advertise over beacon v2 as metamesh.enrich/<id> so meta-sort's Plugins page can find it (Scan -> Add)

src/beacon-advertise.ts

@@ -0,0 +1,188 @@
+/**
+ * Beacon v2 advertiser — makes this plugin findable by meta-sort's Plugins
+ * page (Scan → Add) without anyone typing its URL.
+ *
+ * Advertises one resource, cap `metamesh.enrich/<manifest id>`, on the beacon
+ * group (239.255.99.1:9099): unsolicited every 10 s, and unicast in reply to
+ * any probe whose `want` matches. Says `bye` on shutdown. It keeps no view of
+ * other nodes — a plugin only needs to be found. Spec:
+ * meta-root docs/project-architecture/beacon-v2.md.
+ *
+ * Env:
+ *   ENABLE_UDP_DISCOVERY  false/0/no/off disables
+ *   BEACON_ADVERTISE_URL  endpoints.http (default http://$HOSTNAME:$PORT)
+ *   BEACON_BINDS          the one meta-sort instance allowed to list it
+ *                         (meta-sort sets this on containers it spawns)
+ *   BEACON_GROUP / BEACON_PORT / BEACON_INTERVAL_MS
+ *
+ * ⚠ MIRRORED FILE. Byte-identical copies live in every TypeScript
+ * metamesh-plugin-*; see meta-root `scripts/check-mirrors.sh`. Never fatal: a
+ * host that blocks multicast logs a warning and the plugin serves on.
+ */
+
+import dgram from 'dgram';
+import { hostname, networkInterfaces } from 'os';
+
+const PROTO = 'beacon';
+const VERSION = 2;
+
+interface Resource {
+    id: string;
+    caps: string[];
+    endpoints: Record<string, string>;
+    binds?: string;
+}
+
+interface NodeInfo {
+    name: string;
+    instance: string;
+    version?: string;
+}
+
+function envNum(key: string): number | undefined {
+    const n = Number(process.env[key]);
+    return Number.isFinite(n) && n > 0 ? n : undefined;
+}
+
+function enabled(): boolean {
+    const v = (process.env.ENABLE_UDP_DISCOVERY ?? '').trim().toLowerCase();
+    return !['false', '0', 'no', 'off'].includes(v);
+}
+
+/** IPv4 addresses of every up, non-internal interface. */
+function interfaceAddresses(): string[] {
+    const out: string[] = [];
+    for (const list of Object.values(networkInterfaces())) {
+        for (const ni of list ?? []) {
+            const v4 = (ni as { family: string | number }).family === 'IPv4' || (ni as { family: string | number }).family === 4;
+            if (v4 && !ni.internal && ni.address) out.push(ni.address);
+        }
+    }
+    return out;
+}
+
+/** Same matching rules as the spec: `*`, `x/*`, optional `@contract`. */
+function capMatches(pattern: string, cap: string): boolean {
+    if (pattern === '*') return true;
+    const split = (s: string): [string, string | null] => {
+        const i = s.lastIndexOf('@');
+        return i >= 0 ? [s.slice(0, i), s.slice(i + 1)] : [s, null];
+    };
+    const [pb, pc] = split(pattern);
+    const [cb, cc] = split(cap);
+    if (pc !== null && cc !== pc) return false;
+    if (pb.endsWith('/*')) {
+        const prefix = pb.slice(0, -2);
+        const rest = cb.startsWith(prefix) ? cb.slice(prefix.length) : '';
+        return rest.startsWith('/') && rest.length > 1;
+    }
+    return pb === cb;
+}
+
+let socket: dgram.Socket | null = null;
+let timer: NodeJS.Timeout | null = null;
+let node: NodeInfo | null = null;
+let resource: Resource | null = null;
+let ifaces: string[] = [];
+let group = '239.255.99.1';
+let udpPort = 9099;
+
+function send(msg: object, dest?: { address: string; port: number }): void {
+    const s = socket;
+    if (!s) return;
+    const body = Buffer.from(JSON.stringify(msg));
+    if (dest) {
+        s.send(body, dest.port, dest.address, () => {});
