Skip to content

Latest commit

 

History

History
129 lines (104 loc) · 8.07 KB

README.md

File metadata and controls

129 lines (104 loc) · 8.07 KB
Raw

BootcBlade

Ansible automation for deploying a KVM hypervisor using bootc and CentOS Stream.

BootcBlade

This Ansible automation uses bootc to create "the perfect" KVM hypervisor with ZFS, NFS + Samba, Cockpit, and Sanoid + Syncoid.

Usage - deploy on top of existing system

  1. Install a fresh CentOS Stream to the desired host - use the latest minimal install to save disk space on the resulting deployed machine
  2. Install podman on the host
  3. Generate an SSH key
  4. Create inventory using the example in the docs folder
  5. Make sure you have passwordless sudo or root access to desired host using your new SSH key
  6. Install needed Ansible collections ansible-galaxy install -r collections/requirements.yml
  7. Run ansible-playbook deploy.yml -i inventories/<your_inventory.yml>

Usage - create ISO for automatic installation

  1. You will need an existing machine with Podman and Ansible
  2. The Ansible can be run locally or remotely - just be sure to have root or passwordless sudo working
  3. Create inventory for a local or remote run using the example in the docs folder
  4. Run ansible-playbook iso.yml -i inventories/<your_inventory.yml>
  5. ISO will be created on the local or remote machine, /root/bootcblade-output/bootiso/install.iso

Credentials

Deploy

  • The SSH key (for the user) is added for the root user and baked into the image generation - the resulting BootcBlade image will include your SSH key
  • The user, ssh key, and password (if defined) are configured after deployment via Ansible

ISO

  • The user, SSH key, and password (if defined) are baked into the ISO - the resulting ISO will include your user, SSH key, and hashed password (if defined)

How It Works

Deploy

  1. A new or existing system must exist. This system should be as small as possible because its filesystem will persist in the resulting deployed machine
  2. bootcblade.containerfile is copied to the existing system, then podman build is used to build the image
  3. Once the image is built, the BootcBlade image is deployed to the system - then it is rebooted
  4. Ansible creates the user with (or without) the password and adds the SSH key

ISO

  1. An existing system must exist to build the ISO on, no changes to this system are made
  2. Running the Ansible will create the files necessary to generate the ISO - including the user with (or without) password and the SSH key
  3. Resulting ISO is stored in the /root directory
  4. Configuration files and all used container images are deleted

Updating

Updates happen in two ways. When deploy.yml is used, there is a systemd unit created (bootcblade-rebuild.service) that is created and will run once a week. This service depends on /root/bootcblade.containerfile to exist, as the container is rebuilt. If the build is successful, then the new container is staged. The default update service bootc-fetch-apply-updates.service is masked as well so it will not run.

If deploy.yml is not used (perhaps the system was installed using an ISO created by iso.yml), then there is no automated update process and the default bootc-fetch-apply-updates.service is still enabled. If you want the automatic update method, then ansible-playbook deploy.yml --tags configure-bootcblade will need to be run, either remotely or as localhost, and the required variables will need to be defined.

Troubleshooting

/root/bootcblade.containerfile is gone:

You can use update.yml to recreate this, assuming you have the correct inventory.

BootcBlade will no longer build

The default tag used for centos-bootc is referenced in templates/bootcblade.containerfile.j2 - its possible that there was a kernel update, or a release update, that breaks ZFS. Usually these issues are transient and resolve on their own. If you need a build now (perhaps for a fresh system) you can try and see if there is an older release (tag) from the upstream repo, and adjust it using the bootc_image_tag variable.

https://quay.io/repository/centos-bootc/centos-bootc?tab=tags&tag=latest

Variable Usage

This is a description of each variable, what it does, and a table to determine when it is needed.

  • create_user: This user will be created during deploy.yml and iso.yml
  • create_user_password: This password will be used for the created user
  • create_user_ssh_pub: This is a SSH pubkey that will be added to the created user during deploy.yml and iso.yml, also it is applied to the root user in deploy.yml
  • create_user_shell: This shell setting will be used for the created user only during deploy.yml
  • bootc_image_tag: Override the source image tag for deploy.yml, iso.yml, and update.yml
  • bootc_acknowledge: This setting is only effective when setting it to false, newer versions of bootc require an acknowledgment during deploy.yml but older versions break if this is defined - so this can override the default and remove that
  • ansible_user - This is an Ansible variable, useful for connecting to the initial machine with a different user during deploy.yml
  • ansible_connection - This is an Ansible variable, useful when running Ansible locally with iso.yml and update.yml

deploy.yml

Variable Used Required
create_user X -
create_user_password X -
create_user_ssh_pub X X
create_user_shell X -
bootc_image_tag X -
bootc_acknowledge X -

iso.yml

Variable Used Required
create_user X X
create_user_password X -
create_user_ssh_pub X X
create_user_shell - -
bootc_image_tag X -
bootc_acknowledge - -

update.yml

Variable Used Required
create_user - -
create_user_password - -
create_user_ssh_pub - -
create_user_shell - -
bootc_image_tag X -
bootc_acknowledge - -

Hacks / Workarounds

ZFS

I ran into a few problems with ZFS - the ZFS on Linux packages for CentOS didn't quite work for CentOS Stream. Yes the older kernel was still in use and didn't change major versions however sometimes Red Hat backported changes from newer kernels. I considered using Fedora Server (instead of CentOS Stream) however the problem was reverse then, sometimes Fedora changes major kernel versions mid-release. So I settled with using CentOS Stream for the base and the Fedora ZoL release packages. I may tweak the code or the exact release being used but this seems to be the most stable so far.

Cockpit

I ran into a problem where the cockpit-ws package would not install onto the base image containers/bootc#571. There was some advice in that thread about using the containerized version of cockpit-ws so that is what I am doing, however this is being applied after deployment via Ansible and not baked into the image. https://quay.io/repository/cockpit/ws

Using this containerized version of cockpit-ws also brought problems, using the privileged container caused mount points to be held inside the container. This meant once the container started, ZFS datasets could not be deleted since they were still "mounted" inside the container. To workaround this bastion mode is being used instead. That means to login to Cockpit you have to use the host host.containers.internal. SSL certificates can still be added to the /etc/cockpit/ws-certs.d directory - it is mounted into the container.

This also explains why I'm using rpm vs dnf to install the 45Drives Cockpit packages - they have a dependency on cockpit-ws that I need to override. Once the official cockpit-files package is released I will be using that instead of cockpit-navigator.