Dynamic Inventories

Create dynamic inventories in Azure based on tags, resource groups and more.



In this lab we will move away from static inventory lists and start generating groups automatically.

Dynamic inventories are integrated into Ansible as of version 2.8.

If you stumble across internet sites that reference a Python script called azure_rm.py then this is the way that dynamic inventories were done prior to v2.8. Those pages are now out of date.

Dynamic inventory based on resource group filters

  1. Change directory to the working area

    Go back to your ansible folder for the duration of these labs:

    cd ~/ansible
  2. Create a third VM

    Create another VM in the ansible_vms resource group.

    imageId=$(az image show --resource-group packer_images --name lab1 --output tsv --query id)
    az vm create --name vm3 \
    --resource-group ansible_vms \
    --image $imageId \
    --ssh-key-values "@~/.ssh/id_rsa.pub" \
    --vnet-name vnet \
    --subnet subnet \
    --output jsonc \

    Note that we skipped the tags on this third VM.

  3. Create a dynamic inventory file

    Create a configuration file called inventory.azure_rm.yml:

    plugin: azure_rm
    - ansible_vms
    auth_source: auto

    Note that Azure inventory config filenames must end with either azure_rm.yml or azure_rm.yaml

  4. Verify the dynamic inventory

    Check that the inventory works and the hosts can be managed:

    ansible all -m ping -i inventory.azure_rm.yml

    Ansible is now using hostnames rather than IP addresses. It will suffix your hostnames with a string to ensure uniqueness.

    You can override this behaviour using plain_host_names: yes in your yaml inventory file.

    Once the third VM is up and running, the command should return for all three VMs in the resource group.

The include_vm_resource_groups section in the YAML config file allows multiple resource groups to be specified. If this section is omitted then all resource groups will be included.

The https://docs.ansible.com/ansible/latest/plugins/inventory/azure_rm.html page shows some of the other options for creating dynamic inventories.

You can include vmss_resource_groups, define different authentication methods, and filter by including or excluding based on VM naming convention, tags and their values, location and the current powerstate.

Keyed groups

Keyed groups are very powerful as they will dynamically generate multiple groups based on the prefix and key.

  1. Add keyed groups

    Extend the inventory file with some keyed groups

    # Generate 'tag_(tag name)_(tag value)' for each tag on a VM.
    - prefix: tag
      key: tags
    # Generate 'loc_(location name)', depending on the VM's location
    - prefix: loc
      key: location
  2. Verify the keyed groups

    Check that the dynamic groups work:

    ansible tag_owner_citadel -m ping -i inventory.azure_rm.yml
    ansible loc_westeurope -m ping -i inventory.azure_rm.yml

    Your vm3 won’t be in the tag_owner_citadel group as it was created without any tags.

  3. Default to the dynamic inventory

    Switch the default inventory file to the dynamic inventory. Edit the ansible.cfg file, and set the default inventory away from the static hosts inventory to the new yaml file.

    inventory = ~/ansible/inventory.azure_rm.yml
    roles_path = ~/ansible/roles
    nocows = 1
  4. Clean up the static file

    Remove the old static hosts file:

    rm ~/ansible/hosts
  5. Confirm

    We are now working with dynamic inventories as the default.

    ansible loc_westeurope -m ping

It is up to you how you organise your groups of VMs.

You may wish to maintain named inventory files, each with filtered resource groups and VM types, and with its own keyed groups. You would then use the -i switch to specify the alternate inventory file.

That would be one way to organise your dynamic inventories. Instead we’ll maximise the default dynamic inventory yaml file to generate as many useful groups as possible.

Keyed groups format

In the next section we will add to the number of automatically generated dynamic groups by extending the keyed groups. Let’s take a moment to understand the format for each keyed groups entry in the config file.

The keyed groups all defined under the keyed_groups: section. Let’s look at the one you just added in to group by location:

- prefix: loc
  key: location

The prefix is a text string that prefixes the resulting auto generated groups. It is combined with underscore and then the key, which in this example is location. With our VMs this resulted in a group called loc_westeurope. (You can override the underscore default separator if desired.)

But how does Ansible know the region for our VMs? The answer is the Instance Metadata Service.

Instance Metadata Service

Adding the Azure dynamic inventories functionality has added a wide number of Azure platform values to the hostvars set. These values been pulled from the Azure instance metadata service. Let’s take a quick look at that now.

