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
Replace the current retag-only post-testing workflow with a Dakota/Bluefin-style gate that boots and tests the exact image digest before any moving testing tag advances. Secondary flavors need flavor-specific kernel and NVIDIA assertions.
Acceptance criteria
The main image digest passes smoke and common VM suites before promotion.
All promoted digests come from the same successful build and are promoted atomically.
Gaming and NVIDIA flavors validate their expected kernel, modules, and userspace before promotion.
A failed test leaves existing testing tags unchanged and publishes diagnostics.
The branch promotion workflow cannot bypass the digest gate.
Reviewed both open proposals against their current heads. #161 currently contains only incorrect README citations and no gate implementation; changes requested. #631 adds non-blocking reporting, not promotion gating, and needs immutable testsuite pins; changes requested. Existing LUKS/production-ISO checks do not satisfy every smoke/common, flavor-specific and atomic-promotion criterion in this issue. Keeping the acceptance gap open.
Maintainer decision: #631 is merged as a non-blocking exact-digest QA-reporting improvement only. It does not gate :testing, and the NVIDIA suite still has no runnable hardware assertions. The full acceptance criteria remain open; the hive/covered-by-pr label was removed so this is not mistaken for complete.
Implemented the currently reachable gate in PR #722: exact-digest LUKS/production-ISO/per-flavor testsuite now gate :testing promotion; the status is bound to the source SHA and the branch promoter only runs from trusted main. Local just check passes (1081 tests).
The issue remains open: GHCR tag copies across flavor packages are still sequential, not atomic, and this PR does not satisfy the all-or-nothing multi-tag criterion. Hive/CI review on #722 is pending.
I checked this issue once against main (1a76ea7). It's still open.
#627 added a read-only, opt-in offline ISO test matrix. It doesn't gate testing-tag promotion on exact-digest VM E2E, so the acceptance boxes are still unmet. The work left (promotion gated on exact-digest suites, atomic promotion, and no bypass through branch promotion) is already in progress in #722 by @hanthor. Agents are leaving it alone so there aren't two competing implementations.
This issue asks for a CI "gate" that boots and tests the exact image that will ship before any moving label (:testing, the pointer to the latest tested image) is allowed to advance. Most of the gate already exists and is being finished by the maintainer in open PR #722. The one requirement it does not yet meet — promoting every flavor's tags together as a single all-or-nothing step — is a design choice only the maintainer can make, so no separate contribution should start now.
What is already done or in progress
Open PR ci: gate testing promotion on exact-digest suites #722 ("ci: gate testing promotion on exact-digest suites") turns the per-flavor testsuite into a real promotion gate, adds per-flavor kernel, module and userspace assertions, and stops branch promotion from bypassing the gate by binding the result to the source commit and requiring it before promotion.
On main, the LUKS and production ISO exact-digest VM tests already run, and a superseded build already leaves the tags unchanged.
What remains
The requirement "all promoted digests are promoted atomically" is not met. The image store (GHCR) copies tags one at a time, and the repo's own CI docs already call this a known limitation. Deciding what "atomic" means here — or whether a sequential copy with rollback is acceptable — is a maintainer decision, not new engineering an outside contributor can ship without stepping on ci: gate testing promotion on exact-digest suites #722.
The maintainer's own comment already records that #722 covers the reachable gate and that the atomic box stays open. Agents are asked not to open a competing implementation while #722 is in progress, so this issue is held for the maintainer to close that last criterion.
The park reason was written before these changes. Re-read it against the current state before deciding. This note does not un-park the issue: it stays with a maintainer, and Hive updates this note if more of the referenced work moves.
If the closing note quoted above settles the question, a maintainer can reply /fixed to close this issue as completed, or /hive decision <your instructions> to un-park it.
Resolved by #885 (exact-digest VM E2E promotion ratchet and release gate) and #905. :testing now moves only to digests that passed the LUKS matrix; it has done so 5 times since 11-06.
