YAML Idiosyncrasies

One of Salt's strengths, the use of existing serialization systems for representing SLS data, can also backfire. YAML is a general purpose system and there are a number of things that would seem to make sense in an sls file that cause YAML issues. It is wise to be aware of these issues. While reports or running into them are generally rare they can still crop up at unexpected times.

Spaces vs Tabs

YAML uses spaces, period. Do not use tabs in your SLS files! If strange errors are coming up in rendering SLS files, make sure to check that no tabs have crept in! In Vim, after enabling search highlighting with: :set hlsearch, you can check with the following key sequence in normal mode(you can hit ESC twice to be sure): /, Ctrl-v, Tab, then hit Enter. Also, you can convert tabs to 2 spaces by these commands in Vim: :set tabstop=2 expandtab and then :retab.

Indentation

The suggested syntax for YAML files is to use 2 spaces for indentation, but YAML will follow whatever indentation system that the individual file uses. Indentation of two spaces works very well for SLS files given the fact that the data is uniform and not deeply nested.

Nested Dictionaries

When dictionaries are nested within other data structures (particularly lists), the indentation logic sometimes changes. Examples of where this might happen include context and default options from the file.managed state:

/etc/http/conf/http.conf:
  file:
    - managed
    - source: salt://apache/http.conf
    - user: root
    - group: root
    - mode: 644
    - template: jinja
    - context:
        custom_var: "override"
    - defaults:
        custom_var: "default value"
        other_var: 123

Notice that while the indentation is two spaces per level, for the values under the context and defaults options there is a four-space indent. If only two spaces are used to indent, then those keys will be considered part of the same dictionary that contains the context key, and so the data will not be loaded correctly. If using a double indent is not desirable, then a deeply-nested dict can be declared with curly braces:

/etc/http/conf/http.conf:
  file:
    - managed
    - source: salt://apache/http.conf
    - user: root
    - group: root
    - mode: 644
    - template: jinja
    - context: {
      custom_var: "override" }
    - defaults: {
      custom_var: "default value",
      other_var: 123 }

Here is a more concrete example of how YAML actually handles these indentations, using the Python interpreter on the command line:

>>> import yaml
>>> yaml.safe_load(
...     """mystate:
...   file.managed:
...     - context:
...         some: var"""
... )
{'mystate': {'file.managed': [{'context': {'some': 'var'}}]}}
>>> yaml.safe_load(
...     """mystate:
...   file.managed:
...     - context:
...       some: var"""
... )
{'mystate': {'file.managed': [{'some': 'var', 'context': None}]}}

Note that in the second example, some is added as another key in the same dictionary, whereas in the first example, it's the start of a new dictionary. That's the distinction. context is a common example because it is a keyword arg for many functions, and should contain a dictionary.

Multi-line Strings

Similarly, when a multi-line string is nested within a list item (such as when using the contents argument for a file.managed state), the indentation must be doubled. Take for example the following state:

/tmp/foo.txt:
  file.managed:
    - contents: |
      foo
      bar
      baz

This is invalid YAML, and will result in a rather cryptic error when you try to run the state:

myminion:
    Data failed to compile:
----------
    Rendering SLS 'base:test' failed: could not find expected ':'; line 5

---
/tmp/foo.txt:
  file.managed:
    - contents: |
      foo
      bar    <======================
      baz

---

The correct indentation would be as follows:

/tmp/foo.txt:
  file.managed:
    - contents: |
        foo
        bar
        baz

True/False, Yes/No, On/Off

PyYAML will load these values as boolean True or False. Un-capitalized versions will also be loaded as booleans (true, false, yes, no, on, and off). This can be especially problematic when constructing Pillar data. Make sure that your Pillars which need to use the string versions of these values are enclosed in quotes. Pillars will be parsed twice by salt, so you'll need to wrap your values in multiple quotes, including double quotation marks (" ") and single quotation marks (' '). Note that spaces are included in the quotation type examples for clarity.

Multiple quoting examples looks like this:

- '"false"'
- "'True'"
- "'YES'"
- '"No"'

Note

When using multiple quotes in this manner, they must be different. Using "" "" or '' '' won't work in this case (spaces are included in examples for clarity).

The '%' Sign

The % symbol has a special meaning in YAML, it needs to be passed as a string literal:

cheese:
  ssh_auth.present:
    - user: tbortels
    - source: salt://ssh_keys/chease.pub
    - config: '%h/.ssh/authorized_keys'

Time Expressions

PyYAML will load a time expression as the integer value of that, assuming HH:MM. So for example, 12:00 is loaded by PyYAML as 720. An excellent explanation for why can be found here.

To keep time expressions like this from being loaded as integers, always quote them.