  1. Remind yourself of the public IP address for VM1

    az vm list-ip-addresses --resource-group ansible_vms --output table

    Example output:

    VirtualMachine    PublicIPAddresses    PrivateIPAddresses
    ----------------  -------------------  --------------------
  2. SSH onto VM1

    Example command:

    ssh richeney@
  3. Hit the Instance Metadata Service

    Run the following command to see the Azure environmental information for that Ubuntu server

    curl -sH Metadata:true "" | jq .

    Piping the minified JSON through jq will output coloured and prettified JSON

    Example output:

      "compute": {
        "azEnvironment": "AzurePublicCloud",
        "customData": "",
        "location": "westeurope",
        "name": "vm1",
        "offer": "",
        "osType": "Linux",
        "placementGroupId": "",
        "plan": {
          "name": "",
          "product": "",
          "publisher": ""
        "platformFaultDomain": "0",
        "platformUpdateDomain": "0",
        "provider": "Microsoft.Compute",
        "publicKeys": [
            "keyData": "<<REDACTED>>",
            "path": "/home/richeney/.ssh/authorized_keys"
        "publisher": "",
        "resourceGroupName": "ansible_vms",
        "resourceId": "/subscriptions/2ca40be1-7e80-4f2b-92f7-06b2123a68cc/resourceGroups/ansible_vms/providers/Microsoft.Compute/virtualMachines/vm1",
        "sku": "",
        "subscriptionId": "2ca40be1-7e80-4f2b-92f7-06b2123a68cc",
        "tags": "docker:true;owner:citadel",
        "version": "",
        "vmId": "056ac4bd-d9d8-474a-a990-f18f3c01729c",
        "vmScaleSetName": "",
        "vmSize": "Standard_DS1_v2",
        "zone": ""
      "network": {
        "interface": [
            "ipv4": {
              "ipAddress": [
                  "privateIpAddress": "",
                  "publicIpAddress": ""
              "subnet": [
                  "address": "",
                  "prefix": "24"
            "ipv6": {
              "ipAddress": []
            "macAddress": "000D3A2E8FDC"

    You will see your public key in compute.publicKeys.keyData. I have redacted mine.

    If you are bash scripting on a host then it is common to pull the metadata JSON into a file and then pull out values using jq. Or you can use a fully pathed value directly using the service, e.g.:

    pip=$(curl -sH Metadata:true "")
    echo $pip
  4. Exit the SSH session

    Exit back to your desktop:


Ansible hostvars

A subset of the Azure instance metadata is being pulled into the hostvars now that you are using dynamic inventories. You can view these hostvars using the debug module from Ansible.

  1. View the hostvars visible from one host

    ansible vm1_cf1d -m debug -a 'var=hostvars'

    Excerpt of example output for one of the hosts:

    vm1_cf1d | SUCCESS => {
        "hostvars": {
            "vm1_cf1d": {
                "ansible_check_mode": false,
                "ansible_diff_mode": false,
                "ansible_facts": {},
                "ansible_forks": 5,
                "ansible_host": "",
                "ansible_inventory_sources": [
                "ansible_playbook_python": "/usr/bin/python",
                "ansible_verbosity": 0,
                "ansible_version": {
                    "full": "2.8.3",
                    "major": 2,
                    "minor": 8,
                    "revision": 3,
                    "string": "2.8.3"
                "group_names": [
                "groups": {
                    "all": [
                    "loc_westeurope": [
                    "tag_docker_true": [
                    "tag_owner_citadel": [
                    "ungrouped": []
                "id": "/subscriptions/2ca40be1-7e80-4f2b-92f7-06b2123a68cc/resourceGroups/ansible_vms/providers/Microsoft.Compute/virtualMachines/vm1",
                "image": {
                    "id": "/subscriptions/2ca40be1-7e80-4f2b-92f7-06b2123a68cc/resourceGroups/images/providers/Microsoft.Compute/images/lab1"
                "inventory_dir": "/home/richeney/ansible",
                "inventory_file": "/home/richeney/ansible/inventory.azure_rm.yml",
                "inventory_hostname": "vm1_cf1d",
                "inventory_hostname_short": "vm1_cf1d",
                "location": "westeurope",
                "mac_address": "00-0D-3A-2E-8F-DC",
                "name": "vm1",
                "network_interface": "vm1VMNic",
                "network_interface_id": "/subscriptions/2ca40be1-7e80-4f2b-92f7-06b2123a68cc/resourceGroups/ansible_vms/providers/Microsoft.Network/networkInterfaces/vm1VMNic",
                "omit": "__omit_place_holder__5a75ee2f678d39048d7c4d288c68a897bb00e8df",
                "os_disk": {
                    "name": "vm1_disk1_c3111459458744568f03ba0a078661b4",
                    "operating_system_type": "linux"
                "plan": null,
                "playbook_dir": "/home/richeney/ansible",
                "powerstate": "running",
                "private_ipv4_addresses": [
                "provisioning_state": "succeeded",
                "public_dns_hostnames": [],
                "public_ip_id": "/subscriptions/2ca40be1-7e80-4f2b-92f7-06b2123a68cc/resourceGroups/ansible_vms/providers/Microsoft.Network/publicIPAddresses/vm1PublicIP",
                "public_ip_name": "vm1PublicIP",
                "public_ipv4_addresses": [
                "resource_group": "ansible_vms",
                "resource_type": "Microsoft.Compute/virtualMachines",
                "security_group": "vm1NSG",
                "security_group_id": "/subscriptions/2ca40be1-7e80-4f2b-92f7-06b2123a68cc/resourceGroups/ansible_vms/providers/Microsoft.Network/networkSecurityGroups/vm1NSG",
                "tags": {
                    "docker": "true",
                    "owner": "citadel"
                "virtual_machine_size": "Standard_DS1_v2",
                "vmid": "056ac4bd-d9d8-474a-a990-f18f3c01729c",
                "vmss": {}