Activity
hanthor commentedon Oct 10, 2026
Reviewed both open proposals against their current heads. #161 currently contains only incorrect README citations and no gate implementation; changes requested. #631 adds non-blocking reporting, not promotion gating, and needs immutable testsuite pins; changes requested. Existing LUKS/production-ISO checks do not satisfy every smoke/common, flavor-specific and atomic-promotion criterion in this issue. Keeping the acceptance gap open.
Reviewed with Codex.
hanthor commentedon Oct 12, 2026
Maintainer decision: #631 is merged as a non-blocking exact-digest QA-reporting improvement only. It does not gate
:testing, and the NVIDIA suite still has no runnable hardware assertions. The full acceptance criteria remain open; thehive/covered-by-prlabel was removed so this is not mistaken for complete.hanthor commentedon Oct 12, 2026
Implemented the currently reachable gate in PR #722: exact-digest LUKS/production-ISO/per-flavor testsuite now gate :testing promotion; the status is bound to the source SHA and the branch promoter only runs from trusted main. Local
just checkpasses (1081 tests).The issue remains open: GHCR tag copies across flavor packages are still sequential, not atomic, and this PR does not satisfy the all-or-nothing multi-tag criterion. Hive/CI review on #722 is pending.
hivecommons-hive commentedon Oct 14, 2026
I checked this issue once against
main(1a76ea7). It's still open.#627 added a read-only, opt-in offline ISO test matrix. It doesn't gate testing-tag promotion on exact-digest VM E2E, so the acceptance boxes are still unmet. The work left (promotion gated on exact-digest suites, atomic promotion, and no bypass through branch promotion) is already in progress in #722 by @hanthor. Agents are leaving it alone so there aren't two competing implementations.
🐝 Hive Agent:
scanner| Instance:hosted-projectbluefin-knuckle-gjvq| SHA:unknown— hive: agent=scanner backend=copilot model=claude-opus-5.5 copilot=1.0.88
kylerankin commentedon Oct 15, 2026
Summary
This issue asks for a CI "gate" that boots and tests the exact image that will ship before any moving label (
:testing, the pointer to the latest tested image) is allowed to advance. Most of the gate already exists and is being finished by the maintainer in open PR #722. The one requirement it does not yet meet — promoting every flavor's tags together as a single all-or-nothing step — is a design choice only the maintainer can make, so no separate contribution should start now.What is already done or in progress
main, the LUKS and production ISO exact-digest VM tests already run, and a superseded build already leaves the tags unchanged.What remains
The maintainer's own comment already records that #722 covers the reachable gate and that the atomic box stays open. Agents are asked not to open a competing implementation while #722 is in progress, so this issue is held for the maintainer to close that last criterion.
— hive: backend=pi model=lemonade/ornith-1.5-35b-a3b-gguf-q6_k
🐝 Hive Agent:
contributor| SHA:aa24e6fhivecommons-hive commentedon Oct 26, 2026
The reason this issue is parked may be out of date. It has been parked for a maintainer (
stage/045-needs-decision) since 2026-10-15, and since then:run_e2eswitch would reintroduce the tip-equality dependency ci(promote): ratchet :testing over E2E-green commits and gate :stable on per-digest evidence #885 removed (see the review above). Still worth a fresh PR from main if wan …” (comment).The park reason was written before these changes. Re-read it against the current state before deciding. This note does not un-park the issue: it stays with a maintainer, and Hive updates this note if more of the referenced work moves.
If the closing note quoted above settles the question, a maintainer can reply
/fixedto close this issue as completed, or/hive decision <your instructions>to un-park it.🐛 fix(park-recheck): follow a closed PR to the hive's adopted copy an…
hanthor commentedon Nov 7, 2026
Resolved by #885 (exact-digest VM E2E promotion ratchet and release gate) and #905.
:testingnow moves only to digests that passed the LUKS matrix; it has done so 5 times since 11-06.