Creating generalized images without a provisioning agent

Applies to: ✔️ Linux VMs ✔️ Flexible scale sets

Microsoft Azure provides provisioning agents for Linux VMs in the form of the walinuxagent or cloud-init (recommended). But there could be a scenario when you don't want to use either of these applications for your provisioning agent, such as:

  • Your Linux distro/version doesn't support cloud-init/Linux Agent.
  • You require specific VM properties to be set, such as hostname.


If you do not require any properties to be set or any form of provisioning to happen you should consider creating a specialized image.

This article shows how you can set up your VM image to satisfy the Azure platform requirements and set the hostname, without installing a provisioning agent.

Networking and reporting ready

In order to have your Linux VM communicate with Azure components, a DHCP client is required. The client is used to retrieve a host IP, DNS resolution, and route management from the virtual network. Most distros ship with these utilities out-of-the-box. Tools that are tested on Azure by Linux distro vendors include dhclient, network-manager, systemd-networkd and others.


Currently creating generalized images without a provisioning agent only supports DHCP-enabled VMs.

After networking has been set up and configured, select "report ready". This tells Azure that the VM has been successfully provisioned.


Failing to report ready to Azure will result in your VM being rebooted!


An existing Marketplace image (in this case, a Debian Buster VM) with the Linux Agent (walinuxagent) removed and a custom python script added is the easiest way to tell Azure that the VM is "ready".

Create the resource group and base VM:

$ az group create --location eastus --name demo1

Create the base VM:

$ az vm create \
    --resource-group demo1 \
    --name demo1 \
    --location eastus \
    --ssh-key-value <ssh_pub_key_path> \
    --public-ip-address-dns-name demo1 \
    --image "debian:debian-10:10:latest"

Remove the image provisioning Agent

Once the VM is provisioning, you can connect to it via SSH and remove the Linux Agent:

$ sudo apt purge -y waagent
$ sudo rm -rf /var/lib/waagent /etc/waagent.conf /var/log/waagent.log

Add required code to the VM

Also inside the VM, because we've removed the Azure Linux Agent we need to provide a mechanism to report ready.

Python script

import http.client
import sys
from xml.etree import ElementTree

wireserver_ip = ''
wireserver_conn = http.client.HTTPConnection(wireserver_ip)

print('Retrieving goal state from the Wireserver')
    headers={'x-ms-version': '2012-11-30'}

resp = wireserver_conn.getresponse()

if resp.status != 200:
    print('Unable to connect with wireserver')

wireserver_goalstate ='utf-8')

xml_el = ElementTree.fromstring(wireserver_goalstate)

container_id = xml_el.findtext('Container/ContainerId')
instance_id = xml_el.findtext('Container/RoleInstanceList/RoleInstance/InstanceId')
incarnation = xml_el.findtext('Incarnation')
print(f'ContainerId: {container_id}')
print(f'InstanceId: {instance_id}')
print(f'Incarnation: {incarnation}')

# Construct the XML response we need to send to Wireserver to report ready.
health = ElementTree.Element('Health')
goalstate_incarnation = ElementTree.SubElement(health, 'GoalStateIncarnation')
goalstate_incarnation.text = incarnation
container = ElementTree.SubElement(health, 'Container')
container_id_el = ElementTree.SubElement(container, 'ContainerId')
container_id_el.text = container_id
role_instance_list = ElementTree.SubElement(container, 'RoleInstanceList')
role = ElementTree.SubElement(role_instance_list, 'Role')
instance_id_el = ElementTree.SubElement(role, 'InstanceId')
instance_id_el.text = instance_id
health_second = ElementTree.SubElement(role, 'Health')
state = ElementTree.SubElement(health_second, 'State')
state.text = 'Ready'

out_xml = ElementTree.tostring(
print('Sending the following data to Wireserver:')

        'x-ms-version': '2012-11-30',
        'Content-Type': 'text/xml;charset=utf-8',
        'x-ms-agent-name': 'custom-provisioning'

resp = wireserver_conn.getresponse()
print(f'Response: {resp.status} {resp.reason}')


Bash script


until [ "$attempts" -gt 5 ]
    echo "obtaining goal state - attempt $attempts"
    goalstate=$(curl --fail -v -X 'GET' -H "x-ms-agent-name: azure-vm-register" \
                                        -H "Content-Type: text/xml;charset=utf-8" \
                                        -H "x-ms-version: 2012-11-30" \
    if [ $? -eq 0 ]
       echo "successfully retrieved goal state"
    sleep 5

if [ "$retrieved_goal_state" != "true" ]
    echo "failed to obtain goal state - cannot register this VM"
    exit 1

container_id=$(grep ContainerId <<< "$goalstate" | sed 's/\s*<\/*ContainerId>//g' | sed 's/\r$//')
instance_id=$(grep InstanceId <<< "$goalstate" | sed 's/\s*<\/*InstanceId>//g' | sed 's/\r$//')

ready_doc=$(cat << EOF
<?xml version="1.0" encoding="utf-8"?>
<Health xmlns:xsi="" xmlns:xsd="">

until [ "$attempts" -gt 5 ]
    echo "registering with Azure - attempt $attempts"
    curl --fail -v -X 'POST' -H "x-ms-agent-name: azure-vm-register" \
                             -H "Content-Type: text/xml;charset=utf-8" \
                             -H "x-ms-version: 2012-11-30" \
                             -d "$ready_doc" \
    if [ $? -eq 0 ]
       echo "successfully register with Azure"
    sleep 5 # sleep to prevent throttling from wire server

