execute something only once to ensure that an environment
has been set up, or certain conditions global to the state behavior can be
predefined. This is the realm of the mod_init interface.

A state module can have a function called **mod_init** which executes when the
first state of this type is called. This interface was created primarily to
improve the pkg state. When packages are installed the package metadata needs
to be refreshed, but refreshing the package metadata every time a package is
installed is wasteful. The mod_init function for the pkg state sets a flag down
so that the first, and only the first, package installation attempt will refresh
the package database (the package database can of course be manually called to
refresh via the ``refresh`` option in the pkg state).

The mod_init function must accept the **Low State Data** for the given
executing state as an argument. The low state data is a dict and can be seen by
executing the state.show_lowstate function. Then the mod_init function must
return a bool. If the return value is True, then the mod_init function will not
be executed again, meaning that the needed behavior has been set up. Otherwise,
if the mod_init function returns False, then the function will be called the
next time.

A good example of the mod_init function is found in the pkg state module:

.. code-block:: python

    def mod_init(low):
        '''
        Refresh the package database here so that it only needs to happen once
        '''
        if low['fun'] == 'installed' or low['fun'] == 'latest':
            rtag = __gen_rtag()
            if not os.path.exists(rtag):
                open(rtag, 'w+').write('')
            return True
        else:
            return False

The mod_init function in the pkg state accepts the low state data as ``low``
and then checks to see if the function being called is going to install
packages, if the function is not going to install packages then there is no
need to refresh the package database. Therefore if the package database is
prepared to refresh, then return True and the mod_init will not be called
the next time a pkg state is evaluated, otherwise return False and the mod_init
will be called next time a pkg state is evaluated.

Log Output
==========

You can call the logger from custom modules to write messages to the minion
logs. The following code snippet demonstrates writing log messages:

.. code-block:: python

    import logging

    log = logging.getLogger(__name__)

    log.info('Here is Some Information')
    log.warning('You Should Not Do That')
    log.error('It Is Busted')


Strings and Unicode
===================

A state module author should always assume that strings fed to the module
have already decoded from strings into Unicode. In Python 2, these will
be of type 'Unicode' and in Python 3 they will be of type ``str``. Calling
from a state to other Salt sub-systems, such as execution modules should
pass Unicode (or bytes if passing binary data). In the rare event that a state needs to write directly
to disk, Unicode should be encoded to a string immediately before writing
to disk. An author may use ``__salt_system_encoding__`` to learn what the
encoding type of the system is. For example,
`'my_string'.encode(__salt_system_encoding__')`.


Full State Module Example
=========================

The following is a simplistic example of a full state module and function.
Remember to call out to execution modules to perform all the real work. The
state module should only perform "before" and "after" checks.

1.  Make a custom state module by putting the code into a file at the following
    path: **/srv/salt/_states/my_custom_state.py**.

2.  Distribute the custom state module to the minions:

    .. code-block:: bash

        salt '*' saltutil.sync_states

3.  Write a new state to use the custom state by making a new state file, for
    instance **/srv/salt/my_custom_state.sls**.

4.  Add the following SLS configuration to the file created in Step 3:

    .. code-block:: yaml

        human_friendly_state_id:        # An arbitrary state ID declaration.
          my_custom_state:              # The custom state module name.
            - enforce_custom_thing      # The function in the custom state module.
            - name: a_value             # Maps to the ``name`` parameter in the custom function.
            - foo: Foo                  # Specify the required ``foo`` parameter.
            - bar: False                # Override the default value for the ``bar`` parameter.

Example state module
--------------------

.. code-block:: python

    import salt.exceptions

    def enforce_custom_thing(name, foo, bar=True):
        '''
        Enforce the state of a custom thing

        This state module does a custom thing. It calls out to the execution module
        ``my_custom_module`` in order to check the current system and perform any
        needed changes.

        name
            The thing to do something to
        foo
            A required argument
        bar : True
            An argument with a default value
        '''
        ret = {
            'name': name,
            'changes': {},
            'result': False,
            'comment': '',
            }

        # Start with basic error-checking. Do all the passed parameters make sense
        # and agree with each-other?
        if bar == True and foo.startswith('Foo'):
            raise salt.exceptions.SaltInvocationError(
                'Argument "foo" cannot start with "Foo" if argument "bar" is True.')

        # Check the current state of the system. Does anything need to change?
        current_state = __salt__['my_custom_module.current_state'](name)

        if current_state == foo:
            ret['result'] = True
            ret['comment'] = 'System already in the correct state'
            return ret

        # The state of the system does need to be changed. Check if we're running
        # in ``test=true`` mode.
        if __opts__['test'] == True:
            ret['comment'] = 'The state of "{0}" will be changed.'.format(name)
            ret['changes'] = {
                'old': current_state,
                'new': 'Description, diff, whatever of the new state',
            }

            # Return ``None`` when running with ``test=true``.
            ret['result'] = None

            return ret

        # Finally, make the actual change and return the result.
        new_state = __salt__['my_custom_module.change_state'](name, foo)

        ret['comment'] = 'The state of "{0}" was changed!'.format(name)

        ret['changes'] = {
            'old': current_state,
            'new': new_state,
        }

        ret['result'] = True

        return ret