Salt style guide¶

Read Salt Best Practices and Salt Formulas Style.

State IDs¶

While state IDs with spaces are easier to read, they are also easier to mistype: for example, in require arguments. As such, prefer hyphens to spaces in state IDs.

State functions¶

cmd¶

In general, avoid the cmd.run function. For most system commands, Salt provides a state function.

Our use is limited to:

Activating and running a Python command within a virtual environment
Running a custom script that is specific to our services
Running a system command for which Salt has no relevant function (rare)

Excluding virtual environments, cmd.run is used less than 10 times in the repository.

When using cmd.run, you should set an onchanges requisite or a creates argument. Otherwise, a cmd.run function is run each time its state file is applied.

file¶

If possible, avoid file functions. For many system files, Salt provides a state function.

Our use is limited to:

Adding a system cron job, as these are easier to find in /etc/cron.d/.
Writing a custom file or creating a custom directory that is specific to our services
Updating a system file for which Salt has no relevant function (uncommon)

Note that unarchiving files (whether local or remote) should use the archive.extracted function.

Tip

Use file.keyvalue instead of file.append or file.replace, where possible.

Attention

If skip_verify is True and the source is remote, then the file will never be updated. Use source_hash, instead.

service¶

The Salt documentation states:

By default if a service is triggered to refresh due to a watch statement the service is restarted. If the desired behavior is to reload the service, then set the reload value to True.

Some configuration changes require a reload only, while others require a restart. To support both, we author IDs like:

apache2:
  service.running:
    - name: apache2
    - enable: True

apache2-reload:
  module.wait:
    - name: service.reload
    - m_name: apache2

proxy:
  apache_module.enabled:
    - name: proxy
    - watch_in:
      - service: apache2

enable conf letsencrypt.conf:
  apache_conf.enabled:
    - name: letsencrypt
    - watch_in:
      - module: apache2-reload

In this example, enabling the proxy module causes the apache2 service to restart, whereas enabling the letsencrypt configuration causes it to reload.

Includes¶

As the Salt documentation states, with respect to whether to use an include or the top file:

If a Salt state always needs some other state, then using an include is a better choice. If only some systems should receive both Salt states, including both states in the top file gives you the flexibility to choose which systems receive each.

In other words: If running state.apply my-state fails with an error like:

- Cannot extend ID 'my-id' in 'base:my-state'. It is not part of the high state.
  This is likely due to a missing include statement or an incorrectly typed ID.
  Ensure that a state with an ID of 'my-id' is available
  in environment 'base' and to SLS 'my-state'

then the state file that defines the my-id ID should be included in the my-state file. Otherwise, it shouldn’t.

Requisites¶

Instead of relying on ordering, it’s better to explicitly declare direct requisites. We use exclusively:

require is easier to reason about than require_in, because code typically declares its own dependencies.
watch_in is easier to reason about than watch, because it follows the direction of causation: if this state changes, then restart or reload that service.
onchanges makes the state apply only if its required state generates changes, and is used exclusively with the cmd.run function (which otherwise always applies).

We use require_in in exceptional circumstances: for example, to require a state created by a macro.

Macros¶

As the Salt documentation states:

Jinja macros are useful for one thing and one thing only: creating mini templates that can be reused and rendered on demand.

All macros are defined in lib.sls.

set_firewall() and unset_firewall() make sense as macros, because different state files might want to open or close different ports based on Pillar data. For example, the apache file opens or closes ports 80 and 443 based on the apache.public_access value.
create_user() makes sense as a macro, because users are created in many different contexts, and it is simpler to couple the user’s creation to that context, than to synchronize user creation and service configuration in separate places.
apache() makes sense as a macro, because sites are created in two different contexts: when processing apache.sites data in the apache file, and python_apps data in the python file. See #80.

Looping over Pillar data¶

A few state files loop over Pillar data:

core.rsyslog and core.logrotate
apache, included by the top file if the apache.sites key is set in Pillar data
python_apps, included by the state files of specific services
prometheus, included by the state file of the prometheus service, and by non-development targets in the top file

This pattern allows service-specific configuration values to live in Pillar, rather than in Salt.