fix(anpr): share one camera snapshot across bridge + advisory paths

On a vehicle entry, two paths captured the SAME Hikvision camera within ~1s —
the ANPR bridge (barrier-driving) and the advisory snapshotAsync (evidence/
telemetry) — each from a separate adapter instance. Hikvision serves snapshots
single-threaded, so the second concurrent GET returned HTTP 503; the bridge
then fail-softed and burned its 12s debounce, producing a ~74s "slow" subscriber
entry (observed 2026-06-25, Qazim Mulleti / AB816NN — plate read was instant at
conf 1.000; the delay was the 503/debounce churn, not recognition).

Add captureSnapshotShared() in snapshot.ts: a module-level, deviceId-keyed cache
that both paths call. It coalesces in-flight captures (the 2nd caller awaits the
1st's pull → no concurrent 503), serves a brief freshness window (1500ms) so the
bridge→advisory sequence for one vehicle reuses one frame, never caches a failure
(next caller retries), and keys by deviceId (no cross-camera/stale-vehicle reuse).
Wired into anpr-entry.ts (bridge) and snapshot.ts (advisory).

Tests: snapshot.test.ts (concurrent coalescing, TTL reuse, TTL-lapse re-pull,
failure-not-cached, per-camera keying); anpr-entry.test.ts mock updated. 168
server tests green.

NOTE: this removes the latency (the 503 collision). The separate double-entry
(two signed vehicle_entry for one car) — debounce-too-short / stamp-before-
success — is still open; less likely now but not eliminated.

Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
This commit is contained in:
2026-06-26 08:10:56 +02:00
parent b1c4109045
commit b3cb67188e
4 changed files with 170 additions and 4 deletions
+68 -2
View File
@@ -1,6 +1,6 @@
import { randomUUID } from "node:crypto";
import { deviceEvents as deviceEventsTable, snapshots, type Db } from "@parking/db";
import { registry, type CameraDevice } from "@parking/devices";
import { registry, type CameraDevice, type Snapshot } from "@parking/devices";
import type { FastifyBaseLogger } from "fastify";
import { devicesByDirection, type FlowDirection } from "./device-resolve.js";
import type { VisionClient } from "./vision-client.js";
@@ -62,7 +62,9 @@ export function snapshotAsync(job: SnapshotJob): Promise<string[]> {
return null;
}
try {
const shot = await camera.captureSnapshot({ direction });
// Shared capture: if the ANPR bridge just pulled this camera's frame for the
// same vehicle, reuse it instead of a 2nd concurrent GET (which 503s).
const shot = await captureSnapshotShared(row.id, camera, { direction });
const id: string = randomUUID();
db.insert(snapshots)
.values({
@@ -153,6 +155,70 @@ export function buildCamera(row: { driverId: string; config: unknown }): CameraD
}
}
// --- shared snapshot capture (one HTTP pull per camera per vehicle) -----------
// A Hikvision camera serves /ISAPI/.../picture SINGLE-THREADED: two concurrent
// snapshot GETs to the same unit return HTTP 503 "service busy". On a vehicle entry
// TWO paths capture the SAME camera within ~1s — the ANPR bridge (barrier-driving,
// anpr-entry.ts) and the advisory snapshotAsync (evidence + telemetry, below). They
// each `buildCamera()` a SEPARATE adapter instance, so a per-instance cache can't
// dedupe them. This module-level, deviceId-keyed cache does: it coalesces in-flight
// captures (the 2nd caller awaits the 1st's pull) AND serves a result captured within
// SNAPSHOT_TTL_MS, so the bridge + advisory share ONE frame instead of colliding into
// a 503 (which then burned the bridge's 12s debounce → the slow entry observed
// 2026-06-25; see wiki/concepts/lane-presence-and-anpr-entry.md).
/** How long a fresh capture is reused for the same camera. A car is one event for a
* couple of seconds; 1.5s comfortably spans the bridge→advisory gap without ever
* serving a stale frame for a *different* vehicle (entries are seconds apart). */
const SNAPSHOT_TTL_MS = 1500;
interface CacheEntry {
/** A capture in flight — concurrent callers await this instead of issuing a 2nd GET. */
inflight?: Promise<Snapshot>;
/** The last SUCCESSFUL capture + when it resolved, for the freshness window. */
last?: { shot: Snapshot; at: number };
}
const snapshotCache = new Map<string, CacheEntry>();
/**
* Capture a snapshot for a camera, sharing ONE HTTP pull across concurrent/near-
* simultaneous callers (the ANPR bridge and the advisory snapshot). Same contract as
* `camera.captureSnapshot` (throws on failure) — a failed pull is NOT cached, so the
* next caller retries rather than inheriting the error. Key by the stable `deviceId`.
*/
export function captureSnapshotShared(
deviceId: string,
camera: CameraDevice,
ctx: { direction: FlowDirection },
): Promise<Snapshot> {
const now = Date.now();
let entry = snapshotCache.get(deviceId);
if (!entry) {
entry = {};
snapshotCache.set(deviceId, entry);
}
// Fresh enough → reuse the last frame (same vehicle, no second hardware hit).
if (entry.last && now - entry.last.at < SNAPSHOT_TTL_MS) {
return Promise.resolve(entry.last.shot);
}
// A capture is already running → join it (this is what prevents the 503 collision).
if (entry.inflight) return entry.inflight;
// Otherwise issue the single real pull; record it as the in-flight promise.
const pull = camera
.captureSnapshot(ctx)
.then((shot) => {
entry.last = { shot, at: Date.now() };
return shot;
})
.finally(() => {
// Clear the in-flight slot whether it resolved or threw; a failure is never cached.
if (entry.inflight === pull) entry.inflight = undefined;
});
entry.inflight = pull;
return pull;
}
function recordFailure(
db: Db,
direction: FlowDirection,