You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
desktop: validate Bluefin first-boot services and hooks at runtime #19
Close the gap between copied files and functional first boot. Exercise the Bluefin Flatpak, privileged setup, user setup, statistics, input-remapper, update, and optional Tailscale paths on a booted Utah system, with absent optional packages handled explicitly.
Acceptance criteria
First boot completes without failed setup units or hooks.
Flathub and the declared default Flatpaks are installed or have a visible retryable failure state.
Input-remapper and Bluefin statistics services have intentional enablement policy.
Hooks never invoke a missing Tailscale binary; the deferred Tailscale state is explicit to users and CI.
A repeat boot proves first-boot operations are idempotent.
Live acceptance pass on the utah-test reference VM (booted :testing, Oct 10):
Failed units: system shows only brew-update.service + brew-upgrade.service, both 217/USER — the inherited common units hardcode User=1000 ('Override the user if different UID/User') and this lab VM has no UID 1000. User units: zero failed. Likely lab-only (a real first-boot user is UID 1000), but worth a decision: ship a drop-in or leave to common.
Flatpaks:flatpak-preinstall.service exited SUCCESS (Oct 7, installed its declared set); idempotent — stays exited across reboots.
Fresh review after ChairLift PR #668 merged: post-merge build 38059234215 passed all four flavor images. Its main build log confirms systemctl --global enable brew-preinstall.service and successful desktop verification, including the complete inherited Control Center integration.
Runtime acceptance remains open. Post-Testing E2E run 38061400815 failed in ISO composition for every flavor: DNF tried file:///etc/utah-packages/repodata/repomd.xml, which is absent outside the build-only repository mount. No VM/first-login ChairLift check ran, and promote-to-testing was skipped. Do not treat #668's image wiring or older screenshots as proof of successful first-login GUI installation for this digest.
Summary: The code that makes Bluefin's first-boot services explicit and testable is already merged, but two of the checks in this issue — that a first boot finishes with no failed setup steps, and that a second boot shows those steps are safe to repeat — can only be answered by actually booting the image. That boot runs on a hosted virtual machine that lives outside this repository, so there is no code change left to make here until that boot runs.
What I need from you: Run an exact-SHA first-boot pass on the hosted reference VM (or lift the blocked label once one passes).
What is already done in this repository:
First-boot unit enablement is pinned — PR test: pin first-boot unit enablement policy #1122 (b55fdfb, "test: pin first-boot unit enablement policy") fails CI if any of the first-boot units are accidentally disabled: flatpak-preinstall.service, ublue-system-setup.service, bluefin-stats-refresh.timer, projectbluefin-countme.timer, uupd.timer, tailscaled.service, and input-remapper.service. The companion test_systemd_exit_tolerance.py test guards the "tolerate a missing thing" convention.
Default Flatpaks and the input-remapper contract are asserted in tests/test_desktop_contract.py.
The Flatpak preinstall hook and the Tailscale "missing binary" hook already have tests, so a hook never runs a missing Tailscale binary.
What still needs a boot (cannot be proved by a unit test):
First boot completes without failed setup units or hooks.
A repeat boot proves first-boot operations are idempotent.
The post-testing-e2e workflow only composes the ISO and checks LUKS unlock — it does not boot the installed system and read back first-boot service state. The review-VM path that does first-boot service execution, Flathub/default-Flatpak validation, and a graceful second boot lives on the hosted "utah-test reference VM," which is external to this repository. So nothing here can change until that VM pass runs.
Minor open item (flagged in the issue as likely lab-only): the inherited common units assume a UID-1000 user — ship a drop-in or leave it to common. Separate, low-priority policy choice; does not block runtime validation.
Moved from a silent blocked to needs-human. A contributor relay marked this blocked without naming what it waits on, so nothing could lift it, and the hive's own fix-up was held by its write-back gate (the issue predates the hive's write access). Removing blocked and adding needs-human keeps agents off it and puts it in front of a maintainer, which is what the hive would have done itself.
