Cloud-init

From ArchWiki

Cloud-init is a package that contains utilities for early initialization of cloud instances. It is needed in Arch Linux images that are built with the intention of being launched in cloud environments like OpenStack, AWS etc.

Installation

Install the cloud-init package.

If you intend to use the growpart module you will also need the cloud-guest-utils package.

Configuration

This section only discusses the most basic settings. For a full list of available configuration options, please see the cloud-init documentation.

The main configuration file is /etc/cloud/cloud.cfg. Optionally, additional *.cfg files to be loaded can be placed in /etc/cloud/cloud.cfg.d. All of their contents will be merged.

The default configuration shipped with cloud-init 19.3 and later should work out of the box with most major cloud environments. In rough terms, it does the following:

  • Disable the root user, create a user arch for logging in
  • Rely on cloud-init's built-in detection for data sources
  • Run all modules known to work on Arch Linux

Depending on the use case, the default configuration might need to be adapted.

Default user configuration

The default configuration includes the following contents (comments omitted for brevity):

users:
   - default

The users to be added to the system. The special name default is just a reference to the default_user in the system_info section (see below), but the syntax supports configuring arbitrary users with many options. The first user in this list will be considered the "default" user by other modules, for example the one that sets up SSH keys passed in from the cloud environment.

disable_root: true

Disable root SSH access. You may also delete the root user password on the cloud image:

# passwd -d root
system_info:
   default_user:
     name: arch
     lock_passwd: true
     gecos: arch Cloud User
     groups: [wheel, adm]
     sudo: ["ALL=(ALL) NOPASSWD:ALL"]
     shell: /bin/bash

This is the specification of the distribution's default user:

  • the default user's name will be arch
  • the default user is password locked, which means you can only log into the instance with the SSH keys configured during boot
  • the default user will be added to the groups adm and wheel
  • the default user is allowed passwordless sudo usage
  • the default user's shell is /bin/bash

Note that the user specified here will only be created if the special user "default" is included in the users: section above (or the section is omitted entirely).

Configuring data sources

Data Sources define how the instance metadata is pulled during boot. This depends on the cloud environment (OpenStack, AWS, OpenNebula etc.) you are running your instance in. Under the hood, this translates to a corresponding module which implements a few methods defined in a common interface.

The default configuration specifies no data sources, which means that cloud-init will attempt to auto-detect the cloud environment. However, some environments cannot be detected or may require special configuration to work. In this case, the data sources to be used can be explictly specified and configured. Refer to the list of known data sources in the documentation.

To specify a list of data sources to be used in your /etc/cloud/cloud.cfg add something like this:

datasource_list: [ NoCloud, ConfigDrive, OpenNebula, Azure, AltCloud, OVF, MAAS, GCE, OpenStack, CloudSigma, Ec2, CloudStack, None ]

This instructs cloud-init what modules to load while trying to download instance metadata. Optionally further configuration parameters may be passed specific to each datasource as follows:

datasource:
  OpenStack:
    metadata_urls: [ 'http://169.254.169.254:80' ]
    dsmode: net

The above configuration tells OpenStack datasource to use the url http://169.254.169.254:80 to download metadata and to run after network initialization, both of which are the default behaviour and may be omitted.

Modules

Cloud-init comes with a set of modules that can be enabled or disabled in the configuration. The default configuration enables all modules that are known to work on Arch Linux. Omitted modules include e.g. those specific to other distributions or operating systems.

The fact that a module is enabled usually does not mean that it will actually do anything. It will however check if any configuration relevant to it was passed in, e.g. from the cloud environment via the data source. Only then it will attempt to act. As such, enabling all modules usually helps to maximize compatibility with cloud environments. Nevertheless, modules known to be not needed can be removed from configuration, e.g. to improve start-up times. You can use cloud-init analyze on a booted instance to see how much time was spent on individual modules.

