Salt Best Practices

Salt's extreme flexibility leads to many questions concerning the structure of configuration files.

This document exists to clarify these points through examples and code.

General rules

  1. Modularity and clarity should be emphasized whenever possible.
  2. Create clear relations between pillars and states.
  3. Use variables when it makes sense but don't overuse them.
  4. Store sensitive data in pillar.
  5. Don't use grains for matching in your pillar top file for any sensitive pillars.

Structuring States and Formulas

When structuring Salt States and Formulas it is important to begin with the directory structure. A proper directory structure clearly defines the functionality of each state to the user via visual inspection of the state's name.

Reviewing the MySQL Salt Formula it is clear to see the benefits to the end-user when reviewing a sample of the available states:

/srv/salt/mysql/files/
/srv/salt/mysql/client.sls
/srv/salt/mysql/map.jinja
/srv/salt/mysql/python.sls
/srv/salt/mysql/server.sls

This directory structure would lead to these states being referenced in a top file in the following way:

base:
  'web*':
    - mysql.client
    - mysql.python
  'db*':
    - mysql.server

This clear definition ensures that the user is properly informed of what each state will do.

Another example comes from the vim-formula:

/srv/salt/vim/files/
/srv/salt/vim/absent.sls
/srv/salt/vim/init.sls
/srv/salt/vim/map.jinja
/srv/salt/vim/nerdtree.sls
/srv/salt/vim/pyflakes.sls
/srv/salt/vim/salt.sls

Once again viewing how this would look in a top file:

/srv/salt/top.sls:

base:
  'web*':
    - vim
    - vim.nerdtree
    - vim.pyflakes
    - vim.salt
  'db*':
    - vim.absent

The usage of a clear top-level directory as well as properly named states reduces the overall complexity and leads a user to both understand what will be included at a glance and where it is located.

In addition Formulas should be used as often as possible.

注解

Formulas repositories on the saltstack-formulas GitHub organization should not be pointed to directly from systems that automatically fetch new updates such as GitFS or similar tooling. Instead formulas repositories should be forked on GitHub or cloned locally, where unintended, automatic changes will not take place.

Structuring Pillar Files

Pillars are used to store secure and insecure data pertaining to minions. When designing the structure of the /srv/pillar directory, the pillars contained within should once again be focused on clear and concise data which users can easily review, modify, and understand.

The /srv/pillar/ directory is primarily controlled by top.sls. It should be noted that the pillar top.sls is not used as a location to declare variables and their values. The top.sls is used as a way to include other pillar files and organize the way they are matched based on environments or grains.

An example top.sls may be as simple as the following:

/srv/pillar/top.sls:

base:
  '*':
    - packages

Any number of matchers can be added to the base environment. For example, here is an expanded version of the Pillar top file stated above:

/srv/pillar/top.sls:

base:
  '*':
    - packages
  'web*':
    - apache
    - vim

Or an even more complicated example, using a variety of matchers in numerous environments:

/srv/pillar/top.sls:

base:
  '*':
    - apache
dev:
  'os:Debian':
    - match: grain
    - vim
test:
  '* and not G@os: Debian':
    - match: compound
    - emacs

It is clear to see through these examples how the top file provides users with power but when used incorrectly it can lead to confusing configurations. This is why it is important to understand that the top file for pillar is not used for variable definitions.

Each SLS file within the /srv/pillar/ directory should correspond to the states which it matches.

This would mean that the apache pillar file should contain data relevant to Apache. Structuring files in this way once again ensures modularity, and creates a consistent understanding throughout our Salt environment. Users can expect that pillar variables found in an Apache state will live inside of an Apache pillar:

/srv/pillar/apache.sls:

apache:
  lookup:
    name: httpd
    config:
      tmpl: /etc/httpd/httpd.conf

While this pillar file is simple, it shows how a pillar file explicitly relates to the state it is associated with.

Variable Flexibility

Salt allows users to define variables in SLS files. When creating a state variables should provide users with as much flexibility as possible. This means that variables should be clearly defined and easy to manipulate, and that sane defaults should exist in the event a variable is not properly defined. Looking at several examples shows how these different items can lead to extensive flexibility.

