Micro-config-drive ("ucd" for short) provides a subset of functionality that is provided by cloud-init implementations. Specifically, ucd provides the basic options and functions to handle config-2 config-drive data for systems that have this config-drive data provided as a block device to the container or VM.
cloud-init is the standard way for cloud customers to initialize containers and virtual hosts. These virtual machines are usually provisioned in bulk and provided without any customization to cloud customers, which then require that they are customized for their particular purpose.
Several implementations exist that implement this functionality, and this implementation is fairly similar to the other ones. For reference, we're listing the other implementations here:
While generally it's preferred to use existing implementations, since it reduces duplicate code and makes stronger communities, in this project's case it was decided to forego extending and working on the existing implementations. Several factors were considered, and a short summary of the key points of that decision are listed below.
Speed is a significant factor. Interpreted languages have come a long way and are highly performant, especially if properly used. However, there is a significant cost of provisioning cloud nodes that have increased base storage costs due to the inclusion of libraries of interpreted languages. Since we expect people to prefer cloud nodes are minimal in size, having interpreted language libraries just for the sake of a cloud initialization script makes little sense, and we can reduce the storage need and copy times of cloud images significantly.
Language choice is a minor factor. Generally languages that have good exception handling are preferred. Since execution speed and low complexity are preferred for installation and boot-time critical tasks, lots of libraries should be avoided, as robustness is critical. Missed python exceptions could cause cloud-init execution to halt, resulting in an unusable cloud host, which is to be avoided at all cost. The drawbacks of C being fairly sparse are known, and debugging tools compensate for that.
We can shed additional overhead by eliminating unwanted functionality that a clearlinux cloud node does not need, but we can likely never recoup the initial cost of a base Python installation, as it is generally in the order of 100MB or more. Hence, we've opted to implement a version in C that has reduced library requrements. This has brought the size of this implementation down to under 1mb. Even considering used libraries, the resulting binary is small.
Currently, ucd requires the following prerequisites:
glib-2.0 >= 2.24.1yaml-0.1 >= 0.1.4libparted >= 3.1blkid >= 2.25.0json-glib-1.0As ucd is tooled with autotools, one shouldn't have to do more than:
$ sh autogen.sh
$ make
$ sudo make installmicro-config-drive and user-data-fetch should never be deployed
on existing, or provisioned, installations. Doing so may expose your
system to an attacker on a local network.
If deploying on a public could infrastructure, you should assure that routing to your instance is limited to trusted infrastructure nodes.
Most cloud service providers use LL addresses to serve cloud configs. If your CSP does not, you should verify that your use of non-LL addresses for cloud config service data is secure and does not expose your systems to unnecesary risk.
When running this project in a CSP environment, you should assure you are enabling the correct cloud service type. Do not under any circumstance enable several service types, this may further expose your instances to risk.
micro-config-drive is not immune to man-in-the-middle (MITM)
attacks, since it relies on unencryted HTTP services. This means a
local attacker may manipulate the received cloud service data. This
is a fundamental issue with cloud-config as it currently is offered
by CSP's, and not an implementation bug.
ucd is hosted on github. You can find releases, an issue tracker and git sources on github.com/clearlinux/micro-config-drive. For mailing lists, subscribe to dev@lists.clearlinux.org (via lists.clearlinux.org).
When submitting bug reports, it may be relevant to include the
log output of clouds- init. This can be done with the following
command: journalctl -b -u ucd. Omit the -b option if
you want output from prior to the last boot event.
This project has many contributors. Not all may be mentioned in the AUTHORS file.