Some modules declare to cloud-init which distributions they have been verified for. Even if you specify that you want to run them, they will refuse to run unless the distribution specified in cloud.cfg is one of the verified distributions for that module. If you need to override this behavior to run a module on Arch anyway, add the module to the unverified_modules section in the cloud config, e.g.:

unverified_modules: ['ssh-import-id']

Systemd integration

Package cloud-init provides four systemd.service(5), two systemd.target(5), and a systemd.generator(7), whose dependencies are constructed in a way that they are activated in the sequence listed:

  • cloud-init-generator. Determines availability of any data source and enables or disables cloud-init.target
  • cloud-init-local.service. Only requires the filesystems to be up. Executes cloud-init init --local
  • cloud-init.service. Requires the network to be up. Executes cloud-init init
  • cloud-config.target. Corresponds to the cloud-config upstart event "to inform third parties that cloud-config is available"
  • cloud-config.service. Executes cloud-init modules --mode=config
  • cloud-final.service. Executes cloud-init modules --mode=final
  • cloud-init.target. Reached when all services have been started

The Uplink Labs EC2 images have all of them enabled, although that appears to be overkill due to the dependencies. When preparing an image, enabling cloud-init.service and cloud-final.service should be sufficient. Note that this does not mean that the cloud-init services will actually be run - that still depends on the generator enabling the cloud-init.target on early boot.

See also the cloud-init boot stages documentation for more information.

Using

Archiso

Here is how to test an Archiso with cloud-init (see archlinux/archiso#27) using QEMU:

Create a user-data file in YAML format for cloud-init containing username(s) and public SSH key(s). You can either use the convenient create_cloud-init.sh script proposed in archiso Merge Request #117, or just hand write as below, as-is or adding any additional options shown on the cloud-init documentation. Beware that the #cloud-config is NOT a YAML comment, but is required to be present by cloud-init.

#cloud-config
users:
  - name: vorburger
    ssh_authorized_keys:
      - ssh-rsa (...)

We can use meta-data as well, but do not have to, so we can just make that an empty file (but it has to exist):

$ touch meta-data

Then build a cloud-init.iso containing (only) the user-data and meta-data using xorriso from libisoburn:

$ xorriso -as genisoimage -output cloud-init.iso -volid CIDATA -joliet -rock user-data meta-data

Now add this cloud-init.iso as an additional second drive to the VM, using e.g. qemu-system-x86_64 ... -cdrom cloud-init.iso, or if using archiso then with the run_archiso -c.

You should then be able to SSH into the machine using as the chosen user and public key supplied in user-data. Note that it may take a moment to be available, due to a delay at the boot menu; e.g. the standard releng archlinux*.iso will wait there for 30s at start-up. For frequent re-testing of new instances with changing hostkeys, ssh -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null may be handy.

Tip: Cloud-init's nocloud documenation explains an alternative approach, based on using -smbios type=1,serial=ds=nocloud. This can be accomplished locally on a host running QEMU and an ad hoc Python webserver as described in the cloud-init tutorial. To boot on bare metal you could just "burn" two separate USB sticks, for archlinux*.iso and cloud-init.iso, and boot from archlinux*.iso. When doing this, cloud-init will run using the configuration on cloud-init.iso.

Troubleshooting

FAQ

The cloud-init FAQ has useful information about how to debug cloud-init, including where its log files are, and how to re-run datasource detection and cloud-init during development.

Unmounted image

The first thing to check is often if the cloud-init image (ISO) is seen, with lsblk and mount.

"device /dev/sr0 with label cidata not a valid seed"

datasourcenocloud.py warning device /dev/sr0 with label cidata not a valid seed appears when the nocloud datasource sees an ISO that e.g. only contains user-data but no meta-data - both are required, even if one is empty.

"unhandled non-multipart [...] ssh_authorized_keys"

unhandled non-multipart (text-x-not-multipart) userdata 'b'ssh_authorized_keys appears when user-data YAML does not start with #cloud-config (no space, apparently; even though yamllint will tell you that it is missing starting space in comment).

See also