punimtag/ROADMAP.md
ilia c0c997f796
All checks were successful
CI / skip-ci-check (pull_request) Successful in 5s
CI / docker-ci (pull_request) Successful in 7s
CI / secret-scan (pull_request) Successful in 12s
CI / viewer-unit (pull_request) Successful in 1m29s
CI / e2e (pull_request) Successful in 4m8s
CI / admin-unit (pull_request) Successful in 4m24s
test: timing budgets + CI hardening (artifact v3 pin, npm cache retry)
Closes three items from the outstanding e2e/CI gap list:

- Timing budgets: timings.measure() only ever recorded durations for the
  (still-unwired) Pushgateway export — nothing failed CI when a step got
  slow. Add e2e/timing-budgets.ts (expectWithinBudget + shared BUDGET_MS
  buckets) and wire it into every measure() call site across the suite.
  Mail-wait steps are deliberately left unbudgeted (external mail-trap
  delivery latency, not a code performance signal).

- actions/upload-artifact@v4 doesn't work against this Gitea/act runner's
  artifact backend — pin to v3 for the e2e failure-report upload.

- Shared act_runner npm cache has corrupted platform-native tarballs before
  (@next/swc-linux-x64-musl) and reds viewer-unit/admin-unit/e2e with no
  product bug involved. All three npm ci steps now retry once after
  `npm cache clean --force` on first failure.

Verified: full local suite green against DEV (37 passed, 6 skipped, no
budget assertion failures) before wiring into CI.
2026-07-15 08:56:31 -04:00

9.4 KiB

PunimTag roadmap

Living plan for product quality, auth/email reliability, and automation.

Now — shipped / in flight

  • Fix public-host auth redirects (Kolby #57 family) — NEXTAUTH_URL + verify-email
  • DEV mail trap via Mailpit (not Mailtrap SaaS / not Spamhaus-blocked Mailcow for e2e)
  • Playwright e2e suite (e2e/) on @levkin/playkit against punimtagdev.levkin.ca
  • Dedicated e2e user e2e@levkine.ca (Vault / Infisical / Gitea Actions)
  • Vitest unit tests for viewer-frontend in Gitea Actions
  • API smoke + Zod schema checks (health, people, tags, 401 contracts)
  • storageState setup project + real upload smoke
  • FastAPI e2e userse2e (admin, mirrors e2e@levkine.ca) + e2e-viewer (viewer) in FastAPI main DB; E2E_API_USERNAME/PASSWORD + E2E_API_VIEWER_USERNAME/PASSWORD in CI
  • Search / filter e2e — tag + person filters on public gallery (gallery.search-filters.spec.ts)
  • Role permissions e2e — viewer vs admin write gates (api.role-permissions.spec.ts)
  • Ansible FastAPI-e2e work committed — ansible PR #127 merged (provisioning script, SMTP quote fix, vault/docs)
  • Infisical /playkit/punimtag E2E_API_ keys created* — self-hosted build's /api/v3/secrets/raw create route needs client-side E2EE ciphertext (422 on plaintext); fixed by switching to /api/v4/secrets (plaintext) in provision-punimtag-e2e-user.py — see ansible docs/hardening/SECRETS.md
  • Vaultwarden sync for FastAPI e2e loginsPunimTag FastAPI e2e admin/viewer (DEV) entries created
  • Playkit bumped to v0.3.1 — tagged, released (Gitea release job needed a RELEASE_TOKEN secret rename — GITEA_TOKEN is a reserved name Gitea's Actions API rejects), Outline QA & Dev → Playkit page created
  • Network silent-failure monitorstartNetworkErrorMonitor on gallery home, search+filter, upload, Manage Users (network-errors.spec.ts)
  • Favorites + People filter UI e2egallery.filters.authed.spec.ts (uses storageState)
  • admin-frontend Vitest — separate Vite/React app; admin-unit CI job + fastapi-path/media-path lib tests
  • interceptNetworkCall on upload + search filter UI — typed status/JSON spies replace waitForResponse/text scraping (upload.smoke.spec.ts, gallery.search-filters.spec.ts)
  • Widened Zod schemas — FastAPI login TokenResponse + /auth/me UserResponse (api.fastapi-login.spec.ts), gallery SearchResponse (gallery.search-filters.spec.ts)
  • PROD smoke specprod.smoke.spec.ts, opt-in via PROD_BASE_URL; skips (no-op) until the PROD LXC exists — see Ops/docs debt
  • NextAuth (viewer) write gates — third user store (auth DB, hasWriteAccess=false) provisioned via provision-punimtag-e2e-user.py; viewer.write-gates.spec.ts proves 403 (viewer) vs pass-through (admin) on POST /api/faces/{id}/identify
  • Manage Users real CRUD + causal cross-session e2eadmin.manage-users-actions.spec.ts: admin creates/edits/deletes a user through the actual panel (the older spec only opened/closed it), and deactivating a live user immediately revokes their already open session (jwt callback re-checks isActive); disposable e2e-manage-test-* accounts, always cleaned up
  • Timing budgetse2e/timing-budgets.ts expectWithinBudget() gates every timings.measure() call site against a shared BUDGET_MS bucket (api/uiAction/uiLogin/heavyCrud); a step getting order-of-magnitude slower now reds the test instead of only feeding an unwatched sample
  • CI: actions/upload-artifact@v4 pinned to v3 — v4 doesn't work against this Gitea/act runner's artifact backend; report upload on e2e failure was silently broken
  • CI: npm cache corruption retryviewer-unit/admin-unit/e2e all retry npm ci once after npm cache clean --force on first failure (shared act_runner cache has corrupted @next/swc-linux-x64-musl before, redding CI with no product bug)
  • ROADMAP (this file)