+        return;
+    }
+    // Once per interface: the default route alone reaches one network.
+    for (const addr of ifaces) {
+        try {
+            s.setMulticastInterface(addr);
+            s.send(body, udpPort, group, () => {});
+        } catch (e) {
+            console.warn(`[beacon] send on ${addr} failed:`, e);
+        }
+    }
+}
+
+function advertise(dest?: { address: string; port: number }): void {
+    if (!node || !resource) return;
+    send({ proto: PROTO, v: VERSION, type: 'advertise', node, resources: [resource] }, dest);
+}
+
+/**
+ * Start advertising this plugin. Call once the HTTP server is listening, so a
+ * consumer that hears it can reach it. Idempotent.
+ */
+export function startBeacon(pluginId: string, version: string, httpPort: number): void {
+    if (socket || !enabled()) return;
+    group = process.env.BEACON_GROUP || group;
+    udpPort = envNum('BEACON_PORT') ?? udpPort;
+    const intervalMs = envNum('BEACON_INTERVAL_MS') ?? 10_000;
+    const instance = process.env.HOSTNAME || hostname();
+    const url = (process.env.BEACON_ADVERTISE_URL || `http://${instance}:${httpPort}`).replace(/\/+$/, '');
+    const binds = (process.env.BEACON_BINDS || '').trim();
+
+    node = { name: `metamesh-plugin-${pluginId}`, instance, ...(version ? { version } : {}) };
+    resource = {
+        id: pluginId,
+        caps: [`metamesh.enrich/${pluginId}`],
+        endpoints: { http: url, manifest: '/manifest' },
+        ...(binds ? { binds } : {}),
+    };
+
+    const s = dgram.createSocket({ type: 'udp4', reuseAddr: true });
+    socket = s;
+    s.on('error', (err) => console.warn('[beacon] socket error:', err.message));
+    s.on('message', (buf, rinfo) => {
+        let m: { proto?: string; v?: number; type?: string; from?: string; want?: string[] };
+        try {
+            m = JSON.parse(buf.toString('utf8'));
+        } catch {
+            return; // foreign traffic on the shared group
+        }
+        if (m?.proto !== PROTO || m.v !== VERSION || m.type !== 'probe' || m.from === instance) return;
+        const want = Array.isArray(m.want) ? m.want : [];
+        if (want.length === 0 || resource!.caps.some((c) => want.some((p) => capMatches(p, c)))) {
+            advertise({ address: rinfo.address, port: rinfo.port });
+        }
+    });
+    s.bind(udpPort, '0.0.0.0', () => {
+        try {
+            s.setMulticastTTL(1);
+        } catch {
+            /* best effort */
+        }
+        ifaces = interfaceAddresses();
+        for (const addr of ifaces) {
+            try {
+                s.addMembership(group, addr);
+            } catch (e) {
+                console.warn(`[beacon] join ${group} on ${addr} failed:`, e);
+            }
+        }
+        console.log(`[beacon] advertising metamesh.enrich/${pluginId} at ${url} on ${group}:${udpPort}`);
+        advertise();
+        timer = setInterval(() => advertise(), intervalMs);
+        timer.unref?.();
+    });
+}
+
+/** Say `bye` so consumers drop this plugin at once. */
+export async function stopBeacon(): Promise<void> {
+    if (!socket || !node) return;
+    if (timer) clearInterval(timer);
+    send({ proto: PROTO, v: VERSION, type: 'bye', node });
+    // Let the bye datagrams leave before the socket closes under them.
+    await new Promise((r) => setTimeout(r, 50));
+    const s = socket;
+    socket = null;
+    await new Promise<void>((resolve) => {
+        try {
+            s.close(() => resolve());
+        } catch {
+            resolve();
+        }
+    });
+}

