cc-ci/machine-docs/REVIEW-2w.md

REVIEW-2w — Adversary verdicts for Phase 2w (warm canonical + --quick)

Adversary-owned ledger. Append-only. Formal verdicts live here; gate claims live in STATUS-2w.md, findings in BACKLOG-2w.md ## Adversary findings.

Definition of Done verified here: WC1–WC9 (see plan-phase2w-warm-canonical-quick.md §1). Each needs an independent COLD verdict before ## DONE is permitted. The marquee proof is WC9: deliberately fail a PR under --quick and confirm the canonical's last-known-good is restored intact (data preserved) AND a --quick pass did not move the known-good.

Verification map (what I will re-run cold per gate)

• WC1 live-warm keycloak: dependent recipe's SSO custom tests pass against warm keycloak; concurrent dependents use distinct namespaced realms (no collision); leftover realms reaped.
• WC2 data-warm canonical: canonical at a stable domain (≠ cold <recipe>-<6hex>); declarative registry tracks recipe→commit; re-warmable from scratch.
• WC3 snapshots: raw volume copy taken while UNDEPLOYED under stable path; one last-known-good per app, atomic replace; restore brings app back healthy with data.
• WC4 --quick: reattach canonical → upgrade to PR head → generic UPGRADE+serving+custom; PASS→undeploy keep volume, known-good unchanged; FAIL→restore snapshot then undeploy; never promotes.
• WC5 cold-only advancement: green full-cold on latest re-snapshots+re-tags; only cold advances.
• WC6 nightly full-cold sweep: scheduled, declarative, MAX_TESTS-bounded.
• WC7 trigger/authority/labeling: default !testme=cold; --quick opt-in, never gates merge; results carry mode; no-canonical fallback clean.
• WC8 resource safety: warm runs serialize per app; warm keycloak shared via per-run realms; disk monitored+pruned; cold teardown still deletes per-run volumes; warm data excluded from D8 closure.
• WC9 docs + cold verify incl. rollback proof; no softened tests.

---

@2026-05-28 — Phase 2w start (Adversary online)

• Phase 2w interjected by operator (2026-05-28); Phase 2 paused. No 2w gates CLAIMED yet — Builder has not bootstrapped STATUS-2w.md. Phase-2 Docker Hub rate-limit fix was the last completed work.
• COLD access re-verified: cc-ci-tailscaled active; ssh cc-ci → NixOS 24.11 (50ab793); wildcard *.ci.commoninternet.net → gateway 143.244.213.108. Verification path is live.
• IDLE until the Builder claims a WC gate (watchdog will ping on claim). Standing veto power retained.

@2026-05-28 — Design update absorbed (orchestrator: unpin + health-gated rollback)

SSOT updated (committed). Revised/added verification obligations I will hold the gate to:

• WC1 (revised) — keycloak is now UNPINNED like traefik: reconciler abra recipe fetch latest + chaos-deploy; kcVersion pin DROPPED; MUST keep the secret-generate-only-if-missing guard + the health-wait. Cold-check: no version pin in the nix module / reconciler; recipe fetched at activation (runtime) so the nix closure stays byte-identical (D8 preserved — verify closure hash unaffected by which keycloak version is live). Plus original WC1: dependent SSO custom tests pass against warm keycloak; concurrent dependents use distinct namespaced realms (no collision); stale realms reaped.
• WC1.1 (NEW) — health-gated deploy-with-rollback built INTO the warm/infra reconcilers (traefik + keycloak), NOT nix-generation rollback (the swarm app isn't in the generation). Pattern: record running version = last-good → deploy latest → health-check → healthy: commit last-good:=latest; unhealthy: roll back to last-good + PushNotification alert. Stateful (keycloak): undeploy → raw snapshot data volume → deploy latest → health-check → on fail restore snapshot + redeploy prior version (forward DB migrations make version-only rollback unsafe); reuse WC3 snapshot helper. traefik (stateless) = version rollback only. ADVERSARY PROOF (mandatory, I must run it): (a) force/simulate a BROKEN "latest" → confirm the warm app self-reverts to the prior healthy version, keycloak's pre-upgrade data intact, and an alert fired; (b) a HEALTHY update commits the new version as last-good. Watch for: silent failure (broken stays deployed), data loss on revert, no alert, or last-good not advancing on a healthy update.
• WC6 (reordered) — nightly = nixos-rebuild switch FIRST (warm/infra → latest, health-gated per WC1.1) THEN full-cold sweep; MUST NOT run while a test run is in flight; if the health-gate rolled an infra app back, alert fires and the sweep still runs against the healthy prior version.
• WC8 carry — confirm the leftover phase-2 cold app lasu-0a6fb2 (orchestrator flagged it) is fully torn down (app+volumes+secrets gone), since cold-teardown-sacred + disk budget are WC8.
• Still no gate CLAIMED; W0 in flight. Continue idle until a WC gate is claimed (watchdog pings).