Waits on: a live first-boot and second-boot run of an exact image to prove the first-boot services finish with no failures and are safe to repeat. The code side is merged (see the 2026-11-09 comment). A relay can't run that boot. Maintainer question: who runs that check, or should it become a testsuite scenario so CI proves it?
Owner decision (2026-11-16, delegated): this check runs in CI, through Utah's own LUKS install harness. It is scoped to Dakota's service contract.
The 2026-11-09 relay comment says only a hosted VM can run this check. That's wrong. iso/scripts/luks-e2e.sh already installs to a LUKS disk, boots it, does a graceful second boot and checks setup units. It ran only under the opt-in review_vm path (UTAH_E2E_EXPECT_FLAVOR). Its last run, 43762243257 (2026-11-15, main flavor), passed on both boots: flatpak-preinstall, ublue-system-setup, brew-setup, input-remapper, bluefin-stats-refresh.timer, uupd.timer and ublue-user-setup all succeeded, and the default Flatpaks were still present after the repeat boot.
Plan
test(e2e): check every contract service on the booted system #1282 (this step): the e2e checks every unit in the installed image's contract after boot, from both [services].enabled and [services].user_enabled. Each must be loaded, still enabled after first-boot presets, and not failed. The list is read from /usr/share/utah/bluefin-desktop.toml on the booted system. The first-boot check runs on every Post-Testing E2E candidate, and the review VM adds the repeat boot. Validation: review-VM run 43803298248 on the PR head.
Remaining acceptance items, under the Dakota decisions:
Input-remapper and stats policy: input-remapper is replaced by kanata and ydotool (parity(dakota): replace input-remapper with kanata #1222). The contract's enabled set covers whichever units that change enables. bluefin-stats-refresh.timer and projectbluefin-countme.timer stay enabled as contract units.
Tailscale: Utah's Dakota-aligned image ships the tailscale binary (factory recipe in progress). Once tailscaled.service is in the contract it gets the same check. Until then the existing tests keep the hook from calling a missing binary.
UID-1000 brew units: the lab-only case from 2026-10-10. The installer-created user is UID 1000, which the e2e already exercises. No drop-in.
Close this issue once #1282 has merged and a Post-Testing E2E run after the Dakota stack shows both boots passing with the Dakota contract.
/hive decision Proceed with the 2026-11-16 owner plan in this thread: land #1282 so the LUKS e2e checks every contract unit after first and second boot, then close this issue once a Post-Testing E2E run after the Dakota stack (#1309) shows both boots passing. No hosted-VM run is needed.
Rationale: the remaining work is CI-provable, so it should not sit at needs-human.
Approved by @hanthor. Decision: Proceed with the 2026-11-16 owner plan in this thread: land #1282 so the LUKS e2e checks every contract unit after first and second boot, then close this issue once a Post-Testing E2E run after the Dakota stack (#1309) shows both boots passing. No hosted-VM run is needed.
approved-direction is not applied while needs-human remains. Left in place: needs-human (added by @Danathar). Any hold label is left exactly as it was.
Removed needs-human. I added it on 11-12, and the owner has since approved the plan.
Evidence:@hanthor approved the plan with /hive decision at 11:00 PM EST on 11-17 (#19 (comment)): land #1282, then close this issue once a Post-Testing E2E run after #1309 shows both boots passing. The hive recorded the approval (#19 (comment)), but approved-direction stays withheld while needs-human is present. Nothing here still needs a person to decide.
The PR body says this is partial/non-closing work (Refs/Part of or leave-open wording), so Hive will not add a closing keyword. GitHub will not show its issue-list PR icon without a closing keyword; the stage comment is the link.
