94ace2df0e
4 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
Diseri Pearson
|
94ace2df0e |
Portal frontend: apply Prettier formatting baseline
One-time `prettier --write` across the 22 source files not already touched
by the preceding tooling commit (
|
||
|
Diseri Pearson
|
e2cbb83397 |
Portal: Client Dashboard, Measurements page, Excel exports, Grafana auth, RLS
A bundle of related portal work — picked up while ensuring per-customer
isolation actually works end-to-end and replacing the placeholder Client
landing page. Build green, full test suite 66/66.
Frontend — Client surface
- DashboardPage: replace placeholder with 4 KPI cards (kWh, current kW,
active devices, estimated cost), 24h active-power ECharts mini-chart,
per-device "today/range" table, and a date-range picker with shortcuts
(Today / 7d / 30d / This month / Custom). 30s auto-refresh.
- New Measurements page (/measurements, Client mode, any authenticated
user) with multi-select device filter, full date range incl. an
"All time" shortcut, server-paginated preview, and Excel export.
- "Export to Excel" buttons on: Client Dashboard summary, Client Dashboard
raw measurements, Admin fleet dashboard, Admin customer-detail Cost tab.
- DashboardsPage sidebar items: let the menu item grow and reset
line-height so the two-line title+description doesn't crush.
Frontend — Admin / user mgmt
- RestrictedAdmin role: admin who only sees their assigned customers.
New UserFormDrawer choice + CustomerAccessModal for granting/revoking
per-customer access; surfaced from the Users page.
Backend
- ClosedXML 0.104.2 + ExcelExportService (pure formatter; frozen header,
currency/kWh/kW/date number formats, AdjustToContents).
- DashboardSummaryService computes per-device totals + estimated cost
(hourly bucketing × site's municipality's active tariff, mirroring
FleetCostService for the Admin side).
- New endpoints:
GET /api/dashboard/summary[+/export.xlsx]
GET /api/measurements/raw[+/export.xlsx] (deviceIds, paginated)
GET /api/sites/devices (flat list w/ site name)
GET /api/fleet/dashboard/export.xlsx
GET /api/fleet/customers/{id}/cost/export.xlsx
GET /api/auth/check (cookie-only liveness)
- AdminCustomerAccess: per-user customer scoping for RestrictedAdmin via
Postgres-row-level filter — RlsContext (per-DI-scope state) +
CustomerFilterMiddleware (populates from claims after auth) +
fleet.* DbSets gain HasQueryFilter expressions. Bootstrappers
Elevate() to bypass the filter for trusted system code.
- Migration: 20260518095759_AddAdminCustomerAccess (mapping table,
composite PK on UserId+CustomerId).
Infra / templating (the "spin it up via the template" piece)
- docker-compose.prod.yml + docker-compose.yml: pass WhiteLabel__*,
Application__RunMode, FleetIngest__* through to the container as
${VAR:-default} substitutions. Previously these were silently dropped
in prod — a customer's .env settings for branding/fleet-push never
reached the running process. Latent bug, fixed.
- docker-compose.prod.yml: forwardAuth middleware labels on the
Grafana router pointing at /api/auth/check. Option (a) from the
README's three prod-auth modes — every Grafana request now gates on
a valid portal cookie. Anonymous stays off.
- .env.example rewritten with a Client section, optional FleetIngest
block, and an Admin variant block — annotated on what's required vs.
optional and where the seed-only-on-first-boot caveat applies.
- README "Grafana embedding" table: option (a) now marked active with
an inline note on how to switch modes later.
- OPERATIONS.md step 3 includes the white-label pre-brand .env snippet;
step 4 (formerly "decide Grafana auth mode") updated to reflect
that auth is wired by default.
Tests
- New BrandingSeedFromOptionsTests (5 tests) pins the env-var → IOptions
→ DB seed contract: first read seeds from options; subsequent reads
return the DB row (UI edits survive restarts); EnsureSeededAsync is
idempotent; UpdateAsync falls back to options for blanked fields.
- CustomerTokenGraceTests helper: pass the new RlsContext to
AdminDbContext (SetAll() so existing semantics hold).
Verified end-to-end
- Real Docker spin-up with WhiteLabel__* in a throwaway .env →
/api/branding returned all six fields verbatim (ApplicationName,
LogoUrl, three colors, FooterText).
- curl login → /api/dashboard/summary returned valid JSON →
/api/dashboard/summary/export.xlsx returned a 6.9 KB file the
`file` command identifies as "Microsoft Excel 2007+".