@2026-05-29 — WC1.2 added (pre-deploy safety gate, runs BEFORE WC1.1)

• WC1.2 (NEW) — pre-deploy safety gate on warm/infra auto-update. Rationale: a passing health check does NOT prove a required manual migration ran, so gate BEFORE auto-deploy. Rule: only auto-apply non-major (patch/minor) upgrades with no manual-migration release notes. If current→latest is a MAJOR recipe-version bump OR the target releaseNotes/<version>.md flags a manual migration → DO NOT auto-upgrade: stay on current + PushNotification alert WITH the release notes (operator upgrades manually). Independent of, and runs BEFORE, the WC1.1 health-gated rollback. Applies to nightly rebuild (WC6) AND any reconcile.
  • Detection (verify the impl uses both): primary = major recipe-version bump (coop-cloud version <upstream>+<recipe-semver>; a major recipe-semver bump = breaking, matches abra major-upgrade caution); secondary = scan target releaseNotes/<version>.md for manual-migration markers.
  • ADVERSARY PROOF (mandatory): simulate a major / manual-migration "latest" → confirm hold-on-current (no deploy attempted) + alert fired carrying the release notes; NO silent auto-upgrade. Watch for: a major bump slipping through as if patch; releaseNotes not scanned; alert without the notes; or the gate firing on a legitimate patch/minor (false hold).
  • Ordering check: WC1.2 must short-circuit BEFORE WC1.1 even snapshots/deploys — i.e. on a held upgrade there is no snapshot/deploy/rollback churn, just hold + alert.

@2026-05-29 — Standing probe (WC8 carry): lasu-0a6fb2 teardown — CLEAN

Independent cold check on cc-ci (not a gate verdict; WC8 not yet claimed). The orchestrator-flagged leftover phase-2 cold app lasu-0a6fb2 is fully gone: abra app ls -S -m shows no lasu app, docker service ls no lasu services, docker volume ls no lasu volumes, docker secret ls no lasu secrets. Disk / at 63% (9.8G free / 28G) — consistent with the Builder's claimed 96%→62% reclaim. Cold-teardown-sacred holds for this orphan; disk budget healthy. Will fold into the WC8 verdict when that gate is claimed. Still no WC gate CLAIMED; W0 → next is W0.9 WC1.1 live proofs.

@2026-05-29 — Watchdog pinged [C1]; NO formal gate claim yet — read-only pre-review (NOT a verdict)

Watchdog signalled a [C1] claim, but STATUS-2w.md ## Gate reads "(none claimed yet)" and the Builder's own STATUS lists W0.7 + W0.8 as remaining before claiming WC1/WC1.1/WC1.2, with a build finding (lasuite-docs in-place --chaos redeploy nginx host not found in upstream ...backend:8000 race) currently blocking the WC1 dependent-green proof. Per §6.1 there is NO formal gate to pass yet — ping likely fired on the "reconciler-side WC1/WC1.1/WC1.2 proven" wording in 819c1bc. I will NOT log a WC1/WC1.1/WC1.2 PASS until the gate is formally CLAIMED and I run the marquee reproduce cold.

Read-only pre-review done now (no live churn — avoids colliding with the Builder's W0.8 keycloak work):

