Diving deeper into cloud-init

Applies to: ✔️ Linux VMs ✔️ Flexible scale sets

To learn more about cloud-init or troubleshoot it at a deeper level, you need to understand how it works. This document highlights the important parts, and explains the Azure specifics.

When cloud-init is included in a generalized image and a VM is created from that image, it processes configurations and runs through five stages during the initial boot. These stages show you at what point cloud-init applies its configurations.

Understand Cloud-Init configuration

Configuring a VM to run on a platform uses cloud-init to apply multiple configurations. The main configurations you interact with is User data (customData), which supports multiple formats. For more information, see User-Data Formats & cloud-init 21.2 documentation. You also have the ability to add and run scripts (/var/lib/cloud/scripts) for other configuration.

Preconfigured Azure Marketplace images

Some configurations are already baked into Azure Marketplace images that come with cloud-init.

  • Cloud data source - cloud-init contains code that can interact with cloud platforms, these codes are called 'datasources'. When a VM is created from a cloud-init image in Azure, cloud-init loads the Azure datasource, which interacts with the Azure metadata endpoints to get the VM specific configuration.

  • Runtime config (/run/cloud-init).

  • Image config (/etc/cloud), like /etc/cloud/cloud.cfg, /etc/cloud/cloud.cfg.d/*.cfg. An example of where this configuration is used in Azure, it's common for the Linux OS images with cloud-init to have an Azure datasource directive that tells cloud-init what datasource it should use, this configuration saves cloud-init time:

    sudo cat /etc/cloud/cloud.cfg.d/90_dpkg.cfg
    # to update this file, run dpkg-reconfigure cloud-init
    datasource_list: [ Azure ]

Cloud-init boot stages (processing configuration)

When you're provisioning VMs with cloud-init, there are five configuration boot stages. The output from these stages is visible in the logs.

  1. Generator Stage: The cloud-init systemd generator starts, and determines that cloud-init should be included in the boot goals, and if so, it enables cloud-init.

  2. Cloud-init Local Stage: Here, cloud-init looks for the local "Azure" datasource, which enables cloud-init to interface with Azure, and apply a networking configuration, including fallback.

  3. Cloud-init init Stage (Network): Networking should be online, and the NIC and route table information should be generated. At this stage, the modules listed in cloud_init_modules in /etc/cloud/cloud.cfg are run. The VM in Azure is mounted, the ephemeral disk is formatted, the hostname is set, along with other tasks.

    The following are some of the cloud_init_modules:

    - migrator
    - seed_random
    - bootcmd
    - write-files
    - growpart
    - resizefs
    - disk_setup
    - mounts
    - set_hostname
    - update_hostname
    - ssh

    After this stage, cloud-init sends a signal to the Azure platform that the VM has been provisioned successfully. Some modules may have failed, however not all module failures automatically result in a provisioning failure.

  4. Cloud-init Config Stage: At this stage, the modules in cloud_config_modules defined and listed in /etc/cloud/cloud.cfg runs.

  5. Cloud-init Final Stage: At this final stage, the modules in cloud_final_modules, listed in /etc/cloud/cloud.cfg, runs. Here modules that need to be run late in the boot process run, such as package installations and run scripts etc.

    • During this stage, you can run scripts by placing them in the directories under /var/lib/cloud/scripts:
      • per-boot - scripts within this directory, run on every reboot
      • per-instance - scripts within this directory run when a new instance is first booted
      • per-once - scripts within this directory run only once

Next steps

Troubleshooting cloud-init.