- pnpm add maplibre-gl + -D @types/geojson. - src/map/core/styles.ts: defaultStyle (OSM raster bootstrap; 2.2 replaces with the basemap-switcher descriptor table). - src/map/core/map-view.tsx: module-level Map singleton lazily created on first <MapView> mount, attached to a class="trm-map-host" detached <div> that React refs append/remove on mount/unmount. Style-data lifecycle flips mapReady false on every styledata event, polls loaded() at 33ms intervals, flips ready true once the style is loaded — the canonical MapLibre style-swap dance. - Exports getMap()/getMapReady()/subscribeMapReady()/useMapReady (via useSyncExternalStore for SSR-safe + concurrent-safe reads). getMap() throws if called pre-mount; the explicit failure mode beats a null-able top-level export. - src/routes/_authed/monitor.tsx: new /monitor route, full-viewport <MapView /> for 2.1 (no children — subsequent tasks plug in here). - src/routes/_authed/index.tsx: home-page card now links to /monitor. - eslint.config.js: override for src/map/** + src/live/** disables react-refresh/only-export-components. Same pattern as the existing overrides for shadcn primitives and route files. Deviation: spec sketched a top-level `map` constant export; implemented as `getMap(): MapLibreMap` (a function) so the singleton stays lazy until <MapView> mounts. Top-level constant would either force eager init (breaks SSR/tests) or be nullable (footgun). The function form throws a clear error if called pre-mount. Bundle: /monitor lazy chunk is 1MB raw / 274KB gzipped (MapLibre + CSS). Other routes unaffected. Vite chunk-size warning is harmless.
7.7 KiB
Task 2.9 — Connection status + per-device last-seen indicators
Phase: 2 — Live monitoring map
Status: ⬜ Not started
Depends on: 2.4.
Wiki refs: docs/wiki/concepts/maps-architecture.md §"Visible system state".
Goal
Show operators just enough about system state so they can answer two questions at a glance:
- "Is the SPA still connected to live data?" — global WS status.
- "Is this specific device still reporting?" — per-device last-seen age.
Not noisy. Subtle UI; not banners and modal warnings. The design ethos: operators trust the map until something goes wrong, at which point they need a quick read of what went wrong without a wall of red.
Deliverables
src/ui/components/connection-chip.tsx—<ConnectionChip />rendered in the monitor route's chrome (top-right corner of the map UI, near the basemap switcher). ReadsuseConnectionStore. Three visible states:- Connected — small green dot, hidden text (or "Live" only on hover/expand). Subtle.
- Reconnecting — amber dot pulsing, text "Reconnecting…" with attempt count if useful.
- Disconnected / offline — red dot, text "Offline — last live: 14:02:11" showing the last successful contact time.
src/ui/components/device-last-seen.tsx—<DeviceLastSeen deviceId>component. Renders nothing in normal state; renders a small subscript / icon when the device's last position is older than threshold (default: 60s).- Used inside
<MapPositions>'s feature properties (or as a separate symbol layer keyed off thelastSeenAgederived field), and in any future per-device sidebar.
- Used inside
src/live/connection-store.tsupdated — addslastConnectedAt: number | null(already in 2.4's spec) andlastDisconnectedAt: number | null. Toast / chip use these for "last live: N min ago" formatting.src/live/last-seen.ts— utility:formatLastSeen(ts: number, now = Date.now()): stringreturns"now"/"5s ago"/"2m ago"/"14:02"based on age. Used by both the chip and the per-device indicator.- A "stale-position" derived signal — at coalescer flush time, the position store can compute
staleByDevice: Set<string>for devices whose last update is older thanSTALE_THRESHOLD_MS(default 60s). Map layers reading this Set apply a faded-icon variant. Or: a separate symbol layer renders a "warning" badge over stale devices' markers.