- /api/measurements/raw with and without deviceIds filter returned
correct paginated rows; /export.xlsx with filter produced a valid
7.1 KB xlsx with the meter count in the filename.
- Frontend tsc -b clean; backend dotnet build 0/0; xunit 66/66.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
Diseri Pearson
|
66660364ec |
Cost-feature followups: libgssapi fix, daily CA, dashboard cost
Three small polish items on top of the cost-compute feature. (A) Dockerfile: install libgssapi-krb5-2 in the runtime stage. Silences the "Cannot load library libgssapi_krb5.so.2" warning that Npgsql logs when probing for Kerberos auth on Linux. We don't actually use Kerberos; the lib is ~3 MB and removes confusing log noise. (B) FleetTimescaleBootstrapper: add hierarchical fleet.daily_per_customer continuous aggregate on top of fleet.hourly_per_device. Per-customer daily totals (samples, kwh imported/exported, avg/max/min kW). Realtime, materialized_only=false, 365-day start_offset, hourly refresh. Available for long-range dashboards / billing summaries that don't need device-level or hourly detail. Not yet consumed by any query — exists as a primitive for the next dashboard / report that wants it. (C) Fleet dashboard: include per-customer cost today. - FleetQueryService.GetDashboardAsync invokes FleetCostService.ComputeAsync per customer for today's UTC window. Failures for one customer don't break the dashboard (leave their CostToday null and continue). - FleetCustomerSummary gains nullable CostToday; FleetDashboardDto gains TotalCostToday (sum across customers). - AdminFleetDashboardPage adds a 5th Statistic card "Cost today" in green, and a "Today (cost)" column in the customer summary table. - Perf note added in service comment: per-load cost compute scales with N customers; revisit with a materialised daily_cost view when N grows beyond ~100. Verified on the running stack - Admin container logs no longer contain "libgssapi" anywhere. - timescaledb_information.continuous_aggregates shows both hourly_per_device and daily_per_customer. - /api/fleet/dashboard returns totalCostToday=161.00, the single customer (DEV0001) row shows costToday=161.00 matching the cost endpoint's total for the same UTC day. Deliberately not in this commit: per-customer Postgres RLS for multi-Admin-user setups. That needs design decisions (claim source, fleet-admin bypass model, connection-pooling interaction with session variables) — I'd rather pose those questions and ship it right than sneak in a half-baked version. Wrap-up message has the design qs. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Diseri Pearson
|
c5787a7a7f |
Phase 15: Admin operator surface + fleet dashboards + onboarding docs
The Admin stack now has a usable operator UI for managing the fleet.
End-to-end verified locally: Client pushes → Admin dashboard reflects
the activity within the CA refresh window.
Backend (Admin-only)
- FleetQueryService: dashboard headline (totals, active count, today's
measurements + kWh from the hourly_per_device CA) and per-customer
detail (sites, devices, last 50 measurements, last 20 ingest events).
- /api/fleet/dashboard and /api/fleet/customers/{id}/detail endpoints.
- DTOs added; Program.cs wires the service + endpoints under RunMode=Admin.
Frontend
- DashboardPage now branches on RunMode — Admin renders the fleet
headline (statistic cards + customer summary table with lag tags),
Client keeps the existing placeholder.
- AdminCustomerDetailPage drills into one customer: descriptions card +
tabs for Recent ingest (with rejection counts, batch sizes, time-spread
for visible firmware-replay waves), Recent measurements, Sites, Devices.
- AdminCustomersPage rows are clickable → /admin/customers/:id (skips
the click when target is a button/popover so action buttons still work).
- App.tsx adds the /admin/customers/:id route, RequireRole-gated.
Grafana
- grafana/dashboards-admin/fleet-overview.json — 4 stat panels (active
customers, total, last-24h samples, today's kWh) plus 2 time series
(per-customer active power, per-customer hourly kWh). Reads from
fleet.hourly_per_device CA.
- grafana/dashboards-admin/customer-drilldown.json — parameterized by
$customer (template variable querying fleet.Customers). Per-device
active power, cumulative kWh, recent ingest events table.
Docs
- README: Phase 15 section describing the new admin UI surface +
pointer to dashboard-admin folder.
- OPERATIONS: new "Fleet aggregator (Admin stack)" section covering
one-time provisioning (Admin portal + Admin Grafana), end-to-end
customer-onboarding workflow (register on Admin → drop token in
customer .env → restart → verify in UI/SQL), common ops (rotate
token, disable, investigate, compression stats, force CA refresh,
decommission), and Admin-DB backup notes.
- README decommissioning note now mentions deleting from fleet.Customers
if the customer was registered for aggregation.
Verified end-to-end
- Phase 14's Client + Admin stacks rebuilt with Phase 15 code.
- /api/fleet/dashboard returns correct totals (1 customer, 1 active,
measurements + kWh derived from CA).
- /api/fleet/customers/{id}/detail returns sites, devices, recent
measurements, recent ingest events.
- Ingested a fresh measurement on Client → after CA refresh, totals
in Admin dashboard advance correctly.
- All 53 tests still passing.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|