feat(faces): consolidate look-alike clusters after a scan, and suggest the rest (#1107)
consolidate() has existed since #1074 and described this exact symptom in its own comment, but its only caller was recluster() — i.e. when an admin pressed Re-group people. After a normal background scan the centroids converged and nobody looked, so a gallery settled with 14 people that should have been 8. It now runs when a scan drains. There is no scan-finished event to hook, so an idle worker asks whether the events it touched have actually drained — 'a worker went idle' is deliberately not treated as sufficient, because with concurrency above one the others may still be working. The uncertain band asks instead of acting: pairs between the assignment threshold and the stricter auto-merge one surface as accept/dismiss suggestions, with sticky dismissals. Nothing merges silently — a pass that merged anything reports it and points at Split. Review rounds hardened it against overruling explicit decisions: it no longer absorbs ignored clusters (mergePeople ORs is_ignored onto the survivor, which would have hidden a real person), no longer merges dismissed pairs, no longer undoes a manual Split (which now records a separation), and no longer runs after detection is switched off. The dismissal read fails closed, a failed pass is retried with backoff rather than lost or hot-looped, and the new table follows event_people out of exports and backups. Name autocomplete needs no endpoint — the people list already open is the source, and it is event-scoped on purpose. Known limitation, tracked in #1132: separations are keyed on person ids, so a full re-scan loses them. Reported by @BraynArts.
This commit is contained in:
@@ -0,0 +1,219 @@
|
||||
/**
|
||||
* "The scan finished" is not a thing this queue is told (#1107).
|
||||
*
|
||||
* It claims photos one at a time, so a backfill is just a lot of independent
|
||||
* claims and the only available signal is a worker finding nothing left. That
|
||||
* signal is NOT sufficient on its own — with concurrency above one the other
|
||||
* workers may still be busy, and a photo released back to `pending` by a down
|
||||
* sidecar is still owed — so the drain is tested against the queue directly.
|
||||
*
|
||||
* These are the cases that decide whether consolidation runs too early (a
|
||||
* wasted pass over half-formed clusters) or never (the feature silently does
|
||||
* nothing, which is the state #1107 was filed about).
|
||||
*/
|
||||
|
||||
const path = require('path');
|
||||
const fs = require('fs');
|
||||
const os = require('os');
|
||||
|
||||
process.env.NODE_ENV = 'test';
|
||||
process.env.TEST_DATABASE_PATH = path.join(
|
||||
fs.mkdtempSync(path.join(os.tmpdir(), 'picpeak-facedrain-')), 'db.sqlite',
|
||||
);
|
||||
process.env.JWT_SECRET = process.env.JWT_SECRET || 'facedrain-test-secret';
|
||||
|
||||
const { bootCrmDb } = require('./helpers/crmDb');
|
||||
|
||||
let db; let cleanup; let faceQueue; let clustering;
|
||||
|
||||
async function seedEvent(slug) {
|
||||
const [row] = await db('events').insert({
|
||||
slug,
|
||||
event_type: 'wedding',
|
||||
event_name: slug,
|
||||
event_date: '2026-01-01',
|
||||
host_email: '[email protected]',
|
||||
admin_email: '[email protected]',
|
||||
password_hash: 'x',
|
||||
share_link: `${slug}-share`,
|
||||
expires_at: new Date().toISOString(),
|
||||
// The drain rechecks this before consolidating, so the fixture has to be
|
||||
// a gallery that actually has detection on.