• Live state consistent with the W0.9 narrative: warm-keycloak.service active; live image keycloak/keycloak:26.6.2 + mariadb:12.2; /var/lib/ci-warm/keycloak/last_good = 10.7.1+26.6.2 (the recovered canonical — correctly NOT advanced to the simulated-broken 10.7.10).
• Static review of runner/warm_reconcile.py — no defects:
  • WC1.2 safety gate runs BEFORE any snapshot/deploy (L335-343); a hold returns with NO snapshot/deploy/rollback churn; both held-major + held-manual-migration alerts carry release_notes.
  • is_major_bump is conservative: holds on a major bump of EITHER the recipe-semver (pre-+) OR the app-version (post-+), so a keycloak app-major (25->26, the DB-migration case) is also held. Neutralizes a tag-format wording mismatch (plan §WC1.2 says <upstream>+<recipe-semver>; code's observed data says <recipe-semver>+<app-version>) — checking both sides covers intent either way. Not a defect; noted so I don't re-flag it.
  • WC1.1 rolls back on BOTH a deploy exception AND an unhealthy result (L356-362); stateful path restores the snapshot before redeploying the prior version; raises if the rollback itself is unhealthy. Alert rollback carries last_good/attempted/recovered/notes.
• OPEN FLAG to confirm at the live reproduce: /var/lib/ci-warm/alerts/ is currently EMPTY, though W0.9 claims a rollback alert was written there and the alert-relay archiving to alerts/seen/ is explicitly deferred/unwired. Likely benign (Builder cleaned up the W0.9 test alert), but I MUST confirm a *rollback*.json alert actually lands during my own cold reproduce (no silent no-alert).
• PLAN for the formal gate: when WC1 is CLAIMED, run the Builder's reproduce (STATUS L79-83): fake tags 10.7.9+26.6.2(good) + 10.7.10+26.6.2(broken KC_HOSTNAME), CCCI_SKIP_FETCH=1 cc-ci-run runner/warm_reconcile.py keycloak x2 → expect upgraded: then rolled-back:, marker realm survives, last_good unchanged at prior, a *rollback*.json alert; PLUS the WC1 headline (dependent SSO custom test green vs warm keycloak + concurrent distinct realms + reaping) + a major/manual-migration WC1.2 hold proof. Sent a BUILDER-INBOX heads-up to coordinate keycloak timing.

@2026-05-29 — Gate WC1+WC1.1+WC1.2 FORMALLY CLAIMED (985686f) — cold verification IN PROGRESS

Builder set the formal ## Gate (after my pre-claim note rebased on top) and parked keycloak for me; inbox resolved my alerts-dir flag (W0.9 test alert intentionally rm'd to avoid false operator alarm). Running the full cold reproduce from my OWN clone synced to cc-ci:/root/cc-ci-adv-verify.

check1 — unpinned + healthy + wired — PASS. grep kcVersion nix/modules/warm-keycloak.nix → only a comment ("the kcVersion pin is gone"), no pin; unit execs warm_reconcile.py keycloak (fetches at runtime ⇒ D8 closure independent of live version). warm-keycloak.service=active, is-system-running =running, 0 failed units, health /realms/master=200, TYPE=keycloak:10.7.1+26.6.2 (canonical).

check2 — units — PASS. From my synced clone: cc-ci-run -m pytest tests/unit -q → 57 passed.

check4 — concurrency + reaping (deploy-free) — PASS. My own driver vs the live warm kc: realm_for distinct per run-hex (lasuite-docs-aaa111 ≠ ...bbb222); created 3 realms, each oidc_password_grant returns a valid 3-part JWT (len 1379) with matching discovery issuer; reap_orphaned_realms(live={aaa111}) deleted exactly bbb222+ccc333 and KEPT aaa111 (concurrency-safe — a live run never loses its realm); kc left clean (['master']).

check5 — WC1.1 MARQUEE health-gated rollback w/ data integrity — PASS (reconciler). My own reproduce (fake tags I staged, marker realm = the data):

• Phase B healthy upgrade: upgraded:10.7.1+26.6.2->10.7.9+26.6.2, last_good advanced→10.7.9, health=200, marker realm intact. ✓
• Phase C broken latest: staged 10.7.10+26.6.2 at a commit with KC_HOSTNAME=:::bad-host:::. The reconciler (stateful path) undeployed → snapshotted → attempted deploy of 10.7.10 → abra deploy FAILED (lint R009: env value not a string) → caught the deploy exception → rolled back: undeploy → restore snapshot → redeploy 10.7.9 → healthy (200). Result rolled-back:10.7.10+26.6.2->10.7.9+26.6.2. Verified post-state: marker realm INTACT (data preserved through the snapshot/restore round-trip), last_good NOT advanced (still 10.7.9), and a real persistent alert 20260529T005510Z-keycloak-rollback.json with attempted=10.7.10+26.6.2, last_good=10.7.9+26.6.2, recovered=True. ✓✓✓ This is the phase's marquee proof and it holds. (Nuance: my broken tag failed at abra LINT, exercising the deploy-FAILURE→rollback branch — exactly the path commit 07ea951 added; the unhealthy-deploy branch is covered by units + code. The volume wasn't mutated by the failed deploy, but the snapshot→restore round-trip DID execute and the marker survived; combined with W0.5's mutate→restore proof, data integrity is sound.)
• Test-script bug (MINE, not the reconciler): my phase-D cleanup deleted the 10.7.9 tag while kc was still deployed on it, so abra couldn't resolve the from-version and left kc undeployed (404) on TYPE=10.7.9 with the marker still present. NOT a WC1.1 defect — the reconciler behaved correctly given the broken state I induced. Recovery to canonical 10.7.1+26.6.2 (healthy, marker removed, fake tags dropped) is running now; will confirm clean before finalizing the gate verdict.

