How to Set UserNames in Ansible Dynamic Inventory EC2

I have a simple Ansible dynamic inventory for AWS servers that looks like this.

---
plugin: aws_ec2

regions:
  - eu-west-2

keyed_groups:
  - key: tags.Name

hostnames:
# A list in order of precedence for hostname variables.
  - ip-address


 compose:
     ansible_host: _Applications_
     ansible_user: "'ubuntu'"

This works fine, except that I also have another instance that's Redhat.

Which means that when I try to do a simple ping command on all the hosts, it fails as the username ubuntu is only valid on one of the servers.

Is there a way to set group my inventory file so that I can add in the ec2-user username for a specific group maybe based on it's tag or something else.

I could do this easily with my static inventory but I'm not sure how to do this with a dynamic inventory.

I've tried setting my Ansible inventory as an environment variable

export ANSIBLE_INVENTORY=~/Users/inventory

And placed my aws_ec2.yamlin the inventory directory along with a group vars directory containing my different groups with default usernames in each of the different groups

username: ubuntu

username: ec2-user

and then setting my inventory file as such

compose:
   ansible_user: "{{ username }}"

But when Ansible tries to connect, it's using an admin username and not what's set in my group vars.

Is there a way to set the different usernames needed to connect to the different type of servers?

Solution

Per the example for the constructed plugin, you can use the keyed_group feature to create groups by ansible_distribution:

keyed_groups:
    # this creates a group per distro (distro_CentOS, distro_Debian) and assigns the hosts that have matching values to it,
    # using the default separator "_"
    - prefix: distro
      key: ansible_distribution

And then set ansible_user inside groups_vars/distro_Ubuntu.yaml and group_vars/distro_RedHat.yaml.

Also from the documentation, this requires fact caching to operate (because otherwise Ansible doesn't know the value of ansible_distribution at the time it's processing the keyed_groups setting).

I don't have access to AWS at the moment, but here's how I'm testing everything locally. Given an inventory that looks like:

$ tree inventory
inventory/
├── 00-hosts.yaml
└── 10-constructed.yaml

Where inventory/00-hosts.yaml looks like:

all:
  hosts:
    host0:
      ansible_host: localhost

And inventory/10-constructed.yaml looks like:

plugin: constructed
strict: false
groups:
  ipmi_hosts: ipmi_host|default(false)
keyed_groups:
  - prefix: "distro"
    key: ansible_distribution

And ansible.cfg looks like:

[defaults]
inventory = inventory
enable_plugins = constructed

gathering = smart
fact_caching = jsonfile
fact_caching_connection = ./.facts

The first time I run this playbook:

- hosts: all
  gather_facts: true
  tasks:
    - debug:
        var: group_names

The output of the debug task is:

TASK [debug] ******************************************************************************************************************************************************************************************************
ok: [host0] => {
    "group_names": [
        "ungrouped"
    ]
}

But because of the fact gathering and caching performed by the previous playbook run, the second time I run it the output is:

TASK [debug] ******************************************************************************************************************************************************************************************************
ok: [host0] => {
    "group_names": [
        "distro_Fedora"
    ]
}

Similarly, before the first playbook run ansible-inventory --graph outputs:

@all:
  |--@ungrouped:
  |  |--host0

But after running the playbook once, I get:

@all:
  |--@distro_Fedora:
  |  |--host0
  |--@ungrouped:

I've bundled this all into an example repository.