Activity
chore: document buildroot-toolchain parity debt
feat(desktop): validate Bluefin first-boot services and hooks at runtime
fix(setup): make tailscale rollback jq-only and run the repeat-boot c…
feat(desktop): validate Bluefin first-boot services and hooks at runtime
fix(setup): make tailscale rollback jq-only and run the repeat-boot c…
fix(hooks): validate Bluefin first-boot services and hooks at runtime (…
fix(services): enable bluefin-stats-refresh.timer at first boot
fix(services): enable bluefin-stats-refresh.timer at first boot
fix(services): enable bluefin-stats-refresh.timer at first boot
hanthor commentedon Oct 10, 2026
Live acceptance pass on the utah-test reference VM (booted :testing, Oct 10):
brew-update.service+brew-upgrade.service, both 217/USER — the inherited common units hardcodeUser=1000('Override the user if different UID/User') and this lab VM has no UID 1000. User units: zero failed. Likely lab-only (a real first-boot user is UID 1000), but worth a decision: ship a drop-in or leave to common.flatpak-preinstall.serviceexited SUCCESS (Oct 7, installed its declared set); idempotent — stays exited across reboots.ublue-system-setup.serviceenabled;ublue-update.timernot-found (update policy lives elsewhere — uupd).Not closing: the UID-1000 assumption and the input-remapper/statistics enablement policy still want explicit decisions.
hanthor commentedon Oct 10, 2026
Fresh review after ChairLift PR #668 merged: post-merge build 38059234215 passed all four flavor images. Its main build log confirms
systemctl --global enable brew-preinstall.serviceand successful desktop verification, including the complete inherited Control Center integration.Runtime acceptance remains open. Post-Testing E2E run 38061400815 failed in ISO composition for every flavor: DNF tried
file:///etc/utah-packages/repodata/repomd.xml, which is absent outside the build-only repository mount. No VM/first-login ChairLift check ran, andpromote-to-testingwas skipped. Do not treat #668's image wiring or older screenshots as proof of successful first-login GUI installation for this digest.Build: https://github.com/projectbluefin/utah/actions/runs/38059234215
Failed E2E: https://github.com/projectbluefin/utah/actions/runs/38061400815
37 remaining items
kylerankin commentedon Nov 9, 2026
Summary: The code that makes Bluefin's first-boot services explicit and testable is already merged, but two of the checks in this issue — that a first boot finishes with no failed setup steps, and that a second boot shows those steps are safe to repeat — can only be answered by actually booting the image. That boot runs on a hosted virtual machine that lives outside this repository, so there is no code change left to make here until that boot runs.
What I need from you: Run an exact-SHA first-boot pass on the hosted reference VM (or lift the blocked label once one passes).
What is already done in this repository:
b55fdfb, "test: pin first-boot unit enablement policy") fails CI if any of the first-boot units are accidentally disabled:flatpak-preinstall.service,ublue-system-setup.service,bluefin-stats-refresh.timer,projectbluefin-countme.timer,uupd.timer,tailscaled.service, andinput-remapper.service. The companiontest_systemd_exit_tolerance.pytest guards the "tolerate a missing thing" convention.tests/test_desktop_contract.py.What still needs a boot (cannot be proved by a unit test):
The
post-testing-e2eworkflow only composes the ISO and checks LUKS unlock — it does not boot the installed system and read back first-boot service state. The review-VM path that does first-boot service execution, Flathub/default-Flatpak validation, and a graceful second boot lives on the hosted "utah-test reference VM," which is external to this repository. So nothing here can change until that VM pass runs.Minor open item (flagged in the issue as likely lab-only): the inherited common units assume a UID-1000 user — ship a drop-in or leave it to common. Separate, low-priority policy choice; does not block runtime validation.
— hive: backend=pi model=lemonade/ornith-1.5-35b-a3b-gguf-q6_k
🐝 Hive Agent:
contributor| SHA:aa24e6fDanathar commentedon Nov 12, 2026
Moved from a silent
blockedtoneeds-human. A contributor relay marked thisblockedwithout naming what it waits on, so nothing could lift it, and the hive's own fix-up was held by its write-back gate (the issue predates the hive's write access). Removingblockedand addingneeds-humankeeps agents off it and puts it in front of a maintainer, which is what the hive would have done itself.Waits on: a live first-boot and second-boot run of an exact image to prove the first-boot services finish with no failures and are safe to repeat. The code side is merged (see the 2026-11-09 comment). A relay can't run that boot.