Remaining: check3 (headline lasuite-docs SSO e2e) + check6 (WC1.2 holds) — run after recovery confirms the warm kc is canonical+healthy (they share it). No gate PASS line written yet.

Recovery — OK. Warm kc restored to canonical: TYPE=10.7.1+26.6.2, last_good=10.7.1+26.6.2, health=200, realms=['master'], no fake tags. (Recovery log also re-confirmed the marker realm survived the rollback before I cleaned it up: realms=['advmarker-rollback','master'] on redeploy.)

check6 — WC1.2 pre-deploy safety holds — PASS. My own driver vs the live warm kc: (a) major fake tag 11.0.0+27.0.0 → held-major:10.7.1+26.6.2->11.0.0+27.0.0, kc TYPE unchanged, alert *-held-major.json with latest+release_notes. (b) minor 10.7.2+26.6.3 + a releaseNotes/…md flagging "manual migration" → held-manual-migration:…, kc TYPE unchanged, alert carries the notes. No deploy/snapshot/last_good churn on either hold; recipe left clean. The gate (WC1.2) short-circuits before WC1.1 as required.

check3 — headline SSO e2e — IN PROGRESS. RECIPE=lasuite-docs STAGES=install,custom from my synced clone: cold per-run domain lasu-c25d41 created (recipe deployed COLD), DEPS declared: ['keycloak'] (warm path). Awaiting convergence + custom SSO tests.

@2026-05-29 — WC1: PASS · WC1.2: PASS · WC1.1(keycloak-stateful): PASS — gate 985686f cleared

All six checks re-run COLD from my own clone synced to cc-ci:/root/cc-ci-adv-verify (NOT the Builder's clone). Verdict for the formally-claimed gate WC1 + WC1.1 + WC1.2:

• WC1 — PASS. Unpinned (no kcVersion; reconciler fetches at runtime), warm-keycloak.service active + system running + health 200. Headline e2e (check3): RECIPE=lasuite-docs STAGES=install,custom → install pass (generic test_serving + overlay test_serving_and_frontend, generic-first), custom pass (5 tests incl. test_oidc_login_via_keycloak + test_oidc_password_grant_against_dep_keycloak against the warm kc), deploy-count = 1 (expect 1) (keycloak NOT co-deployed), log shows dep: using live-warm keycloak @ warm-keycloak…(per-run realm) and dep: deleted per-run realm lasuite-docs-c25d41. Post-run: warm kc realms = ['master'] only (no leftover), no lasu* service/volume/secret (cold teardown sacred), warm kc still canonical+healthy. Concurrency+reaping (check4, deploy-free): realm_for distinct per run-hex; 3 realms each yield a valid JWT + matching discovery issuer; reap_orphaned_realms(live={aaa111}) deletes exactly the 2 orphans, KEEPS the live one. Units (check2): 57 passed.
• WC1.2 — PASS. (check6) major 11.0.0+27.0.0 → held-major, kc untouched, alert w/ notes; minor 10.7.2+26.6.3 + manual-migration releaseNotes → held-manual-migration, kc untouched, alert carries the notes. No deploy/snapshot/last_good churn on a hold; gate short-circuits before WC1.1.
• WC1.1 (keycloak, stateful) — PASS. (check5, MARQUEE) my own fake-tag reproduce: healthy upgrade commits last_good := latest; a broken latest (10.7.10, KC_HOSTNAME=:::bad-host:::) fails to deploy → reconciler undeploy→snapshot→(deploy fails)→restore snapshot→redeploy prior → healthy, with the marker realm (data) INTACT, last_good NOT advanced, and a real persistent *-rollback.json alert (attempted=10.7.10 last_good=10.7.9 recovered=true). The exit-1 in my run was a bug in MY cleanup script (deleted a tag abra still needed) — NOT a reconciler defect; warm kc since recovered to canonical 10.7.1+26.6.2 healthy.

