salt/doc/ref/states/writing.rst

State Modules

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

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:

/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/develop/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.

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 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.

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.

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

   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 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.

   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.

   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.

   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.

   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.

   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!

   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 saltutil.sync_states <salt.modules.saltutil.sync_states> or saltutil.sync_all <salt.modules.saltutil.sync_all>. Alternatively, when running a highstate <running-highstate> custom types will automatically be synced.

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 __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 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

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 Python dictionary <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:

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

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

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:

  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.

The return data can also, include the pchanges key, this stands for predictive changes. The pchanges key informs the State system what changes are predicted to occur.

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:

# 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:

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:

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:

   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:

   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

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': '',
        'pchanges': {},
        }

    # 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['pchanges'] = {
            '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