Support bootc #830
Support bootc #830
Conversation
|
Skipping CI for Draft Pull Request. |
roles/edpm_kernel/defaults/main.yml
Outdated
| @@ -36,3 +36,4 @@ edpm_nova_compute_config_dir: /var/lib/config-data/ansible-generated/nova_libvir | |||
|
|
|||
| # KSM control | |||
| edpm_kernel_enable_ksm: false | |||
| edpm_use_bootc: false | |||
There was a problem hiding this comment.
We probably need a better way to implement this globally. But at least for testing purposes, this is what I've used to get something that deploys.
There was a problem hiding this comment.
This is why I added the edpm_bootc role in #813 so that we had a way to do it consistently across anywhere that needs it.
There was a problem hiding this comment.
Also, I created a new bootc branch for edpm-ansible: https://github.com/openstack-k8s-operators/edpm-ansible/tree/bootc
Can you propose this PR to the bootc branch instead?
I'll be reverting #813 from main until we are ready to merge all bootc support into main.
There was a problem hiding this comment.
Branch thing is done. But that role would need to be called from each and every playbook to detect and set the bootc variable right? I guess we can just add it as a ansibleVar and avoid calling the role each and every time we start a new service.
| - name: Push script | ||
| ansible.builtin.copy: | ||
| dest: /usr/local/sbin/containers-tmpwatch | ||
| dest: /var/lib/openstack/cron/containers-tmpwatch |
There was a problem hiding this comment.
/usr is immutable with bootc deployments. So I've proposed doing this in two different ways. 1, we bake the scripts into the container file:
https://github.com/openstack-k8s-operators/install_yamls/pull/950/files#diff-f8fb9af5355b45b9ca8936bf0d721c6f0e37e13b637f5598e2be19995dea23e7R45-R46
And 2. Which is this method of writing to /var/lib/openstack. I personally prefer doing it this way if we can agree on a common place for any scripts that we want to use. That saves us baking things into images and then trying to keep them in sync. Better imo to have them in edpm-ansible for now.
| ansible.builtin.include_role: | ||
| name: osp.edpm.edpm_container_standalone | ||
| vars: | ||
| edpm_container_standalone_service: ovn_controller | ||
| edpm_container_standalone_container_defs: | ||
| ovn_controller: "{{ lookup('template', 'ovn_controller.yaml.j2') | from_yaml }}" | ||
| edpm_container_standalone_kolla_config_files: | ||
| ovn_controller: "{{ lookup('template', 'kolla_ovn_controller.yaml.j2') | from_yaml }}" |
There was a problem hiding this comment.
All of this needs to stay in order to support both deployment methodologies. It can just be conditional like:
https://github.com/openstack-k8s-operators/edpm-ansible/pull/830/files#diff-34e3323585e197e806d463771e3b5132716048c41818b1318fecb2c0d8e36cd6R45
bshephar
force-pushed
the
support-bootc
branch
from
dd3d198 to
cde9667
Compare
bshephar
force-pushed
the
support-bootc
branch
12 times, most recently
from
ffd86e6 to
54f7101
Compare
roles/edpm_ovn/tasks/run.yml
Outdated
| {{ | ||
| ovn_controller_pod_spec | combine({ | ||
| 'spec': { | ||
| 'containers': ovn_controller_pod_spec.spec.containers | zip_longest([], [{'image': edpm_ovn_controller_agent_image}]) | map('combine') | list, |
There was a problem hiding this comment.
I believe that if we were to customize the image like this then by definition the container is no longer "logically bound". Instead, it would be considered a "floating" container per https://containers.github.io/bootc/logically-bound-images.html#comparison-with-default-podman-systemd-units
and also:
There is no mechanism to inject arbitrary arguments to the podman pull (or equivalent) invocation used by bootc.
which seems to apply additional mounts or other options passed to podman pull are not possible.
The dynamically-injected ConfigMaps[1][2] may provide some customization, but that is still not likely for the app container image itself, b/c once that is changed to some other image, then that no longer fits into how logically bound images should be managed with the lifecycle of the base bootc image itself.
Point being...if we choose to allow the ability to podman run any arbitrary image at runtime, then these really aren't logically bound images at all, but are considered "floating".
The question becomes, should we adopt logically bound images, and require our end users to be building new bootc images before deploying EDPM nodes, depending on if they need to customize any of the container images? We could ship a bootc image that had all the images logically bound, but if a user wanted to run a different one (from a partner, etc) they they would need to rebuild that image.
I do like the quadlet/systemd design, and I think we can still adopt that either way.
[1] https://containers.github.io/bootc/building/guidance.html?highlight=configmap#configuration
[2] containers/bootc#22
bshephar
force-pushed
the
support-bootc
branch
2 times, most recently
from
a0569c4 to
b590229
Compare
|
I created a new bootc branch for edpm-ansible: https://github.com/openstack-k8s-operators/edpm-ansible/tree/bootc Can you propose this PR to the bootc branch instead? I reverted #813 from main in #844 I think that was the only other bootc related PR that has merged. |
bshephar
force-pushed
the
support-bootc
branch
4 times, most recently
from
1afffaf to
51a8b34
Compare
Signed-off-by: Brendan Shephard <bshephar@redhat.com> dnf yum-utils Signed-off-by: Brendan Shephard <bshephar@redhat.com> nvme-package Signed-off-by: Brendan Shephard <bshephar@redhat.com>
This change writes systemd files to etc instead of /usr/share along with adding support for Python libraries baked into the bootc image. Signed-off-by: Brendan Shephard <bshephar@redhat.com>
This change moves the script we're using for the logs cronjob into the /var/lib/openstack/cron directory. This facilitates the bootc immutable filesystem where we can't write to /usr, while also consolidating scripts relevant to our deployment in a common place. Signed-off-by: Brendan Shephard <bshephar@redhat.com>
Signed-off-by: Brendan Shephard <bshephar@redhat.com>
Signed-off-by: Brendan Shephard <bshephar@redhat.com>
bshephar
force-pushed
the
support-bootc
branch
from
51a8b34 to
c04ac6a
Compare
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: bshephar The full list of commands accepted by this bot can be found here. The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
Signed-off-by: Brendan Shephard <bshephar@redhat.com>
bshephar
force-pushed
the
support-bootc
branch
from
ebb8305 to
64316b3
Compare
Signed-off-by: Brendan Shephard <bshephar@redhat.com>
Signed-off-by: Brendan Shephard <bshephar@redhat.com>
bshephar
force-pushed
the
support-bootc
branch
from
64316b3 to
216ec24
Compare
There was a problem hiding this comment.
The entire handling of kernel args needs to be reconsidered for bootc:
https://containers.github.io/bootc/building/kernel-arguments.html
Signed-off-by: Brendan Shephard <bshephar@redhat.com>
This PR adds a number of changes to roles in order to facilitate the use of image mode RHEL.