|
||||
face_recognition_enabled: true,
|
||||
}).returning('id');
|
||||
return typeof row === 'object' ? row.id : row;
|
||||
}
|
||||
|
||||
/** Both halves of the "two deliberate actions" rule have to be on. */
|
||||
async function enableFacesGlobally() {
|
||||
const existing = await db('feature_flags').where({ key: 'faces' }).first();
|
||||
if (existing) await db('feature_flags').where({ key: 'faces' }).update({ value: true });
|
||||
else await db('feature_flags').insert({ key: 'faces', value: true });
|
||||
}
|
||||
|
||||
async function insertPhoto(eventId, faceStatus) {
|
||||
const [row] = await db('photos').insert({
|
||||
event_id: eventId,
|
||||
filename: `${Math.random()}.jpg`,
|
||||
path: '/tmp/x.jpg',
|
||||
type: 'individual',
|
||||
face_status: faceStatus,
|
||||
}).returning('id');
|
||||
return typeof row === 'object' ? row.id : row;
|
||||
}
|
||||
|
||||
describe('faceQueue drain consolidation (#1107)', () => {
|
||||
beforeAll(async () => {
|
||||
({ db, cleanup } = await bootCrmDb());
|
||||
faceQueue = require('../../src/services/faceQueue');
|
||||
clustering = require('../../src/services/faceClustering');
|
||||
await enableFacesGlobally();
|
||||
}, 120000);
|
||||
|
||||
afterAll(async () => { if (cleanup) await cleanup(); });
|
||||
|
||||
beforeEach(() => {
|
||||
faceQueue.touchedEvents.clear();
|
||||
faceQueue.consolidationRetryAt.clear();
|
||||
faceQueue.inFlightByEvent.clear();
|
||||
jest.restoreAllMocks();
|
||||
});
|
||||
|
||||
it('does nothing at all when no photo has been scanned', async () => {
|
||||
const spy = jest.spyOn(clustering, 'consolidate');
|
||||
await faceQueue.drainConsolidation();
|
||||
expect(spy).not.toHaveBeenCalled();
|
||||
});
|
||||
|
||||
it('waits while the event still has photos queued', async () => {
|
||||
const eventId = await seedEvent('drain-pending');
|
||||
await insertPhoto(eventId, 'done');
|
||||
await insertPhoto(eventId, 'pending');
|
||||
faceQueue.touchedEvents.add(eventId);
|
||||
|
||||
const spy = jest.spyOn(clustering, 'consolidate');
|
||||
await faceQueue.drainConsolidation();
|
||||
|
||||
expect(spy).not.toHaveBeenCalled();
|
||||
// Still owed, so it must keep its place for the next idle tick — dropping
|
||||
// it here would mean the gallery never consolidates at all.
|
||||
expect(faceQueue.touchedEvents.has(eventId)).toBe(true);
|
||||
});
|
||||
|
||||
it('waits while a photo is still being processed by another worker', async () => {
|
||||
const eventId = await seedEvent('drain-processing');
|
||||
await insertPhoto(eventId, 'done');
|
||||
await insertPhoto(eventId, 'processing');
|
||||
faceQueue.touchedEvents.add(eventId);
|
||||
|
||||
const spy = jest.spyOn(clustering, 'consolidate');
|
||||
await faceQueue.drainConsolidation();
|
||||
|
||||
expect(spy).not.toHaveBeenCalled();
|
||||
expect(faceQueue.touchedEvents.has(eventId)).toBe(true);
|
||||
});
|
||||
|
||||
it('consolidates once the queue is empty, and does not repeat itself', async () => {
|
||||
const eventId = await seedEvent('drain-empty');
|
||||
await insertPhoto(eventId, 'done');
|
||||
await insertPhoto(eventId, 'failed');
|
||||
await insertPhoto(eventId, 'skipped');
|
||||
faceQueue.touchedEvents.add(eventId);
|
||||
|
||||
const spy = jest.spyOn(clustering, 'consolidate').mockResolvedValue([]);
|
||||
await faceQueue.drainConsolidation();
|
||||
|
||||
expect(spy).toHaveBeenCalledTimes(1);
|
||||
expect(spy).toHaveBeenCalledWith(eventId);
|
||||
// Drained and handled, so a second idle tick must not pay for it again.
|
||||
expect(faceQueue.touchedEvents.has(eventId)).toBe(false);
|
||||
|
||||
await faceQueue.drainConsolidation();
|
||||
expect(spy).toHaveBeenCalledTimes(1);
|
||||
});
|
||||
|
||||
it('a failing consolidation never propagates into the worker loop, and is retried', async () => {
|
||||
const eventId = await seedEvent('drain-throws');
|
||||
await insertPhoto(eventId, 'done');
|
||||
faceQueue.touchedEvents.add(eventId);
|
||||
|
||||
const spy = jest.spyOn(clustering, 'consolidate').mockRejectedValue(new Error('boom'));
|
||||
|
||||
await expect(faceQueue.drainConsolidation()).resolves.toBeUndefined();
|
||||
|
||||
// A transient database error must not cost the gallery its consolidation
|
||||
// outright — the event keeps its place so a later tick retries.