Although it is possible to set variables locally, this is generally not preferred:

/srv/salt/apache/conf.sls:

{% set name = 'httpd' %}
{% set tmpl = 'salt://apache/files/httpd.conf' %}

include:
  - apache

apache_conf:
  file.managed:
    - name: {{ name }}
    - source: {{ tmpl }}
    - template: jinja
    - user: root
    - watch_in:
      - service: apache

When generating this information it can be easily transitioned to the pillar where data can be overwritten, modified, and applied to multiple states, or locations within a single state:

/srv/pillar/apache.sls:

apache:
  lookup:
    name: httpd
    config:
      tmpl: salt://apache/files/httpd.conf

/srv/salt/apache/conf.sls:

{% from "apache/map.jinja" import apache with context %}

include:
  - apache

apache_conf:
  file.managed:
    - name: {{ salt['pillar.get']('apache:lookup:name') }}
    - source: {{ salt['pillar.get']('apache:lookup:config:tmpl') }}
    - template: jinja
    - user: root
    - watch_in:
      - service: apache

This flexibility provides users with a centralized location to modify variables, which is extremely important as an environment grows.

Modularity Within States

Ensuring that states are modular is one of the key concepts to understand within Salt. When creating a state a user must consider how many times the state could be re-used, and what it relies on to operate. Below are several examples which will iteratively explain how a user can go from a state which is not very modular to one that is:

/srv/salt/apache/init.sls:

httpd:
  pkg.installed: []
  service.running:
    - enable: True

/etc/httpd/httpd.conf:
  file.managed:
    - source: salt://apache/files/httpd.conf
    - template: jinja
    - watch_in:
      - service: httpd

The example above is probably the worst-case scenario when writing a state. There is a clear lack of focus by naming both the pkg/service, and managed file directly as the state ID. This would lead to changing multiple requires within this state, as well as others that may depend upon the state.

Imagine if a require was used for the httpd package in another state, and then suddenly it's a custom package. Now changes need to be made in multiple locations which increases the complexity and leads to a more error prone configuration.

There is also the issue of having the configuration file located in the init, as a user would be unable to simply install the service and use the default conf file.

Our second revision begins to address the referencing by using - name, as opposed to direct ID references:

/srv/salt/apache/init.sls:

apache:
  pkg.installed:
    - name: httpd
  service.running:
    - name: httpd
    - enable: True

apache_conf:
  file.managed:
    - name: /etc/httpd/httpd.conf
    - source: salt://apache/files/httpd.conf
    - template: jinja
    - watch_in:
      - service: apache

The above init file is better than our original, yet it has several issues which lead to a lack of modularity. The first of these problems is the usage of static values for items such as the name of the service, the name of the managed file, and the source of the managed file. When these items are hard coded they become difficult to modify and the opportunity to make mistakes arises. It also leads to multiple edits that need to occur when changing these items (imagine if there were dozens of these occurrences throughout the state!). There is also still the concern of the configuration file data living in the same state as the service and package.

In the next example steps will be taken to begin addressing these issues. Starting with the addition of a map.jinja file (as noted in the Formula documentation), and modification of static values:

/srv/salt/apache/map.jinja:

{% set apache = salt['grains.filter_by']({
    'Debian': {
        'server': 'apache2',
        'service': 'apache2',
        'conf': '/etc/apache2/apache.conf',
    },
    'RedHat': {
        'server': 'httpd',
        'service': 'httpd',
        'conf': '/etc/httpd/httpd.conf',
    },
}, merge=salt['pillar.get']('apache:lookup')) %}

/srv/pillar/apache.sls:

apache:
  lookup:
    config:
      tmpl: salt://apache/files/httpd.conf

/srv/salt/apache/init.sls:

{% from "apache/map.jinja" import apache with context %}

apache:
  pkg.installed:
    - name: {{ apache.server }}
  service.running:
    - name: {{ apache.service }}
    - enable: True

apache_conf:
  file.managed:
    - name: {{ apache.conf }}
    - source: {{ salt['pillar.get']('apache:lookup:config:tmpl') }}
    - template: jinja
    - user: root
    - watch_in:
      - service: apache

The changes to this state now allow us to easily identify the location of the variables, as well as ensuring they are flexible and easy to modify. While this takes another step in the right direction, it is not yet complete. Suppose the user did not want to use the provided conf file, or even their own configuration file, but the default apache conf. With the current state setup this is not possible. To attain this level of modularity this state will need to be broken into two states.