Note that this output only shows the information for one host. Each host actually sees the hostvar info for others in the inventory.

You can see that location and tags are in the list, and this is where the current keyed groups are taking their information.

Adding simple hostvar keyed groups

We can use any of these hostvars to create additional keyed groups.

Let’s add keyed groups based on resource group, operating system type and VM size.

  1. Add metadata keyed groups

    Extend the keyed_groups: section of your inventory yaml file:

    # Generate rg_(resource_group)
    - prefix: rg
      key: resource_group
    # Generate os_type_(os_type)
    - prefix: os
      key: os_disk.operating_system_type
    # Generate vm_size_(vm_size)
    - prefix: vm_size
      key: virtual_machine_size
  2. List the groups and members

    List out the auto generate groups and the membership:

    ansible localhost -m debug -a 'var=groups.keys()'
    ansible localhost -m debug -a 'var=groups'

Using nested values and Jinja2 filters

If you want to use a nested value then just use the standard dot notation.

For example, the operating system type is within the os_disk hostvar:

    "os_disk": {
        "name": "vm1_disk1_c3111459458744568f03ba0a078661b4",
        "operating_system_type": "linux"

So your keyed group would look like:

# Generate os_type_(os_type)
- prefix: os
  key: os_disk.operating_system_type


Assuming that you have followed the labs all the way through, then one of your three VMs will not have any tags. We recommend using Azure Policy for creating default tags and flagging those that are non-compliant. You can also use defaults in the keyed groups to create dynamic groups of servers that do not have certain tags.

If you remember the tags keyed group then it was in this format:

# Generate 'tag_(tag name)_(tag value)' for each tag on a VM.
- prefix: tag
  key: tags

This is powerful as it creates groups for all combinations of tags and tag values. But can you use Ansible to generate a group of servers that are missing a required tag and value? What if we wanted to look at all servers that didn’t have an owners tag?

You can use the following format to get a specific set of owner tags keyed groups including an tag_owner_none group for those that do not have a defined owner value:

# Generate 'tag_owner_(owner value)'
- prefix: tag_owner
  key: tags.owner | default('none')

Note that the groups this generates overlap with those created by the first keyed group, but they will be merged naturally.

There is far more that you can do with the dynamic inventory file for Azure, so take time to review the other functionality shown in the documentation.

Complete the dynamic inventory file

  1. Finalise the dynamic inventory

    If you haven’t already done so, add those two blocks into the keyed groups area:

    # Generate os_type_(os_type)
    - prefix: os
      key: os_disk.operating_system_type
    # Generate 'tag_owner_(owner value)'
    - prefix: tag_owner
      key: tags.owner | default('none')
  2. List the groups

    ansible localhost -m debug -a 'var=groups'

Debug messages

We can now also use the extended set of hostvars in debug messages. This will be useful in the next lab within the Ansible Playbook.

  1. Display a custom debug message

{% raw %}

ansible all -m debug -a "msg='{{ inventory_hostname }} ({{ ansible_host }})) is {{ powerstate }}.'"

{% endraw %}

These use the Jinja2 expressions.

Coming up next

In the next section we will look at playbooks and publishing to the shared image gallery.


Help us improve

Azure Citadel is a community site built on GitHub, please contribute and send a pull request

Make a change