Skip to content

"Bluefin server pivot" #249

Closed
Closed

Description

@castrojo

These are notes for jeff:

a) We had a pure DDI image at first, Bluefin Server.
b) Using systemd's new installer. Keep this for homelab usage. This is the simple TUI thing.

  • Any automated config should be easy to bolt onto this.
  • ignition/cloud-init, etc. chef's choice - honestly I'd rather go pure systemd and drop all this stuff (cloud init requires python, etc) but "everyone uses this so we'll support it" is also fine.
  • Most of that stuff has a modern systemd equiv that we'd like to use:

c) Keep the UKIs, this let's us swap out kernels - we need to make UKIs for these (it's well documented and we can do it in GHA)

  • Default to Fedora CoreOS's kernel - this covers all the hardware we have
  • Flatcar LTS kernel - for slow usage
  • Ubuntu's Server kernel - easy swap, tons of users, weird hardware support, etc.
    d) Stick to DDI, bootc won't work for us here.
  • I force my agents to use context7 and follow systemd recommendations. Flatcar does this but isn't as up to date so you may find yourself doing something Flatcar may eventually adopt.

The first cut was just building the server OS with buildstream and mostly matching the versions in Flatcar. No API/ABI guarantees, just a fresh cut version. Andy gave me the idea of using Flatcar's actual raw files that are tested etc, and consuming those directly and then assembling with BST.

My long term goal is to rebuild Flatcar with buildstream in a way that would be official. (This is a prototype, I want to just show Thilo and crew how it would work - there's no "Someday we will move this to flatcar" since it's a rebuild.)

But also it could just work and be fine. It boots on VMs, bare metal is kicking my ass and I have a brand new minisforum sitting idling. Destroy what you need.

Activity

  1. Danathar commented on Sep 26, 2026

    @Danathar
    Contributor

    Triage from the hive contributor queue: no PR opened for this one. It's written as direction notes for a specific maintainer (Jeff). Nothing in it can be implemented yet without design decisions that conflict with current repo policy:

    • Kernel choice (c): The notes say "Default to Fedora CoreOS's kernel", with Flatcar LTS and Ubuntu Server kernels as alternate UKIs. Today the target UKI is built from the imported Flatcar /usr kernel (elements/flatcar/flatcar-usr.bst, patches/flatcar-kernel/, refactor: switch OS payload to Flatcar base, drop displaced FSDK components #132/feat(flatcar): add flatcar-usr.bst importing Flatcar /usr as the OS payload base #140). Changing the default kernel, and choosing a way to select or ship several UKIs (per-kernel DDIs? multiple UKIs in the ESP? one CI matrix?), is an architecture decision. AGENTS.md hard rule 1 would also need to change.
    • Provisioning (b): The notes leave "pure systemd (systemd-creds / first-boot)" vs "also support Ignition/cloud-init" to the implementer. The current tree removes Ignition and coreos-cloudinit in favour of systemd-creds provisioning (flatcar-usr.bst, tests/unit/test_flatcar_usr.py). Adding either one back is a policy call.
    • Long-term (d): Rebuilding Flatcar with BuildStream vs consuming Flatcar's tested raw files is the core direction question. revert: remove unapproved appliance identity design (#208) #235 reverted an unapproved design, so agents shouldn't pick a direction here without approval.

    Already true on main, so no change needed: DDI-only (no bootc), systemd-sysinstall installer, UKIs.

    Suggested next step: once a maintainer settles (1) the default kernel and how multiple kernels are delivered, and (2) the provisioning stance, split the work into scoped issues, e.g. "build Fedora CoreOS kernel UKI in GHA" or "add kernel-flavour selection to installer". Each of those can then be picked up directly.

    — hive: backend=omp model=anthropic/claude-opus-5-5 effort=medium

    🐝 Hive Agent: contributor | SHA: unknown

  2. Danathar commented on Sep 26, 2026

    @Danathar
    Contributor

    Correction to my comment above: changing the kernel doesn't necessarily mean changing AGENTS.md hard rule 1. That rule limits FSDK composition to the installer and allows digest-pinned Flatcar imports for the OS payload, so pinned Fedora CoreOS or Ubuntu kernel imports might fit within it. Which kernels to adopt, and how several are delivered, is still a maintainer choice. docs/skills/architecture-roadmap.md and CONTEXT.md have no approved kernel or provisioning plan yet.

    — hive: backend=omp model=anthropic/claude-opus-5-5 effort=medium

    🐝 Hive Agent: contributor | SHA: unknown

  3. hivecommons-hive commented on Oct 1, 2026

    @hivecommons-hive
    Contributor

    What to reply

    • /hive approve: go with the recommendation
    • /hive decision <your instructions>: go ahead, following your instructions
    • /hive help: show these again

    If this issue is waiting only for reporter confirmation after a merged fix, the reporter or a maintainer can reply /fixed instead.

    Only a maintainer with write access can un-park; everything else is ignored. Full command reference: https://github.com/hivecommons/hive/blob/v5/src/docs/maintainer-commands.md.

  4. hivecommons-hive commented on Oct 1, 2026

    @hivecommons-hive
    Contributor

    To start work, reply /hive approve (or /hive decision A). A plain reply acknowledges the issue but leaves it parked.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions