Python renderer that includes a Pythonic Object based interface


Evan Borgstrom <>

Let's take a look at how you use pyobjects in a state file. Here's a quick example that ensures the /tmp directory is in the correct state.

1 #!pyobjects
3 File.managed("/tmp", user='root', group='root', mode='1777')

Nice and Pythonic!

By using the "shebang" syntax to switch to the pyobjects renderer we can now write our state data using an object based interface that should feel at home to python developers. You can import any module and do anything that you'd like (with caution, importing sqlalchemy, django or other large frameworks has not been tested yet). Using the pyobjects renderer is exactly the same as using the built-in Python renderer with the exception that pyobjects provides you with an object based interface for generating state data.

Creating state data

Pyobjects takes care of creating an object for each of the available states on the minion. Each state is represented by an object that is the CamelCase version of its name (i.e. File, Service, User, etc), and these objects expose all of their available state functions (i.e. File.managed, Service.running, etc).

The name of the state is split based upon underscores (_), then each part is capitalized and finally the parts are joined back together.

Some examples:

  • postgres_user becomes PostgresUser

  • ssh_known_hosts becomes SshKnownHosts

Context Managers and requisites

How about something a little more complex. Here we're going to get into the core of how to use pyobjects to write states.

1 #!pyobjects
3 with Pkg.installed("nginx"):
4     Service.running("nginx", enable=True)
6     with Service("nginx", "watch_in"):
7         File.managed("/etc/nginx/conf.d/mysite.conf",
8                      owner='root', group='root', mode='0444',
9                      source='salt://nginx/mysite.conf')

The objects that are returned from each of the magic method calls are setup to be used a Python context managers (with) and when you use them as such all declarations made within the scope will automatically use the enclosing state as a requisite!

The above could have also been written use direct requisite statements as.

1 #!pyobjects
3 Pkg.installed("nginx")
4 Service.running("nginx", enable=True, require=Pkg("nginx"))
5 File.managed("/etc/nginx/conf.d/mysite.conf",
6              owner='root', group='root', mode='0444',
7              source='salt://nginx/mysite.conf',
8              watch_in=Service("nginx"))

You can use the direct requisite statement for referencing states that are generated outside of the current file.

1 #!pyobjects
3 # some-other-package is defined in some other state file
4 Pkg.installed("nginx", require=Pkg("some-other-package"))

The last thing that direct requisites provide is the ability to select which of the SaltStack requisites you want to use (require, require_in, watch, watch_in, use & use_in) when using the requisite as a context manager.

1 #!pyobjects
3 with Service("my-service", "watch_in"):
4     ...

The above example would cause all declarations inside the scope of the context manager to automatically have their watch_in set to Service("my-service").

Including and Extending

To include other states use the include() function. It takes one name per state to include.

To extend another state use the extend() function on the name when creating a state.

1 #!pyobjects
3 include('http', 'ssh')
5 Service.running(extend('apache'),
6                 watch=[File('/etc/httpd/extra/httpd-vhosts.conf')])

Importing from other state files

Like any Python project that grows you will likely reach a point where you want to create reusability in your state tree and share objects between state files, Map Data (described below) is a perfect example of this.

To facilitate this Python's import statement has been augmented to allow for a special case when working with a Salt state tree. If you specify a Salt url (salt://...) as the target for importing from then the pyobjects renderer will take care of fetching the file for you, parsing it with all of the pyobjects features available and then place the requested objects in the global scope of the template being rendered.

This works for all types of import statements; import X, from X import Y, and from X import Y as Z.

1 #!pyobjects
3 import salt://myfile.sls
4 from salt://something/data.sls import Object
5 from salt://something/data.sls import Object as Other

See the Map Data section for a more practical use.


  • Imported objects are ALWAYS put into the global scope of your template, regardless of where your import statement is.

Salt object

In the spirit of the object interface for creating state data pyobjects also provides a simple object interface to the __salt__ object.

A function named salt exists in scope for your sls files and will dispatch its attributes to the __salt__ dictionary.