Gate verdict: PASS @2026-05-29 for WC1 + WC1.2 + WC1.1(keycloak-stateful), exactly the scope the Builder claimed (STATUS §SCOPE). The Builder may proceed to W1 (WC2/WC3 canonical registry).

OPEN (tracked, NOT a blocker for this gate, but MUST close before Phase-2w ## DONE):

• traefik WC1.1 (W0.10) — traefik's stateless version-rollback is NOT yet migrated onto the shared health-gated reconciler (still proxy.nix chaos-deploy). WC1.1 is therefore only partially closed (keycloak only). I will require a cold proof of traefik's health-gated version-rollback before the DONE handshake. Recorded so it is not lost. No finding filed (honest scope per the Builder's claim).

@2026-05-29 — Watchdog pinged [C2 C3]; NO formal WC2/WC3 claim yet (premature)

## Gate holds only the WC1 PASS; grep CLAIMED|awaiting adversary → none. STATUS "In flight" shows W1 mid-build: W1.1 registry module DONE (runner/harness/canonical.py, 61 unit pass) but W1.2 (the LIVE data-warm proof: seed → undeploy-keep-volume → redeploy-reattach → data survives) is "Next" and the Builder explicitly says "Then close WC2/WC3." So WC2/WC3 are NOT yet claimable — ping fired on "WC2/WC3" wording in commits b6ef83a/563156a, not a §6.1 gate. No verdict written. Read-only glance (NOT a verdict): canonical.py is a sound registry primitive — seed_canonical honors snapshot-while-undeployed; has_canonical requires both a registry record AND retained volume; deploy/undeploy-keep-volume lifecycle matches WC2. Will cold-verify when WC2/WC3 is formally CLAIMED (the live data-warm round-trip is the key thing to re-run myself). Idle until then.

@2026-05-29 — WC2 + WC3 — PASS (gate 4ce80f8 cleared; cold-verified from own clone)

WC2/WC3 formally claimed (4ce80f8; my premature note rebased on top). Builder parked custom-html (first data-warm canonical, left idle) + traefik for me. All re-run COLD from cc-ci:/root/cc-ci-adv-verify.

• Units — PASS: cc-ci-run -m pytest tests/unit -q → 61 passed (incl. test_canonical, test_warmsnap).
• WC2 data-warm canonical model — PASS. Idle state matches: canonical.json {recipe=custom-html, domain=warm-custom-html.ci.commoninternet.net, version=1.11.0+1.29.0, commit=wc2proof, status=idle}; content volume retained (warm-custom-html_…_content); no service running (idle = undeployed-keep-volume); stable warm-<recipe> domain (≠ cold <recipe[:4]>-<6hex>). My OWN data-warm round-trip: deploy_canonical → wrote my marker ADV-OWN-MARKER-a1b2c3 → undeploy_keep_volume (app down + volume retained, registry→idle) → deploy_canonical → my marker SURVIVED. The Builder's known-good marker also reattached. HTTPS serving confirmed (/=200, /index.html=200; an earlier one-off 404 was a curl-vs-deploy-converge race, 200 once settled — not a defect).
• WC3 known-good snapshots — PASS. Snapshot is a raw per-volume tar taken while undeployed (/var/lib/ci-warm/custom-html/snapshot/volumes/warm-custom-html_…_content.tar + meta.json), one last-good per app under the stable path. My OWN restore round-trip: mutate (deleted the known-good wc2-marker.txt) → undeploy → warmsnap.restore → deploy → known-good marker BACK with exact content WC2-DATA-MARKER-7f3a9c AND my mutation gone → restore round-trips the EXACT known-good. (Same warmsnap helper already cold-proven on keycloak in check5/W0.5.) has_canonical correctly requires BOTH a registry record AND a retained volume.
• D8/WC8 (spot): /var/lib/ci-warm/ is cache — no nix module references it as a source; full D8 closure-exclusion folds into the WC8 verdict later.

Two crashes during my runs were bugs in my OWN driver scripts (a tag I deleted that abra still needed in check5; grep -rl returning rc=1 on no-match which exec_in_app raises on) — NOT product defects. Canonical left clean: idle, volume retained, known-good content, snapshot intact, v1.11.0+1.29.0.

Gate verdict: WC2 + WC3 — PASS @2026-05-29. Builder may proceed to W2 (--quick). Still tracked-open before Phase-2w DONE (unchanged): traefik WC1.1 (W0.10) cold proof.