Manipulating data

In many cases, you will need to perform complex operations with your variables. While Ansible is not recommended as a data processing/manipulation tool, you can use the existing Jinja2 templating in conjunction with the many added Ansible filters, lookups and tests to perform some very complex transformations.

Let’s start with a quick definition of each type of plugin:

lookups: Mainly used to query ‘external data’, in Ansible these were the primary part of loops using the with_<lookup> construct, but they can be used independently to return data for processing. They normally return a list due to their primary function in loops as mentioned previously. Used with the lookup or query Jinja2 operators.
filters: used to change/transform data, used with the | Jinja2 operator.
tests: used to validate data, used with the is Jinja2 operator.

Loops and list comprehensions

Most programming languages have loops (for, while, and so on) and list comprehensions to do transformations on lists including lists of objects. Jinja2 has a few filters that provide this functionality: map, select, reject, selectattr, rejectattr.

map: this is a basic for loop that just allows you to change every item in a list, using the ‘attribute’ keyword you can do the transformation based on attributes of the list elements.
select/reject: this is a for loop with a condition, that allows you to create a subset of a list that matches (or not) based on the result of the condition.
selectattr/rejectattr: very similar to the above but it uses a specific attribute of the list elements for the conditional statement.

Use a loop to create exponential backoff.

- name: try wait_for_connection up to 10 times with exponential delay
  ansible.builtin.wait_for_connection:
    delay: '{{ item | int }}'
    timeout: 1
  loop: '{{ range(1, 11) | map("pow", 2) }}'
  loop_control:
    extended: true
  ignore_errors: "{{ not ansible_loop.last }}"
  register: result
  when: result is not defined or result is failed

Extract keys from a dictionary matching elements from a list

The Python equivalent code would be:

chains = [1, 2]
for chain in chains:
    for config in chains_config[chain]['configs']:
        print(config['type'])

There are several ways to do it in Ansible, this is just one example:

Way to extract matching keys from a list of dictionaries

 tasks:
   - name: Show extracted list of keys from a list of dictionaries
     ansible.builtin.debug:
       msg: "{{ chains | map('extract', chains_config) | map(attribute='configs') | flatten | map(attribute='type') | flatten }}"
     vars:
       chains: [1, 2]
       chains_config:
           1:
               foo: bar
               configs:
                   - type: routed
                     version: 0.1
                   - type: bridged
                     version: 0.2
           2:
               foo: baz
               configs:
                   - type: routed
                     version: 1.0
                   - type: bridged
                     version: 1.1

Results of debug task, a list with the extracted keys

   ok: [localhost] => {
       "msg": [
           "routed",
           "bridged",
           "routed",
           "bridged"
       ]
   }

Get the unique list of values of a variable that vary per host

   vars:
       unique_value_list: "{{ groups['all'] | map ('extract', hostvars, 'varname') | list | unique}}"

Find mount point

In this case, we want to find the mount point for a given path across our machines, since we already collect mount facts, we can use the following:

Use selectattr to filter mounts into the list I can then sort and select the last from

  - hosts: all
    gather_facts: True
    vars:
       path: /var/lib/cache
    tasks:
    - name: The mount point for {{path}}, found using the Ansible mount facts, [-1] is the same as the 'last' filter
      ansible.builtin.debug:
       msg: "{{(ansible_facts.mounts | selectattr('mount', 'in', path) | list | sort(attribute='mount'))[-1]['mount']}}"

Omit elements from a list

The special omit variable ONLY works with module options, but we can still use it in other ways as an identifier to tailor a list of elements:

Inline list filtering when feeding a module option

   - name: Enable a list of Windows features, by name
     ansible.builtin.set_fact:
       win_feature_list: "{{ namestuff | reject('equalto', omit) | list }}"
     vars:
       namestuff:
         - "{{ (fs_installed_smb_v1 | default(False)) | ternary(omit, 'FS-SMB1') }}"
         - "foo"
         - "bar"

Another way is to avoid adding elements to the list in the first place, so you can just use it directly:

Using set_fact in a loop to increment a list conditionally

   - name: Build unique list with some items conditionally omitted
     ansible.builtin.set_fact:
        namestuff: ' {{ (namestuff | default([])) | union([item]) }}'
     when: item != omit
     loop:
         - "{{ (fs_installed_smb_v1 | default(False)) | ternary(omit, 'FS-SMB1') }}"
         - "foo"
         - "bar"

Combine values from same list of dicts

Combining positive and negative filters from the examples above, you can get a ‘value when it exists’ and a ‘fallback’ when it doesn’t.

