Task 05: Kubernetes settings experience

Acceptance: admins can create/test/save Kubernetes executors and edit raw Pod-template/storage profiles; members see a read-only configuration boundary.
Acceptance: pure parse/serialize/dirty-state/API logic is tested and all executor capability/filter/presentation switches recognize k8s.
Acceptance: desktop and phone preserve capability parity, visible touch actions, one document scroll owner, 44 px primary targets, and no document horizontal overflow; all copy is translated in six catalogs.
Verification: cd apps && pnpm install --frozen-lockfile && pnpm --filter @kandev/web test -- --run && pnpm --filter @kandev/web lint && cd web && pnpm run typecheck && pnpm run i18n:check && pnpm run i18n:ratchet
Files likely touched: apps/web/app/settings/executors/**, focused components/settings/kubernetes-*.tsx, config serializers/baselines, settings API/types, executor capability/presentation files, and six locale catalogs.
Dependencies: Task 03 API contract.
Parallelism: sequential UI contract owner; E2E and docs consume it.
Inputs: spec What/API/Permissions/mobile scenario; mobile UI language; SSH create/connection/profile patterns.

Results

Implemented the Kubernetes settings experience against the final Task 03 HTTP contract. Administrators can create, edit, delete, and test Kubernetes executors and profiles; members retain profile use and sanitized active-session status while every Kubernetes mutation/test surface is visibly read-only or disabled with an administrator explanation. The shared settings save coordinator owns executor/profile saves and components reach the backend only through domain hooks and API clients.

Executor/profile creation is coordinated and compensating: the hook returns both created identities, the page routes to the canonical executor connection URL, and a failed profile create deletes the just-created executor. Rejected or failed rollback paths retain both causes and surface localized, actionable copy. Member visits to the direct create route remain readable but do not register dirty state or trap navigation.

The responsive cards cover kubeconfig and in-cluster authentication, a fixed namespace, request timeout, platform, main container, strict raw PodTemplate YAML, conditional managed-PVC/emptyDir/existing-claim fields, admission warnings, causal test steps, and 90-second active-session refresh. Phone and desktop share state and handlers; phone actions use 44 px targets, YAML/table content contains its own horizontal overflow, and page shells clip document overflow without adding a second vertical scroll owner. Executor routes, breadcrumbs, icons, labels, cleanup copy, task controls, remote-workspace capabilities, agent credential filtering, and supporting type switches now recognize k8s.

The integration routing audit also closed every legacy entry path. Settings tree executor/profile rows and ExecutorProfilesCard now choose the dedicated Kubernetes connection route or Kubernetes-aware flat profile editor while SSH and other executors retain their scoped routes. Direct or bookmarked singular Kubernetes routes replace themselves before mounting legacy editable controls, including for members. The unused parallel executor/settings declarations in lib/settings/types.ts were removed after rg confirmed that only Theme, KeyValue, and AgentProfile have consumers; canonical HTTP executor types remain authoritative.

Kubernetes profile editing participates in the existing remote-executor flow, including credentials, configuration bundles, and Git identity. Workload and storage parsing and serialization preserve unrelated profile keys. Saves first refresh the persisted session inventory and require explicit confirmation when sessions are active: current connection configuration applies on reconnect, while workload edits apply to new sessions and the saved workload snapshot continues to govern existing/replacement Pods. The ready baseline image is ghcr.io/kdlbs/kandev:latest, with localized guidance to pin a tag or digest for production and to retain the required bootstrap and agent dependencies.

Localization was authored in en, pt-pt, and zh-cn; zh-hk and zh-tw were generated with pnpm run i18n:zh-hant, and the pseudo catalog was regenerated with pnpm run i18n:pseudo. The final generators reported 17,492 real web messages with zero residual Simplified Chinese warnings and 8,746 pseudo messages.

Verification completed on 2026-08-24:

cd apps && pnpm install --frozen-lockfile passed; the lockfile and install were already up to date (pnpm 9.15.9).
The final strict-TDD command was cd apps/web && pnpm exec vitest run components/settings/kubernetes-config.test.ts components/settings/kubernetes-validation.test.ts components/settings/kubernetes-create-error.test.ts components/settings/kubernetes-save-confirmation.test.ts components/settings/kubernetes-settings-cards.test.tsx components/settings/profile-edit/serialize-executor-config.test.ts components/settings/profile-edit/profile-edit-page-chrome.test.tsx hooks/domains/settings/use-kubernetes-settings.test.tsx lib/api/domains/kubernetes-api.test.ts src/settings-routes.test.ts; it passed 10 files and 86 tests in 35.47 seconds.
The integration-routing RED run covered the settings tree, profile card, and singular route delegation: 5 expected failures and 64 passes. A second RED isolated direct/bookmarked routing: the two Kubernetes redirects failed while both SSH preservation cases passed. After GREEN and stale-type cleanup, cd apps/web && pnpm exec vitest run components/app-sidebar/sections/settings/settings-menu-branches.test.ts components/settings/executor-profiles-card.test.tsx components/settings/legacy-executor-settings-route.test.tsx src/settings-routes.test.ts hooks/domains/settings/use-kubernetes-settings.test.tsx passed 5 files and 82 tests in 75.88 seconds.
The completed full assertion run was cd apps && timeout --signal=TERM --kill-after=10s 60m env VITEST_MAX_WORKERS=3 pnpm --filter @kandev/web test -- --run --pool=forks --testTimeout=30000 --hookTimeout=30000 --dangerouslyIgnoreUnhandledErrors; Vitest 4.1.10 exited 0 after 1,543 files, 13,053 passing tests, 4 skipped tests, and 1,656.98 seconds. It still reported three unrelated Monaco lazy-import EnvironmentTeardownError diagnostics; the identical non-suppressing fork run passed every assertion but exited 1 solely for three such diagnostics. The thread-pool aggregate passed every assertion too but amplified this existing teardown problem to 135 diagnostics.
Earlier bounded aggregate attempts were stopped rather than left orphaned: the default single-worker run exited 143 at the integration reviewer's request; 3-worker thread runs reached their 7, 12, and 30 minute guards with no Task 05 failure; a 4-worker 20-minute probe hit two unrelated 5-second load timeouts. Those two files then passed alone: 2 files and 31 tests in 58.15 seconds.
cd apps && pnpm --filter @kandev/web lint passed with zero warnings.
cd apps/web && pnpm run typecheck passed (tsc --noEmit).
cd apps/web && pnpm run i18n:check passed: 7,228 referenced keys, 8,746 English entries, 16 existing orphan warnings, pseudo in sync, and all four real translated catalogs complete.
cd apps/web && pnpm run i18n:ratchet passed: 0 added and 23 modified source files clean; the 639-entry guard allowlist remained intact.
cd apps/web && pnpm run build:e2e passed under Vite 8.0.16 with only the repository's existing chunk-size/dynamic-import warnings (14,796 modules, 9.03 seconds).
Focused Playwright desktop and Pixel 5 smoke coverage passed: 2 tests in 10.6 seconds (chromium and mobile-chrome). E2E files were not edited.
git diff --check passed.
Per the integration reviewer's instruction, no additional full aggregate was started after the routing/type cleanup; the completed aggregate result and Monaco teardown caveat above remain the terminal full-suite result.

No commit, push, or pull request was created.

8.2 KiB Raw Blame History

Task 05: Kubernetes settings experience

Results

8.2 KiB

Raw Blame History