/srv/salt/apache/map.jinja:

{% set apache = salt['grains.filter_by']({
    'Debian': {
        'server': 'apache2',
        'service': 'apache2',
        'conf': '/etc/apache2/apache.conf',
    },
    'RedHat': {
        'server': 'httpd',
        'service': 'httpd',
        'conf': '/etc/httpd/httpd.conf',
    },
}, merge=salt['pillar.get']('apache:lookup')) %}

/srv/pillar/apache.sls:

apache:
  lookup:
    config:
      tmpl: salt://apache/files/httpd.conf

/srv/salt/apache/init.sls:

{% from "apache/map.jinja" import apache with context %}

apache:
  pkg.installed:
    - name: {{ apache.server }}
  service.running:
    - name: {{ apache.service }}
    - enable: True

/srv/salt/apache/conf.sls:

{% from "apache/map.jinja" import apache with context %}

include:
  - apache

apache_conf:
  file.managed:
    - name: {{ apache.conf }}
    - source: {{ salt['pillar.get']('apache:lookup:config:tmpl') }}
    - template: jinja
    - user: root
    - watch_in:
      - service: apache

This new structure now allows users to choose whether they only wish to install the default Apache, or if they wish, overwrite the default package, service, configuration file location, or the configuration file itself. In addition to this the data has been broken between multiple files allowing for users to identify where they need to change the associated data.

Storing Secure Data

Secure data refers to any information that you would not wish to share with anyone accessing a server. This could include data such as passwords, keys, or other information.

As all data within a state is accessible by EVERY server that is connected it is important to store secure data within pillar. This will ensure that only those servers which require this secure data have access to it. In this example a use can go from an insecure configuration to one which is only accessible by the appropriate hosts:

/srv/salt/mysql/testerdb.sls:

testdb:
  mysql_database.present:
    - name: testerdb

/srv/salt/mysql/user.sls:

include:
  - mysql.testerdb

testdb_user:
  mysql_user.present:
    - name: frank
    - password: "test3rdb"
    - host: localhost
    - require:
      - sls: mysql.testerdb

Many users would review this state and see that the password is there in plain text, which is quite problematic. It results in several issues which may not be immediately visible.

The first of these issues is clear to most users -- the password being visible in this state. This means that any minion will have a copy of this, and therefore the password which is a major security concern as minions may not be locked down as tightly as the master server.

The other issue that can be encountered is access by users on the master. If everyone has access to the states (or their repository), then they are able to review this password. Keeping your password data accessible by only a few users is critical for both security and peace of mind.

There is also the issue of portability. When a state is configured this way it results in multiple changes needing to be made. This was discussed in the sections above but it is a critical idea to drive home. If states are not portable it may result in more work later!

Fixing this issue is relatively simple, the content just needs to be moved to the associated pillar:

/srv/pillar/mysql.sls:

mysql:
  lookup:
    name: testerdb
    password: test3rdb
    user: frank
    host: localhost

/srv/salt/mysql/testerdb.sls:

testdb:
  mysql_database.present:
    - name: {{ salt['pillar.get']('mysql:lookup:name') }}

/srv/salt/mysql/user.sls:

include:
  - mysql.testerdb

testdb_user:
  mysql_user.present:
    - name: {{ salt['pillar.get']('mysql:lookup:user') }}
    - password: {{ salt['pillar.get']('mysql:lookup:password') }}
    - host: {{ salt['pillar.get']('mysql:lookup:host') }}
    - require:
      - sls: mysql.testerdb

Now that the database details have been moved to the associated pillar file, only machines which are targeted via pillar will have access to these details. Access to users who should not be able to review these details can also be prevented while ensuring that they are still able to write states which take advantage of this information.