Use selectattr and rejectattr to get the ansible_host or inventory_hostname as needed

   - hosts: localhost
     tasks:
       - name: Check hosts in inventory that respond to ssh port
         wait_for:
           host: "{{ item }}"
           port: 22
         loop: '{{ has_ah + no_ah }}'
         vars:
           has_ah: '{{ hostvars|dictsort|selectattr("1.ansible_host", "defined")|map(attribute="1.ansible_host")|list }}'
           no_ah: '{{ hostvars|dictsort|rejectattr("1.ansible_host", "defined")|map(attribute="0")|list }}'

Custom Fileglob Based on a Variable

This example uses Python argument list unpacking to create a custom list of fileglobs based on a variable.

Using fileglob with a list based on a variable.

  - hosts: all
    vars:
      mygroups:
        - prod
        - web
    tasks:
      - name: Copy a glob of files based on a list of groups
        copy:
          src: "{{ item }}"
          dest: "/tmp/{{ item }}"
        loop: '{{ q("fileglob", *globlist) }}'
        vars:
          globlist: '{{ mygroups | map("regex_replace", "^(.*)$", "files/\1/*.conf") | list }}'

Complex Type transformations

Jinja provides filters for simple data type transformations (int, bool, and so on), but when you want to transform data structures things are not as easy. You can use loops and list comprehensions as shown above to help, also other filters and lookups can be chained and used to achieve more complex transformations.

Create dictionary from list

In most languages, it is easy to create a dictionary (also known as map/associative array/hash and so on) from a list of pairs. In Ansible there are a couple of ways to do it, and the best one for you might depend on the source of your data.

These example produces {"a": "b", "c": "d"}

Simple list to dict by assuming the list is [key, value , key, value, …]

 vars:
     single_list: [ 'a', 'b', 'c', 'd' ]
     mydict: "{{ dict(single_list[::2] | zip_longest(single_list[1::2])) }}"

It is simpler when we have a list of pairs:

 vars:
     list_of_pairs: [ ['a', 'b'], ['c', 'd'] ]
     mydict: "{{ dict(list_of_pairs) }}"

Both end up being the same thing, with zip_longest transforming single_list to a list_of_pairs generator.

A bit more complex, using set_fact and a loop to create/update a dictionary with key value pairs from 2 lists:

Using set_fact to create a dictionary from a set of lists

    - name: Uses 'combine' to update the dictionary and 'zip' to make pairs of both lists
      ansible.builtin.set_fact:
        mydict: "{{ mydict | default({}) | combine({item[0]: item[1]}) }}"
      loop: "{{ (keys | zip(values)) | list }}"
      vars:
        keys:
          - foo
          - var
          - bar
        values:
          - a
          - b
          - c

This results in {"foo": "a", "var": "b", "bar": "c"}.

You can even combine these simple examples with other filters and lookups to create a dictionary dynamically by matching patterns to variable names:

Using ‘vars’ to define dictionary from a set of lists without needing a task

   vars:
       xyz_stuff: 1234
       xyz_morestuff: 567
       myvarnames: "{{ q('varnames', '^xyz_') }}"
       mydict: "{{ dict(myvarnames|map('regex_replace', '^xyz_', '')|list | zip(q('vars', *myvarnames))) }}"

A quick explanation, since there is a lot to unpack from these two lines:

The varnames lookup returns a list of variables that match “begin with xyz_”.

Then feeding the list from the previous step into the vars lookup to get the list of values. The * is used to ‘dereference the list’ (a pythonism that works in Jinja), otherwise it would take the list as a single argument.

Both lists get passed to the zip filter to pair them off into a unified list (key, value, key2, value2, …).

The dict function then takes this ‘list of pairs’ to create the dictionary.

An example of how to use facts to find a host’s data that meets condition X:

vars:
  uptime_of_host_most_recently_rebooted: "{{ansible_play_hosts_all | map('extract', hostvars, 'ansible_uptime_seconds') | sort | first}}"

An example to show a host uptime in days/hours/minutes/seconds (assuming facts were gathered).

- name: Show the uptime in days/hours/minutes/seconds
  ansible.builtin.debug:
   msg: Uptime {{ now().replace(microsecond=0) - now().fromtimestamp(now(fmt='%s') | int - ansible_uptime_seconds) }}