Generic steps (if not using Python or Bash)

If your VM doesn't have Python installed or available, you can programmatically reproduce this above script logic with the following steps:

  1. Retrieve the ContainerId, InstanceId, and Incarnation by parsing the response from the WireServer: curl -X GET -H 'x-ms-version: 2012-11-30'

  2. Construct the following XML data, injecting the parsed ContainerId, InstanceId, and Incarnation from the above step:

  3. Post this data to WireServer: curl -X POST -H 'x-ms-version: 2012-11-30' -H "x-ms-agent-name: WALinuxAgent" -H "Content-Type: text/xml;charset=utf-8" -d "$REPORT_READY_XML"

Automating running the code at first boot

This demo uses systemd, which is the most common init system in modern Linux distros. So the easiest and most native way to ensure this report ready mechanism runs at the right time is to create a systemd service unit. You can add the following unit file to /etc/systemd/system (this example names the unit file azure-provisioning.service):

Description=Azure Provisioning

ExecStart=/usr/bin/python3 /usr/local/
ExecStart=/bin/bash -c "hostnamectl set-hostname $(curl \
    -H 'metadata: true' \
ExecStart=/usr/bin/systemctl disable azure-provisioning.service


This systemd service does three things for basic provisioning:

  1. Reports ready to Azure (to indicate that it came up successfully).
  2. Renames the VM based off of the user-supplied VM name by pulling this data from Azure Instance Metadata Service (IMDS). Note IMDS also provides other instance metadata, such as SSH Public Keys, so you can set more than the hostname.
  3. Disables itself so that it only runs on first boot and not on subsequent reboots.

With the unit on the filesystem, run the following to enable it:

$ sudo systemctl enable azure-provisioning.service

Now the VM is ready to be generalized and have an image created from it.

Completing the preparation of the image

Back on your development machine, run the following to prepare for image creation from the base VM:

$ az vm deallocate --resource-group demo1 --name demo1
$ az vm generalize --resource-group demo1 --name demo1

And create the image from this VM:

$ az image create \
    --resource-group demo1 \
    --source demo1 \
    --location eastus \
    --name demo1img

Now we're ready to create a new VM from the image. This can also be used to create multiple VMs:

$ IMAGE_ID=$(az image show -g demo1 -n demo1img --query id -o tsv)
$ az vm create \
    --resource-group demo12 \
    --name demo12 \
    --location eastus \
    --ssh-key-value <ssh_pub_key_path> \
    --public-ip-address-dns-name demo12 \
    --image "$IMAGE_ID"
    --enable-agent false


It is important to set --enable-agent to false because walinuxagent doesn't exist on this VM that is going to be created from the image.

The VM should be provisioned successfully. After Logging into the newly provisioning VM, you should be able to see the output of the report ready systemd service:

$ sudo journalctl -u azure-provisioning.service
-- Logs begin at Thu 2020-06-11 20:28:45 UTC, end at Thu 2020-06-11 20:31:24 UTC. --
Jun 11 20:28:49 thstringnopa systemd[1]: Starting Azure Provisioning...
Jun 11 20:28:54 thstringnopa python3[320]: Retrieving goal state from the Wireserver
Jun 11 20:28:54 thstringnopa python3[320]: ContainerId: 7b324f53-983a-43bc-b919-1775d6077608
Jun 11 20:28:54 thstringnopa python3[320]: InstanceId: fbb84507-46cd-4f4e-bd78-a2edaa9d059b._thstringnopa2
Jun 11 20:28:54 thstringnopa python3[320]: Sending the following data to Wireserver:
Jun 11 20:28:54 thstringnopa python3[320]: <Health><GoalStateIncarnation>1</GoalStateIncarnation><Container><ContainerId>7b324f53-983a-43bc-b919-1775d6077608</ContainerId><RoleInstanceList><Role><InstanceId>fbb84507-46cd-4f4e-bd78-a2edaa9d059b._thstringnopa2</InstanceId><Health><State>Ready</State></Health></Role></RoleInstanceList></Container></Health>
Jun 11 20:28:54 thstringnopa python3[320]: Response: 200 OK
Jun 11 20:28:56 thstringnopa bash[472]:   % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
Jun 11 20:28:56 thstringnopa bash[472]:                                  Dload  Upload   Total   Spent    Left  Speed
Jun 11 20:28:56 thstringnopa bash[472]: [158B blob data]
Jun 11 20:28:56 thstringnopa2 systemctl[475]: Removed /etc/systemd/system/
Jun 11 20:28:56 thstringnopa2 systemd[1]: azure-provisioning.service: Succeeded.
Jun 11 20:28:56 thstringnopa2 systemd[1]: Started Azure Provisioning.


If you implement your own provisioning code/agent, then you own the support of this code, Microsoft support will only investigate issues relating to the provisioning interfaces not being available. We're continually making improvements and changes in this area, so you must monitor for changes in cloud-init and Azure Linux Agent for provisioning API changes.

Next steps

For more information, see Linux provisioning.