2026-07-17 dashboard compute ui components migration and deploy

Ran PR #248 (Compute Dashboard) locally in mock mode, migrated its frontend to reuse @virdx/ui-components, and drove the build+deploy of dashboard-frontend/dashboard-api images.

Work Done

Context: virdx/mono PR #248 “feat: Compute Dashboard” (branch alex.leake/dashboard_compute) added a COMPUTE section (Argo workflows, Tekton builds, Coder workspaces) — ~3.6k lines across dashboard-api (backend: auth, collectors, routers, services, models, mock_data, tests) and dashboard-frontend.

Ran the features locally without cluster access. dashboard-api has a full mock mode (DASHBOARD_MOCK=true): every compute endpoint (/builds, /workflows, /coder) has an if settings.mock: fast-path returning mock_data.py, and require_user returns a fake dev.user (auth fully bypassed). Minimal local stack = dashboard-api (mock) + frontend; no vxdata/postgres needed for compute tabs.
UI-components reuse migration. The compute code rolled its own UI via a local apps/dashboard-frontend/src/components/ui/* shadcn copy (18 files, pre-existing on main), NOT the shared @virdx/ui-components (56-component superset). Migrated 7 files (StatusBadge, BuildTaskDetail, BuildsList, Builds, WorkflowsList, Workflows, AuthStatus) to import Badge/ Button/Card/Table/Skeleton/Dialog from @virdx/ui-components/ui. Committed + pushed 7ec5679 to the branch; merged to main via PR #248 (merge 2be32cf).
Kept ui-components untouched (user pref). Shared Badge lacks a success variant (has default/secondary/destructive/outline/ghost/link). First added success to the shared badge, but user preferred NOT to touch the shared lib from a feature PR — reverted, instead tinted a neutral secondary pill green via className='bg-success/10 text-success dark:bg-success/20' in StatusBadge. Verified: prod build passes, typecheck/lint clean on touched files (2 pre-existing tsc errors in DataSourcePicker/useBenchmarkSeries are unrelated).
Deploy. dashboard apps deploy by image tag, NOT semver — no version bump needed. Manifests infra_k8s/current/dashboard/{api,frontend}.yaml pin :latest + Keel (keel.sh/trigger: poll, @every 5m, policy: force). Builds normally fire from a scheduled argo-events poll (dashboard-poll) building from main. Triggered manual builds via apps/dashboard-{frontend,api}/deployments/build.sh (Tekton buildkit-image PipelineRun, revision main, tag latest). First frontend build failed on a transient zot registry push flake (error writing layer blob: unknown: blob upload unknown to registry) — resubmitted. At session end: frontend retry f72ms + api jhvzh running; Keel rolls pods within ~5 min of push.

Lessons Learned: Pitfalls

“API unchanged” trap. I initially told the user only the frontend needed rebuilding because my refactor commit was frontend-only. But the merged PR #248 added ~2100 lines to dashboard-api — that image needed rebuilding too. When advising on rollout, scope to the whole merged diff (git diff 731c551 origin/main -- apps/<app>/), not just your own commit.
process-compose TUI dies in background. scripts/dashboard-stack-up.sh in the background fails with open /dev/tty: device not configured. Pass -t=false to run headless. Also the full script relaunches vxdata+postgres and collides if a dev already has their own vxdata stack up (their process-compose REST API defaults to 8080, colliding with dashboard-api). For compute-only tryouts, skip the script: run dashboard-api (mock) on a free port + frontend with VITE_DASHBOARD_API_PROXY_TARGET.
@virdx/ui-components needs building before the frontend runs. No dist/ exists after a fresh bun install; Vite fails to resolve @virdx/ui-components/ui. Run bunx nx build ui-components first. After editing/rebuilding it, clear node_modules/.vite and restart the dev server (Vite won’t re-optimize a workspace dep’s dist automatically).
Misleading build-failure snippet. A pasted “bun install –frozen-lockfile: Failed to install 1 package” looked like a lockfile/Dockerfile bug, but reading the actual build-pod logs (kubectl logs <pr>-build-pod -n tekton-builds) showed install SUCCEEDED (3171 pkgs) and the real failure was the zot push. Always read the full build-pod log before diagnosing; buildkit step order and retries make single snippets deceptive.

Lessons Learned: Improvements

Add an infrastructure SOP for the dashboard app covering: (a) local dev — mock mode + minimal 2-process launch, the bunx nx build ui-components prerequisite, headless dashboard-stack-up.sh -t=false; (b) rollout — image-tag + Keel + argo-events poll model (no version bump), and deployments/build.sh for a manual build from main; (c) the known transient zot blob upload unknown to registry push flake → just resubmit.
Document the local-real-auth blocker. Real Keycloak login locally is blocked because the dashboard client (infra_ansible inventories/group_vars/keycloak/keycloak.yaml) only whitelists https://dashboard.fra.virdx.dev/* as a redirect URI — no http://localhost:5173/*. Worth an SOP note so nobody rediscovers this. Also needs DASHBOARD_TLS_CA_FILE on macOS and a separate DASHBOARD_CODER_TOKEN for the Coder card.
Convergence debt: dashboard-frontend has a split-brain UI layer — local @/components/ui/* copy vs shared @virdx/ui-components. Worth a tracked task to migrate the whole app onto the shared lib and delete the local copy (this session only did the compute-PR files).