|
||||
expect(faceQueue.touchedEvents.has(eventId)).toBe(true);
|
||||
|
||||
// ...but not on the very next tick. The worker idles every couple of
|
||||
// seconds, so an immediate retry would hot-loop a permanently broken event
|
||||
// and warn every time.
|
||||
expect(faceQueue.consolidationRetryAt.get(eventId)).toBeGreaterThan(Date.now());
|
||||
const callsBefore = spy.mock.calls.length;
|
||||
await faceQueue.drainConsolidation();
|
||||
expect(spy).toHaveBeenCalledTimes(callsBefore);
|
||||
|
||||
// Once the backoff elapses it really does try again, and succeeds.
|
||||
faceQueue.consolidationRetryAt.set(eventId, Date.now() - 1);
|
||||
spy.mockResolvedValue([]);
|
||||
await faceQueue.drainConsolidation();
|
||||
expect(faceQueue.touchedEvents.has(eventId)).toBe(false);
|
||||
expect(faceQueue.consolidationRetryAt.has(eventId)).toBe(false);
|
||||
});
|
||||
|
||||
it('waits while another worker is still inside processPhotoFaces', async () => {
|
||||
const eventId = await seedEvent('drain-inflight');
|
||||
// Every row already reads as drained: the last photo is committed 'done'
|
||||
// inside the transaction, and auto-categorisation runs afterwards. Only
|
||||
// the in-flight count knows a worker is still there.
|
||||
await insertPhoto(eventId, 'done');
|
||||
faceQueue.touchedEvents.add(eventId);
|
||||
faceQueue.inFlightByEvent.set(eventId, 1);
|
||||
|
||||
const spy = jest.spyOn(clustering, 'consolidate').mockResolvedValue([]);
|
||||
await faceQueue.drainConsolidation();
|
||||
|
||||
// Consolidating here would record its count, and the busy worker would
|
||||
// then re-mark the event — the next pass merges nothing and overwrites the
|
||||
// real number with zero.
|
||||
expect(spy).not.toHaveBeenCalled();
|
||||
expect(faceQueue.touchedEvents.has(eventId)).toBe(true);
|
||||
|
||||
faceQueue.inFlightByEvent.delete(eventId);
|
||||
await faceQueue.drainConsolidation();
|
||||
expect(spy).toHaveBeenCalledTimes(1);
|
||||
});
|
||||
|
||||
it('does not consolidate an event whose detection was switched off mid-drain', async () => {
|
||||
const eventId = await seedEvent('drain-disabled');
|
||||
await insertPhoto(eventId, 'done');
|
||||
await db('events').where({ id: eventId }).update({ face_recognition_enabled: false });
|
||||
faceQueue.touchedEvents.add(eventId);
|
||||
|
||||
const spy = jest.spyOn(clustering, 'consolidate').mockResolvedValue([]);
|
||||
await faceQueue.drainConsolidation();
|
||||
|
||||
// An earlier photo legitimately marked the event before the toggle went
|
||||
// off. Merging someone's clusters just after they disabled the feature is
|
||||
// not a thing to do quietly.
|
||||
expect(spy).not.toHaveBeenCalled();
|
||||
// Dropped rather than retried — it is not coming back on its own.
|
||||
expect(faceQueue.touchedEvents.has(eventId)).toBe(false);
|
||||
});
|
||||
|
||||
it('treats events independently — a busy gallery does not hold up a finished one', async () => {
|
||||
const busy = await seedEvent('drain-busy');
|
||||
const done = await seedEvent('drain-done');
|
||||
await insertPhoto(busy, 'pending');
|
||||
await insertPhoto(done, 'done');
|
||||
faceQueue.touchedEvents.add(busy);
|
||||
faceQueue.touchedEvents.add(done);
|
||||
|
||||
const spy = jest.spyOn(clustering, 'consolidate').mockResolvedValue([]);
|
||||
await faceQueue.drainConsolidation();
|
||||
|
||||
expect(spy).toHaveBeenCalledTimes(1);
|
||||
expect(spy).toHaveBeenCalledWith(done);
|
||||
expect(faceQueue.touchedEvents.has(busy)).toBe(true);
|
||||
});
|
||||
});
|
||||
Reference in New Issue
Block a user