Next (near-term)

Ops / docs debt

  • Wire QA/PROD SMTP on live guests when LXCs 9102/9103 exist (make punimtag-sync-smtp ENV=qa|prod) — confirmed via pct list on 10.0.10.201: only DEV (9101) exists today, so this (and PROD smoke actually running, and the PROD NEXTAUTH_URL hostname) stays dormant-but-ready until QA/PROD LXCs are provisioned
  • Stop seeding admin@admin.com in docs as the day-to-day login; keep bootstrap scripts but point operators at Vaultwarden PunimTag e2e
  • Set the new E2E_VIEWER_EMAIL/E2E_VIEWER_PASSWORD Gitea secrets from Infisical/Vaultwarden if CI doesn't already have them synced (script sets Gitea directly, so likely already OK — verify on next CI run)
  • git-ci-01 still OOM-kills under multi-repo bursts even at 8 GB — bumped LXC 241 3→8 GB (ansible PR #128) after act_runner OOM-killed mid-job; that fix landed, but pve10 the hypervisor itself is overcommitted (~75 GB allocated across guests vs 62 GB physical RAM, host swap observed 7.9/8.0 GB used, ~8-9 GB free) — a burst of concurrent CI across ansible/punimtag/playkit/maCopy OOM-killed act_runner three times in 13 minutes on 2026-07-15 even with the larger allocation. Real fix is host-wide: audit over-allocated idle guests (e.g. automationlab 4 GB stopped, immich/paperless at 6 GB each) and either trim or move some guests to pve201. Out of scope for a CI-runner-only fix — flagging for a deliberate pass.
  • DEV LXC (9101) redeploypct exec 9101 shows viewer-frontend checked out at e54c609, 19 commits behind dev HEAD (missing e.g. c011c80 accessibility/aria-label pass). Two uncommitted local files from an earlier session (e2e/tests/admin.manage-users-actions.spec.ts, e2e/pages/ManageUsersPanel.ts — real Manage-Users CRUD + a causal cross-session deactivation check) time out because the deployed UI's row action buttons lack the aria-labels those specs target; every other spec here is unaffected since it doesn't depend on that commit. Redeploy DEV, then commit + verify those two files.

Playkit adoption (pin is v0.3.1 today)

Kit feature In pin? PunimTag use
Zod schema on ApiClient yes health, people/tags catalog, FastAPI login TokenResponse + /auth/me, gallery SearchResponse
saveStorageState / storageStateUse yes setup uses saveStorageState for admin and viewer (hasWriteAccess=false); specs use raw test.use({ storageState }) (storageStateUse optional cosmetic)
playkitFailureArtifacts() yes playwright.config.ts
Mailpit / createMailInbox yes mail specs
interceptNetworkCall / startNetworkErrorMonitor yes both in use — network-errors.spec.ts, upload.smoke.spec.ts, gallery.search-filters.spec.ts
Timing → Pushgateway yes API not wired — needs Pushgateway deployed (ansible deploy/observability/) + CI PLAYKIT_METRICS_* env

Product / coverage gaps

  • PROD smoke actually running — spec exists (prod.smoke.spec.ts) and is verified against a real public host, but stays skipped until PROD_BASE_URL is set (blocked on the LXC existing — see Ops/docs debt)
  • Consider a non-admin manager/editor-role FastAPI + NextAuth user if finer-grained role coverage is ever needed (today's viewer/admin split covers the binary write-access gate)
  • No latency/perf budget assertions anywhere — closed via expectWithinBudget() (see "Now — shipped / in flight"). Still no Core Web Vitals (LCP/CLS/INP) collection — separate gap, not addressed.

Deferred (soak policy)

  • Burn-in / deploy-smoke CLI (playkit ROADMAP v0.4/v0.5+) — playkit's own ROADMAP has an explicit "adoption pause" until the v0.3.1 network helpers + release job soak for a few days in this repo's CI; revisit once that settles rather than adding more moving parts at once.

Later

  • Proper DEV deploy (next start + CI image) instead of long-lived next dev
  • Ansible/app_setup path aligned with /opt/punimtag/viewer-frontend/.env (not /srv/app)
  • OpenAPI-driven contract suite (playkit / consumer)
  • Multi-browser matrix (firefox/webkit)
  • Hermes/Mattermost report on e2e failure
  • Enable playkit timing metrics in CI once Pushgateway is live

Test map

Layer Where Status
Viewer unit viewer-frontend Vitest CI viewer-unit
Backend unit/integration tests/ pytest existing
E2E browser + mail e2e/tests/* CI e2e
E2E API (unauth + catalog) e2e/tests/api.* CI
E2E FastAPI authed api.fastapi-login.spec.ts CI (secrets set)
E2E role-permission gates api.role-permissions.spec.ts CI (secrets set)
E2E gallery search filters gallery.search-filters.spec.ts CI (public, no login)
E2E favorites + people filter UI gallery.filters.authed.spec.ts CI (storageState)
E2E network silent-failure monitor network-errors.spec.ts CI
E2E NextAuth write gates (viewer vs admin) viewer.write-gates.spec.ts CI (secrets set)
E2E PROD smoke (health + login) prod.smoke.spec.ts dormant — needs PROD_BASE_URL (PROD LXC)
Admin UI unit admin-frontend Vitest CI admin-unit
Timing metrics push gap (ops + CI env)

See also: e2e/README.md, playkit ROADMAP.md / docs/NETWORK.md / docs/IDEAS.md.