Dockerfile-coreos-pxe

This is a Radial Wheel repository for running dnsmasq as PXE+DHCP server serving Coreos images.

Setup

• Add your public ssh-key to the file "hub/config/pxelinux.cfg/default" where it says YOUR_SSH_PUBLIC_KEY_HERE
  • Also make sure to modify any of the other parameters as needed. They aren't one-size-fits-all. If you want to use cloud-config, you will need to modify this file heavily. Refer to the CoreOS documentation for that.
• Configure dnsmasq configuration in "hub/config/dnsmasq.conf" to suit your network.
  • DHCP must be deactivated on your router and no other DHCP server can be running.
  • The template configuration here has dnsmasq setup as DNS, DHCP, tftp and PXE server. At the bare minimum, this container should handle DHCP and PXE. You could theoretically have DNS and tftp handled by other means (or other containers), but thats left up to the user to optimize for their network.
  • Drop in additional configuration files into "hub/config/dnsmasq.d" for lists of dns hosts or other configuration segmenting.
• If you have additional files/folders to bake into the coreos image (typically, a pre-loaded cloud-config.yml in /usr/share/oem for example), you can create a tarball (with the root of the tarball the same as '/'), upload it into the hub and specify it's location with $AMEND_IMAGE. The entrypoint script will automatically amend it before serving it out.

A very important note: in order for DHCP to work, this container must run using the docker run --net host option. This option uses the hosts network stack as the containers own. That means you must choose ports in your dnsmasq configuration that will not conflict with any ports already used on the host. The --net host option is not as secure as other options, so make sure to use appropriately.

Tunables

Tunable environment variables; modify at runtime. Italics are defaults.

• $REFRESH_IMAGES: [True|False] Refresh images/files on container restart.
• $CACHE_IMAGES: [True|False] Store downloaded images/files; one per release channel. Useful when switching between release channels for testing.
• $RELEASE: ["stable"|"beta"|"alpha"] Which release to download/use.
• $SRV_DIR: ["/data/tftpboot"] Path for the folder to serve the tftpboot files from.
• $CONF_FILE: ["/config/dnsmasq.conf"] Path to dnsmasq.conf file.
• $DNS_CHECK: [True|False] Check for an active DNS service before attempting to download anything. Useful in preventing race conditions when this instance of dnsmasq is not configured to handle DNS, but another container/service/machine is.
• $AMEND_IMAGE: [nothing] Location of compressed or uncompressed tarball to merge with the coreos image before serving it via PXE.

Radial

Radial is a Docker container topology strategy that seeks to put the canon of Docker best-practices into simple, re-usable, and scalable images, dockerfiles, and repositories. Radial categorizes containers into 3 types: Axles, Hubs, and Spokes. A Wheel is a repository used to recreate an application stack consisting of any combination of all three types of containers. Check out the Radial documentation for more.

One of the main design goals of Radial containers is simple and painless modularity. All Spoke (application/binary) containers are designed to be run by themselves as a service (a Wheel consisting of a Hub container for configuration and a Spoke container for the running binary) or as part of a larger stack as a Wheel of many Spokes all joined by the Hub container (database, application code, web server, backend services etc.). Check out the Wheel tutorial for some more details on how this works.

Note also that for now, Radial makes use of Fig for all orchestration, demonstration, and testing. Radial is just a collection of images and strategies, so technically, any orchestration tool can work. But Fig was the leanest and most logical to use for now.

How to Use

Static Build

In case you need to modify the entrypoint script, the Dockerfile itself, create your "config" branch for dynamic building, or just prefer to build your own from scratch, then you can do the following:

1. Clone this repository
2. Make whatever changes needed to configuration and add whatever files
3. fig up

Dynamic Build

A standard feature of all Radial images is their ability to be used dynamically. This means that since great care is made to separate the application code from it's configuration, as long as you make your application configuration available as a git repository, and in it's own "config" branch as per the guidelines in the Wheel template, no building of any images will be necessary at deploy time. This has many benefits as it allows rapid deployment and configuration without any wait time in the building process. However:

Dynamic builds will not commit your configuration files into any resulting images like static builds.

Static builds do a "COPY" of files into the image before exposing the directories as volumes. Dynamic builds do a git fetch at run time and the resulting data is downloaded to an already existing volume location, which is now free from Docker versioning. Both methods have their advantages and disadvantages. Deploying the same exact configuration might benefit from a single image built statically whereas deploying many different disposable configurations rapidly are best done dynamically with no building.

To run dynamically:

1. Modify the fig-dynamic.yml file to point at your own Wheel repository location by setting the $WHEEL_REPO variable. When run, the Hub container will pull the "config" branch of that repository and use it to run the Spoke container with your own configuration.
2. fig -f fig-dynamic.yml up

License

MIT

Credits

Much thanks to Jérôme Petazzoni for PXE, which this container is mainly based off of.