Why can't we fully use a declarative approach in all playbooks in Ansible?

Ansible strives for declarativity by describing the desired state of the system. However, full declarativity is limited due to the need to perform a sequence of operations and dependencies between tasks, which sometimes requires an imperative approach to correctly achieve the end state. Some modules or tasks may require actions to be performed in a specific order or depend on the output of previous tasks.

For example:

1. Conditional task execution: Using when: to execute a task only under certain conditions can violate pure declarativity, as the condition often relies on the current (not desired final) state or the result of a previous task.
2. Using register and changed_when: Saving the output of a task in a variable (register) and using this output in subsequent tasks or to determine a 'change' (changed_when) is also an imperative element.
3. Modules that execute commands: Modules like command, shell, script are inherently imperative because they simply execute a given command, not guaranteeing or describing the final state.

An example of an imperative element:

- name: Ensure service is running
  systemd:
    name: my_service
    state: started
  register: service_status

- name: Restart service if started
  command: systemctl restart my_service
  when: service_status.changed

Here, the second task executes only if the first task changed the service state (e.g., started it). This is an imperative condition dependent on the result of the previous action.

Overall, Ansible offers a hybrid approach: it tries to be as declarative as possible but leaves room (sometimes necessary) for using imperative constructs where deployment and configuration management logic dictate.