This datasource finds metadata and user-data from the Azure cloud platform.


walinuxagent has several functions within images. For cloud-init specifically, the relevant functionality it performs is to register the instance with the Azure cloud platform at boot so networking will be permitted. For more information about the other functionality of walinuxagent, see Azure’s documentation for more details. (Note, however, that only one of walinuxagent’s provisioning and cloud-init should be used to perform instance customisation.)

If you are configuring walinuxagent yourself, you will want to ensure that you have Provisioning.UseCloudInit set to y.

Builtin Agent

An alternative to using walinuxagent to register to the Azure cloud platform is to use the __builtin__ agent command. This section contains more background on what that code path does, and how to enable it.

The Azure cloud platform provides initial data to an instance via an attached CD formatted in UDF. That CD contains a ‘ovf-env.xml’ file that provides some information. Additional information is obtained via interaction with the “endpoint”.

To find the endpoint, we now leverage the dhcp client’s ability to log its known values on exit. The endpoint server is special DHCP option 245. Depending on your networking stack, this can be done by calling a script in /etc/dhcp/dhclient-exit-hooks or a file in /etc/NetworkManager/dispatcher.d. Both of these call a sub-command ‘dhclient_hook’ of cloud-init itself. This sub-command will write the client information in json format to /run/cloud-init/dhclient.hook/<interface>.json.

In order for cloud-init to leverage this method to find the endpoint, the cloud.cfg file must contain:

    set_hostname: False
    agent_command: __builtin__

If those files are not available, the fallback is to check the leases file for the endpoint server (again option 245).

You can define the path to the lease file with the ‘dhclient_lease_file’ configuration.


Azure provides the instance metadata service (IMDS) which is a REST service on providing additional configuration information to the instance. Cloud-init uses the IMDS for:

  • network configuration for the instance which is applied per boot
  • a preprovisioing gate which blocks instance configuration until Azure fabric is ready to provision


The following configuration can be set for the datasource in system configuration (in /etc/cloud/cloud.cfg or /etc/cloud/cloud.cfg.d/).

The settings that may be configured are:

  • agent_command: Either __builtin__ (default) or a command to run to getcw metadata. If __builtin__, get metadata from walinuxagent. Otherwise run the provided command to obtain metadata.

  • apply_network_config: Boolean set to True to use network configuration described by Azure’s IMDS endpoint instead of fallback network config of dhcp on eth0. Default is True. For Ubuntu 16.04 or earlier, default is False.

  • data_dir: Path used to read metadata files and write crawled data.

  • dhclient_lease_file: The fallback lease file to source when looking for custom DHCP option 245 from Azure fabric.

  • disk_aliases: A dictionary defining which device paths should be interpreted as ephemeral images. See cc_disk_setup module for more info.

  • hostname_bounce: A dictionary Azure hostname bounce behavior to react to metadata changes. The ‘hostname_bounce: command’ entry can be either the literal string ‘builtin’ or a command to execute. The command will be invoked after the hostname is set, and will have the ‘interface’ in its environment. If set_hostname is not true, then hostname_bounce will be ignored. An example might be:

    command:  ["sh", "-c", "killall dhclient; dhclient $interface"]

  • hostname_bounce: A dictionary Azure hostname bounce behavior to react to metadata changes. Azure will throttle ifup/down in some cases after metadata has been updated to inform dhcp server about updated hostnames.

  • set_hostname: Boolean set to True when we want Azure to set the hostname based on metadata.

Configuration for the datasource can also be read from a dscfg entry in the LinuxProvisioningConfigurationSet. Content in dscfg node is expected to be base64 encoded yaml content, and it will be merged into the ‘datasource: Azure’ entry.

An example configuration with the default values is provided below:

    agent_command: __builtin__
    apply_network_config: true
    data_dir: /var/lib/waagent
    dhclient_lease_file: /var/lib/dhcp/dhclient.eth0.leases
      ephemeral0: /dev/disk/cloud/azure_resource
      interface: eth0
      command: builtin
      policy: true
      hostname_command: hostname
    set_hostname: true


Userdata is provided to cloud-init inside the ovf-env.xml file. Cloud-init expects that user-data will be provided as base64 encoded value inside the text child of a element named UserData or CustomData which is a direct child of the LinuxProvisioningConfigurationSet (a sibling to UserName) If both UserData and CustomData are provided behavior is undefined on which will be selected.

In the example below, user-data provided is ‘this is my userdata’, and the datasource config provided is {"agent_command": ["start", "walinuxagent"]}. That agent command will take affect as if it were specified in system config.




When the user launches an instance, they provide a hostname for that instance. The hostname is provided to the instance in the ovf-env.xml file as HostName.

Whatever value the instance provides in its dhcp request will resolve in the domain returned in the ‘search’ request.

The interesting issue is that a generic image will already have a hostname configured. The ubuntu cloud images have ‘ubuntu’ as the hostname of the system, and the initial dhcp request on eth0 is not guaranteed to occur after the datasource code has been run. So, on first boot, that initial value will be sent in the dhcp request and that value will resolve.

In order to make the HostName provided in the ovf-env.xml resolve, a dhcp request must be made with the new value. Walinuxagent (in its current version) handles this by polling the state of hostname and bouncing (‘ifdown eth0; ifup eth0’ the network interface if it sees that a change has been made.

cloud-init handles this by setting the hostname in the DataSource’s ‘get_data’ method via ‘hostname $HostName’, and then bouncing the interface. This behavior can be configured or disabled in the datasource config. See ‘Configuration’ above.