Note

When using a jinja load_yaml map, items must be quoted twice. For example:

{% load_yaml as wsus_schedule %}

FRI_10:
  time: '"23:00"'
  day: 6 - Every Friday
SAT_10:
  time: '"06:00"'
  day: 7 - Every Saturday
SAT_20:
  time: '"14:00"'
  day: 7 - Every Saturday
SAT_30:
  time: '"22:00"'
  day: 7 - Every Saturday
SUN_10:
  time: '"06:00"'
  day: 1 - Every Sunday
{% endload %}

YAML does not like "Double Short Decs"

If I can find a way to make YAML accept "Double Short Decs" then I will, since I think that double short decs would be awesome. So what is a "Double Short Dec"? It is when you declare a multiple short decs in one ID. Here is a standard short dec, it works great:

vim:
  pkg.installed

The short dec means that there are no arguments to pass, so it is not required to add any arguments, and it can save space.

YAML though, gets upset when declaring multiple short decs, for the record...

THIS DOES NOT WORK:

vim:
  pkg.installed
  user.present

Similarly declaring a short dec in the same ID dec as a standard dec does not work either...

ALSO DOES NOT WORK:

fred:
  user.present
  ssh_auth.present:
    - name: AAAAB3NzaC...
    - user: fred
    - enc: ssh-dss
    - require:
      - user: fred

The correct way is to define them like this:

vim:
  pkg.installed: []
  user.present: []

fred:
  user.present: []
  ssh_auth.present:
    - name: AAAAB3NzaC...
    - user: fred
    - enc: ssh-dss
    - require:
      - user: fred

Alternatively, they can be defined the "old way", or with multiple "full decs":

vim:
  pkg:
    - installed
  user:
    - present

fred:
  user:
    - present
  ssh_auth:
    - present
    - name: AAAAB3NzaC...
    - user: fred
    - enc: ssh-dss
    - require:
      - user: fred

YAML supports only plain ASCII

According to YAML specification, only ASCII characters can be used.

Within double-quotes, special characters may be represented with C-style escape sequences starting with a backslash ( \ ).

Examples:

- micro: "\u00b5"
- copyright: "\u00A9"
- A: "\x41"
- alpha: "\u0251"
- Alef: "\u05d0"

List of usable Unicode characters will help you to identify correct numbers.

Python can also be used to discover the Unicode number for a character:

repr("Text with wrong characters i need to figure out")

This shell command can find wrong characters in your SLS files:

find . -name '*.sls'  -exec  grep --color='auto' -P -n '[^\x00-\x7F]' \{} \;

Alternatively you can toggle the yaml_utf8 setting in your master configuration file. This is still an experimental setting but it should manage the right encoding conversion in salt after yaml states compilations.

Underscores stripped in Integer Definitions

If a definition only includes numbers and underscores, it is parsed by YAML as an integer and all underscores are stripped. To ensure the object becomes a string, it should be surrounded by quotes. More information here.

Here's an example:

>>> import yaml
>>> yaml.safe_load("2013_05_10")
20130510
>>> yaml.safe_load('"2013_05_10"')
'2013_05_10'

Automatic datetime conversion

If there is a value in a YAML file formatted 2014-01-20 14:23:23 or similar, YAML will automatically convert this to a Python datetime object. These objects are not msgpack serializable, and so may break core salt functionality. If values such as these are needed in a salt YAML file (specifically a configuration file), they should be formatted with surrounding strings to force YAML to serialize them as strings:

>>> import yaml
>>> yaml.safe_load("2014-01-20 14:23:23")
datetime.datetime(2014, 1, 20, 14, 23, 23)
>>> yaml.safe_load('"2014-01-20 14:23:23"')
'2014-01-20 14:23:23'

Additionally, numbers formatted like XXXX-XX-XX will also be converted (or YAML will attempt to convert them, and error out if it doesn't think the date is a real one). Thus, for example, if a minion were to have an ID of 4017-16-20 the minion would not start because YAML would complain that the date was out of range. The workaround is the same, surround the offending string with quotes:

>>> import yaml
>>> yaml.safe_load("4017-16-20")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python2.7/site-packages/yaml/__init__.py", line 93, in safe_load
    return load(stream, SafeLoader)
  File "/usr/local/lib/python2.7/site-packages/yaml/__init__.py", line 71, in load
    return loader.get_single_data()
  File "/usr/local/lib/python2.7/site-packages/yaml/constructor.py", line 39, in get_single_data
    return self.construct_document(node)
  File "/usr/local/lib/python2.7/site-packages/yaml/constructor.py", line 43, in construct_document
    data = self.construct_object(node)
  File "/usr/local/lib/python2.7/site-packages/yaml/constructor.py", line 88, in construct_object
    data = constructor(self, node)
  File "/usr/local/lib/python2.7/site-packages/yaml/constructor.py", line 312, in construct_yaml_timestamp
    return datetime.date(year, month, day)
ValueError: month must be in 1..12
>>> yaml.safe_load('"4017-16-20"')
'4017-16-20'

Keys Limited to 1024 Characters

Simple keys are limited by the YAML Spec to a single line, and cannot be longer that 1024 characters. PyYAML enforces these limitations (see here), and therefore anything parsed as YAML in Salt is subject to them.