The following lines are functionally equivalent:

1 #!pyobjects
3 ret =
4 ret = __salt__[''](bar)

Pillar, grain, mine & config data

Pyobjects provides shortcut functions for calling pillar.get, grains.get, mine.get & config.get on the __salt__ object. This helps maintain the readability of your state files.

Each type of data can be access by a function of the same name: pillar(), grains(), mine() and config().

The following pairs of lines are functionally equivalent:

 1 #!pyobjects
 3 value = pillar('foo:bar:baz', 'qux')
 4 value = __salt__['pillar.get']('foo:bar:baz', 'qux')
 6 value = grains('pkg:apache')
 7 value = __salt__['grains.get']('pkg:apache')
 9 value = mine('os:Fedora', 'network.interfaces', 'grain')
10 value = __salt__['mine.get']('os:Fedora', 'network.interfaces', 'grain')
12 value = config('foo:bar:baz', 'qux')
13 value = __salt__['config.get']('foo:bar:baz', 'qux')

Opts dictionary and SLS name

Pyobjects provides variable access to the minion options dictionary and the SLS name that the code resides in. These variables are the same as the opts and sls variables available in the Jinja renderer.

The following lines show how to access that information.

1 #!pyobjects
3 test_mode = __opts__["test"]
4 sls_name = __sls__

Map Data

When building complex states or formulas you often need a way of building up a map of data based on grain data. The most common use of this is tracking the package and service name differences between distributions.

To build map data using pyobjects we provide a class named Map that you use to build your own classes with inner classes for each set of values for the different grain matches.

 1 #!pyobjects
 3 class Samba(Map):
 4     merge = 'samba:lookup'
 5     # NOTE: priority is new to 2017.7.0
 6     priority = ('os_family', 'os')
 8     class Ubuntu:
 9         __grain__ = 'os'
10         service = 'smbd'
12     class Debian:
13         server = 'samba'
14         client = 'samba-client'
15         service = 'samba'
17     class RHEL:
18         __match__ = 'RedHat'
19         server = 'samba'
20         client = 'samba'
21         service = 'smb'


By default, the os_family grain will be used as the target for matching. This can be overridden by specifying a __grain__ attribute.

If a __match__ attribute is defined for a given class, then that value will be matched against the targeted grain, otherwise the class name's value will be be matched.

Given the above example, the following is true:

  1. Minions with an os_family of Debian will be assigned the attributes defined in the Debian class.

  2. Minions with an os grain of Ubuntu will be assigned the attributes defined in the Ubuntu class.

  3. Minions with an os_family grain of RedHat will be assigned the attributes defined in the RHEL class.

That said, sometimes a minion may match more than one class. For instance, in the above example, Ubuntu minions will match both the Debian and Ubuntu classes, since Ubuntu has an os_family grain of Debian and an os grain of Ubuntu. As of the 2017.7.0 release, the order is dictated by the order of declaration, with classes defined later overriding earlier ones. Additionally, 2017.7.0 adds support for explicitly defining the ordering using an optional attribute called priority.

Given the above example, os_family matches will be processed first, with os matches processed after. This would have the effect of assigning smbd as the service attribute on Ubuntu minions. If the priority item was not defined, or if the order of the items in the priority tuple were reversed, Ubuntu minions would have a service attribute of samba, since os_family matches would have been processed second.

To use this new data you can import it into your state file and then access your attributes. To access the data in the map you simply access the attribute name on the base class that is extending Map. Assuming the above Map was in the file samba/map.sls, you could do the following.

1 #!pyobjects
3 from salt://samba/map.sls import Samba
5 with Pkg.installed("samba", names=[Samba.server, Samba.client]):
6     Service.running("samba", name=Samba.service)
class salt.renderers.pyobjects.PyobjectsModule(name, attrs)

This provides a wrapper for bare imports.


This loads our states into the salt __context__

salt.renderers.pyobjects.render(template, saltenv='base', sls='', salt_data=True, **kwargs)