Specification
<ConnectionChip />
export function ConnectionChip() {
const status = useConnectionStore((s) => s.status);
const lastConnectedAt = useConnectionStore((s) => s.lastConnectedAt);
if (status === 'connected') {
return (
<div className="flex items-center gap-1.5 text-xs">
<span className="w-2 h-2 rounded-full bg-green-500" aria-hidden />
<span className="text-muted-foreground">Live</span>
</div>
);
}
if (status === 'connecting' || status === 'reconnecting') {
return (
<div className="flex items-center gap-1.5 text-xs">
<span className="w-2 h-2 rounded-full bg-amber-500 animate-pulse" aria-hidden />
<span className="text-muted-foreground">
{status === 'connecting' ? 'Connecting…' : 'Reconnecting…'}
</span>
</div>
);
}
return (
<div className="flex items-center gap-1.5 text-xs">
<span className="w-2 h-2 rounded-full bg-destructive" aria-hidden />
<span className="text-destructive">
Offline {lastConnectedAt ? `· last live ${formatLastSeen(lastConnectedAt)}` : ''}
</span>
</div>
);
}
Position: top-right of the monitor route, in a small horizontal bar with the basemap switcher and other controls. Inline-flex.
Per-device last-seen — two strategies
Strategy A: Faded marker. A separate symbol layer with a 'icon-opacity' expression based on a staleSec property on each feature:
'icon-opacity': [
'interpolate', ['linear'], ['get', 'staleSec'],
0, 1.0, // fresh
60, 1.0, // 60s — full opacity
300, 0.4, // 5min — faded
1800, 0.2, // 30min — very faded
],
The position store includes staleSec in each feature's properties at coalescer-flush time.
Strategy B: Badge / overlay. A second symbol layer drawing a small "no-signal" icon over stale markers, filtered to ['>=', ['get', 'staleSec'], 60].
Pick A for v1. Simpler, fewer layers, cleaner visually. B can layer on top later if A isn't legible enough.
Updating staleSec
Recomputed every second by a setInterval inside the position store (not on every position update — that would invalidate latestByDevice constantly):
// In src/live/position-store.ts
let staleTickerId: ReturnType<typeof setInterval> | null = null;
function startStaleTicker() {
if (staleTickerId) return;
staleTickerId = setInterval(() => {
const now = Date.now();
set((state) => {
// Only bump versionTick if any device's stale-bucket changed.
// ... or just trigger subscribers wholesale every second.
return { stalenessTick: now };
});
}, 1000);
}
Map layers re-derive staleSec from latestByDevice + stalenessTick on every flush. The store doesn't mutate position data; it just bumps a version tick that triggers selectors that compute staleSec themselves.
Logout / unmount cleanup
The stale ticker runs as long as the SPA is mounted. Stop on logout (status flips to anonymous) or when navigating away from /monitor. Use a useEffect in <MonitorPage>:
useEffect(() => {
startStaleTicker();
return () => stopStaleTicker();
}, []);
What this task does NOT include
- Loud "device offline!" notifications. No toast, no modal. The faded icon is the signal. If operators need louder, Phase 3.4's per-device detail panel can add an alert badge.
- Per-event "X devices offline" summary. Phase 3 polish.
- Auto-recovery testing for the WS reconnect. That's covered by 2.4's acceptance.
- Notification API integration. Browser push on disconnect — Phase 4.
Acceptance criteria
pnpm typecheck,pnpm lint,pnpm format:check,pnpm buildclean.- On
/monitor, the connection chip top-right shows "Live" with a green dot during normal operation. - Disconnecting the network: the chip flips to "Reconnecting…" within a few seconds; on reconnect, back to "Live"; on permanent disconnect (close + no reconnect for 30s+), shows "Offline · last live HH:MM:SS".
- A device that hasn't reported in 5+ minutes appears noticeably faded on the map (not invisible — operators still see where it was last seen).
- On a fresh page load with no positions yet, the chip says "Connecting…" briefly, then "Live" once the snapshot arrives.
- No banner / toast spam during normal operation.
Risks / open questions
- Threshold tuning. 60s = "fresh", 5min = "fading", 30min = "stale". For Teltonika devices reporting at 1Hz this is generous; for 0.2Hz devices the 60s window catches a device that just paused at a checkpoint. Watch dogfood data and adjust.
- Setinterval ticker accuracy.
setInterval(1000ms)is approximate; on a slow tab it can drift. The visual effect doesn't need wall-clock precision — "this marker is faded because it's been a while" is still correct even if the timer drifted by a few seconds. - Race with reconnect on offline. If the SPA is offline and reconnects, the snapshot replays; some markers' last-seen will jump from "stale" to "fresh" in one frame. Verify the visual transition is clean (no flicker).
- Connection chip during boot. Right after login, the WS hasn't connected yet — chip says "Connecting…". Should this be its own state, or just folded into "reconnecting"? Spec above keeps
connectingdistinct for clarity; either is defensible.
Done
(Filled in when the task lands.)