Maintainer question: who runs that check, or should it become a testsuite scenario so CI proves it?
hanthor commentedon Nov 16, 2026
Owner decision (2026-11-16, delegated): this check runs in CI, through Utah's own LUKS install harness. It is scoped to Dakota's service contract.
The 2026-11-09 relay comment says only a hosted VM can run this check. That's wrong.
iso/scripts/luks-e2e.shalready installs to a LUKS disk, boots it, does a graceful second boot and checks setup units. It ran only under the opt-inreview_vmpath (UTAH_E2E_EXPECT_FLAVOR). Its last run, 43762243257 (2026-11-15, main flavor), passed on both boots:flatpak-preinstall,ublue-system-setup,brew-setup,input-remapper,bluefin-stats-refresh.timer,uupd.timerandublue-user-setupall succeeded, and the default Flatpaks were still present after the repeat boot.Plan
[services].enabledand[services].user_enabled. Each must be loaded, still enabled after first-boot presets, and not failed. The list is read from/usr/share/utah/bluefin-desktop.tomlon the booted system. The first-boot check runs on every Post-Testing E2E candidate, and the review VM adds the repeat boot. Validation: review-VM run 43803298248 on the PR head.docs/skills/desktop-contract/references/dakota-services.md: rtkit, upower, udisks2, cups, pcscd, gnome-remote-desktop, malcontent, oomd, podman.socket, the systemd sockets, timesyncd and the rest. Because test(e2e): check every contract service on the booted system #1282 reads the installed contract, those units are checked on a booted system as soon as the stack lands, with no second list to maintain. If a Dakota unit legitimately can't start in an offline QEMU guest, add it to the e2e's offline exemption with a reason. Don't remove it from the contract.bluefin-stats-refresh.timerandprojectbluefin-countme.timerstay enabled as contract units.tailscalebinary (factory recipe in progress). Oncetailscaled.serviceis in the contract it gets the same check. Until then the existing tests keep the hook from calling a missing binary.Close this issue once #1282 has merged and a Post-Testing E2E run after the Dakota stack shows both boots passing with the Dakota contract.
test(e2e): check every contract service on the booted system
hanthor commentedon Nov 18, 2026
/hive decision Proceed with the 2026-11-16 owner plan in this thread: land #1282 so the LUKS e2e checks every contract unit after first and second boot, then close this issue once a Post-Testing E2E run after the Dakota stack (#1309) shows both boots passing. No hosted-VM run is needed.
Rationale: the remaining work is CI-provable, so it should not sit at needs-human.
Triaged on behalf of @hanthor with Claude Code
hivecommons-hive commentedon Nov 18, 2026
Approved by @hanthor. Decision: Proceed with the 2026-11-16 owner plan in this thread: land #1282 so the LUKS e2e checks every contract unit after first and second boot, then close this issue once a Post-Testing E2E run after the Dakota stack (#1309) shows both boots passing. No hosted-VM run is needed.
approved-directionis not applied whileneeds-humanremains. Left in place:needs-human(added by @Danathar). Anyholdlabel is left exactly as it was.Danathar commentedon Nov 18, 2026
Removed
needs-human. I added it on 11-12, and the owner has since approved the plan.Evidence: @hanthor approved the plan with
/hive decisionat 11:00 PM EST on 11-17 (#19 (comment)): land #1282, then close this issue once a Post-Testing E2E run after #1309 shows both boots passing. The hive recorded the approval (#19 (comment)), butapproved-directionstays withheld whileneeds-humanis present. Nothing here still needs a person to decide.hivecommons-hive commentedon Nov 18, 2026
Hive status — updated in place; edits do not notify. Full history is in the Activity Monitor.
PR #1282 covers this issue: #1282
The PR body says this is partial/non-closing work (
Refs/Part ofor leave-open wording), so Hive will not add a closing keyword. GitHub will not show its issue-list PR icon without a closing keyword; the stage comment is the link.