cloud-seed

cloud-seed is a very minimal alternative to cloud-init. It can set the hostname and write files based on directives provided in user data.

Motivation

Launching a server should be deterministic: launching the same image with the same configuration should have the same result. To that end, server images should include all software and base configuration required for the function of the server. Installing software and performing configuration via scripts on startup introduces opportunities for failure or non-determinism. To update software or fundamentally change the configuration of the server, a new image should be built. The image build process should be automated and frictionless to allow for rapid deployment when necessary.

cloud-seed exists because it is often necessary to provide some "seed" data to an otherwise-ready-to-run image: some values should not be baked into the image due to them varying from server to server.

User data format

The user data consists of #cloud-seed followed by a newline and then a JSON object. The entire user data including the #cloud-seed header can optionally be compressed with gzip, which is automatically detected.

Two keys are currently defined for the JSON object: files, which is an array of file objects, and hostname, which is a string. Both are optional. All other keys are ignored.

For example:

#cloud-seed
{
  "hostname": "myserver",
  "files": [
    {
      "path": "/etc/foobar/foobar.conf",
      "content": "Zm9vYmFyCg==",
      "encoding": "base64",
      "owner": "root:root",
      "permissions": "0644",
      "append": false
    }
  ]
}

If hostname is set, it will be passed to sethostname(2) to set the system hostname when cloud-seed runs. For this to work, cloud-seed must run as root or have the CAP_SYS_ADMIN capability.

If files is set, each object in the array describes a file that will be written by cloud-seed. The supported fields are:

Note about sensitive data

The user data on cloud systems is not generally appropriate for the storage of sensitive data, and therefore tools like cloud-seed (or cloud-init) should not be used to seed credentials or other sensitive data onto servers. Use functionality such as instance roles and temporary credentials to grant permissions to servers.

Supported data sources

cloud-seed can currently fetch user data from the metadata servers of:

• Alibaba Cloud
• AWS
• Exoscale
• Google Cloud
• OpenStack
• Oracle Cloud
• Vultr

Pull requests are welcome to add additional data sources.

DMI data is used to automatically detect which cloud cloud-seed is running in.

Compatibility with cloud-init

For compatibility with cloud-init, YAML is supported, the #cloud-config shebang is accepted, fqdn is accepted as an alias for hostname, and write_files is accepted as an alias for files. The encoding field for each file also supports cloud-init values b64, gz+base64, gzip+base64, gz+b64 and gzip+b64 as aliases for the corresponding cloud-seed values.

All other cloud-init directives are ignored. If cloud-init compatibility is not required, it is recommended to use cloud-seed's JSON format described above.

License

MIT or Apache 2.0, at your option.