Template Processing

YAML Merge directives

When a YAML configuration is loaded, will look for dictionary keys that match the following pattern:

'+'['?']['include'][*anchor][relative path][absolute path]

and treat them as merge directives that update the dictionary by processing the directive and merging in its resolved value.

Merge directives can have the following components:

merge key     ::=  "+"["?"]["include"][anchor][relative_path][absolute_path]
anchor        ::=  "*"[PCHAR]*[PCHAR except "."]
relative_path ::=  "."+
absolute_path ::=  ["/" PCHAR*]+
PCHAR         ::=  any printable character except "/"

Each of these components are optional but at least one needs to be present, otherwise the key is ignored and included in the final document.

A leading ‘?’ indicates that reference maybe missing, otherwise the processing will abort with an error.

include indicates that value of the merge key includes a yaml or json file to load.

*anchor: a reference to YAML anchor that appears in either the current document or, a file was specified, from that file.

relative path: ‘.’+ The path starting at current location this key appears in.

absolute path: [/path]+ A path that is resolved following <jsonpointer> RFC

If the directive contains “include”, its value can be a map or a string. If it is a string, treat the value as a file path or a URL to a YAML or JSON file. If it is a map, it must have a file key whose value is file path or URL, and optionally a “repository” key.

For other directives, if the value of the merge directive is empty, merge the result using algorithm described below. If the value is “raw”, include the result without any further processing.

The resolved value is merged into the directive’s dictionary using the following rules:

If the result of the lookup is not a JSON object or YAML map:

If the map containing the merge key has no other keys:

it will replaced by the result, otherwise abort processing with a merge error

If the map being replaced appears as an item in a list and the result of the merge is also a list:

the list is spliced in place.

(If you don’t want that behavior just wrap the include in another list, e.g [{+/list1: null}]

Otherwise, recursively merge the maps:

for each key in the result object

if the key doesn’t exist in the target:

add the key and value

if value of the target key is null treat it like an empty object

otherwise, if value of the target key is not an object:

ignore this key

if value of the target key is an object with a merge directive with the value “whiteout” ({+%: whiteout}):

omit the target key from the result object

if value of the target key is an object with a merge directive with the value “nullout” ({+%: nullout}):

set the target key to null in the result object

otherwise, merge two values following these rules.

Restoring merge directives

When saving YAML config file that contained merge directive will attempt restore them even if the configuration has changed – if the target object has changed, a new diff will be generated to reflect those changes.

Ansible Jinja2 Templates

Unfurl will process any Ansible-flavored Jinja2 templates it encounters in strings while processing an Ensemble template. It support the full suite of filters and lookup plugins provided by Ansible as well as the following predefined variables, filters, and lookup plugins:

Filters

eval

Evaluates the given expression or function. Equivalent to resolve_one For example: {{ "::instance1::anAttribute" | eval }}

map_value

Resolves any Eval Expressions or template strings embedded in the given map or list. Equivalent to unfurl.eval.map_value().

abspath

see abspath

get_dir

see get_dir

which

Returns the full path to the given executable, like the which shell command.

Lookup plugins

unfurl

Evaluates the given expression For example: {{ lookup("unfurl", "::instance1::anAttribute") }}

Variables

__unfurl

The current RefContext. This can be used to call Expression Functions as Jinja2 functions, for example: {{ __unfurl.to_label('a','b', sep='.') }}

__now

The current time in seconds since the epoch (1970) (Python’s time.time())

__python_executable

The location of the current python executable (Python’s sys.executable)

Eval Expressions

Contents

• Template Processing

  • YAML Merge directives

    • Restoring merge directives

  • Ansible Jinja2 Templates

    • Filters

    • Lookup plugins

    • Variables

  • Eval Expressions

    • Expression Query Syntax

    • Expression Function Syntax

    • Evaluation Semantics

      • Results flattened

When jobs are run Unfurl expressions that appear in the YAML configuration files are processed.

Expression Query Syntax

expr    ::=  segment? ("::" segment)*
segment ::=  [key] ("[" filter "]")* ["?"]
key     ::=  name | integer | var | "*"
filter  ::=  ['!'] [expr] [("!=" | "=") test]
test    ::=  var | ([^$[]:?])+
var     ::=  "$" name

Expression Function Syntax

Key	Value	Description
eval	expr or func	the expression to evaluate
vars?	map	define variables for the expression
select?	expr	apply expression to the result
foreach?	{key?, value?}	apply expression to each item in result
trace?	integer	enable detailed logging of evaluation
strict?	boolean	overrides strict evaluation

Evaluation Semantics

Each segment specifies a key in a resource or JSON/YAML object. :: is used as the segment delimitated to allow for keys that contain “.” and “/”

Path expressions evaluations always start with a list of one or more instances. and each segment selects the value associated with that key. If segment has one or more filters each filter is applied to that value – each is treated as a predicate that decides whether value is included or not in the results. If the filter doesn’t include a test the filter tests the existence or non-existence of the expression, depending on whether the expression is prefixed with a !. If the filter includes a test the left side of the test needs to match the right side. If the right side is not a variable, that string will be coerced to left side’s type before comparing it. If the left-side expression is omitted, the value of the segment’s key is used and if that is missing, the current value is used.

If the current value is a list and the key looks like an integer it will be treated like a zero-based index into the list. Otherwise the segment is evaluated again all values in the list and resulting value is a list. If the current value is a dictionary and the key is *, all values will be selected.

If a segment ends in ?, it will only include the first match. In other words, a?::b::c is a shorthand for a[b::c]::0::b::c. This is useful to guarantee the result of evaluating expression is always a single result.

The first segment: If the first segment is a variable reference the current value is set to that variable’s value. If the key in the first segment is empty (e.g. the expression starts with ::) the current value will be set to the evaluation of .all. If the key in the first segment starts with . it is evaluated against the initial “current resource”. Otherwise, the current value is set to the evaluation of .ancestors?. In other words, the expression will be the result of evaluating it against the first ancestor of the current resource that it matches.

If key or test needs to be a non-string type or contain reserved characters use a var reference instead.

Results flattened

When multiple steps resolve to lists the resultant lists are flattened. However if the final set of matches contain values that are lists those values are not flattened.

For example, given:

{x: [ {
        a: [{c:1}, {c:2}]
      },
      {
        a: [{c:3}, {c:4}]
      }
    ]
}

x::a::c resolves to: [1,2,3,4] not [[1,2], [3,4]])

(Justification: It is inconvenient and fragile to tie data structures to the particular form of a query. If you want preserve structure (e.g. to know which values are part of which parent value or resource) use a less deep path and iterate over results.)