src/index.ts

@@ -24,6 +24,7 @@
 import Fastify from 'fastify';
 import type { HealthResponse, ProcessRequest, ProcessResponse, CallbackPayload, ConfigureRequest, ConfigureResponse } from './types.js';
 import { manifest, process as processFile } from './plugin.js';
+import { startBeacon, stopBeacon } from './beacon-advertise.js';
 
 const app = Fastify({ logger: true });
 let ready = false;
@@ -47,5 +48,5 @@ app.post<{ Body: ProcessRequest }>('/process', async (request, reply) => {
 });
 
 const port = parseInt(process.env.PORT || '8080', 10);
-app.listen({ port, host: '0.0.0.0' }).then(() => { ready = true; console.log(`[${manifest.id}] Listening on port ${port}`); });
-process.on('SIGTERM', async () => { ready = false; await app.close(); process.exit(0); });
+app.listen({ port, host: '0.0.0.0' }).then(() => { startBeacon(manifest.id, manifest.version, port); ready = true; console.log(`[${manifest.id}] Listening on port ${port}`); });
+process.on('SIGTERM', async () => { await stopBeacon(); ready = false; await app.close(); process.exit(0); });

src/plugin.ts

@@ -21,11 +21,17 @@
  * Matroska stream tags already carry. A `fra` member would be dropped by a
  * `languages:fre` filter.
  *
- * ⚠ Audio only. Consumers read `languages/*` as the AUDIO-language set
- * (meta-watch `cards::audio_languages`); subtitle-track languages belong to
- * `subtitleLanguages/*` (subtitle-extractor). Title keys are not a source
- * either: `titles/<lang>` names the language of a TITLE (tmdb files
- * `originalTitle` under `original_language`), not of the file's audio.
+ * ⚠ Audio only as a SOURCE — but it writes two keys, not one. This plugin reads
+ * audio tracks, so the codes it finds are audio languages and land in
+ * `audioLanguages/<lang3>`; every one of them is ALSO written to
+ * `languages/<lang3>`, the union that queries filter on (METADATA_KEYS §9,
+ * rule #7 — "a union field is written, never computed"). Nothing reconciles the
+ * two after the fact, so a record with the split set and the union unset is
+ * invisible to every language filter on every peer. Subtitle-track languages
+ * belong to `subtitleLanguages/*` (subtitle-extractor), which owes the union
+ * member for the same reason. Title keys are not a source either:
+ * `titles/<lang>` names the language of a TITLE (tmdb files `originalTitle`
+ * under `original_language`), not of the file's audio.
  *
  * ⚠ No `titles/*` write. The old `titles/{firstAudioLang || 'eng'}` =
  * `originalTitle` guess overwrote better identity (tmdb, jellyfin-nfo,
@@ -41,8 +47,8 @@ import { MetaCoreClient } from './meta-core-client.js';
 export const manifest: PluginManifest = {
     id: 'language',
     name: 'Language Aggregator',
-    version: '1.1.0',
-    description: 'Adds audio-track languages to the languages/<lang3> key-set',
+    version: '1.2.0',
+    description: 'Adds audio-track languages to the audioLanguages/<lang3> and languages/<lang3> key-sets',
     author: 'MetaMesh',
     dependencies: ['ffmpeg'],
     priority: 40,
@@ -50,6 +56,7 @@ export const manifest: PluginManifest = {
     defaultQueue: 'fast',
     timeout: 30000,
     schema: {
+        audioLanguages: { label: 'Audio languages', type: 'array', readonly: true },
         languages: { label: 'Languages', type: 'array', readonly: true },
     },
     config: {},
@@ -170,23 +177,41 @@ export function audioLanguagesFromMeta(existingMeta: Record<string, unknown> | u
 }
 
 /**
- * Whether the record already carries `languages/<code>`, in either shape: the
- * flat key, or the nested `languages: { <code>: ... }` object meta-sort's
- * reconstructed document holds.
+ * Whether the record already carries `<field>/<code>`, in either shape: the flat
+ * key, or the nested `<field>: { <code>: ... }` object meta-sort's reconstructed
+ * document holds.
  */
-function hasMember(existingMeta: Record<string, unknown> | undefined, code: string): boolean {
+function hasMember(
+    existingMeta: Record<string, unknown> | undefined,
+    field: string,
+    code: string,
+): boolean {
     if (!existingMeta) return false;
-    if (existingMeta[`languages/${code}`] !== undefined) return true;
-    const nested = existingMeta['languages'];
+    if (existingMeta[`${field}/${code}`] !== undefined) return true;
+    const nested = existingMeta[field];
     return !!nested && typeof nested === 'object' && !Array.isArray(nested)
         && (nested as Record<string, unknown>)[code] !== undefined;
 }
 
-/** The `languages/<code>` members to add — never one the record already has. */
+/**
+ * The members to add — never one the record already has.
+ *
+ * Each audio code is written TWICE: once to `audioLanguages/<code>` (what this
+ * plugin actually observed — a spoken track) and once to `languages/<code>`
+ * (the union every query filters on). METADATA_KEYS §9 rule #7 makes the second
+ * write mandatory, not a convenience: the union is stored, never computed, and
+ * a query cannot ask for `audioLanguages OR subtitleLanguages` — the gateway
+ * wire ANDs across keys, so an OR of the splits would return strictly less.
+ */
 export function languageWrites(existingMeta: Record<string, unknown> | undefined): Record<string, string> {
     const writes: Record<string, string> = {};
     for (const code of audioLanguagesFromMeta(existingMeta)) {
-        if (!hasMember(existingMeta, code)) writes[`languages/${code}`] = 'true';
+        if (!hasMember(existingMeta, 'audioLanguages', code)) {
+            writes[`audioLanguages/${code}`] = 'true';
+        }
+        if (!hasMember(existingMeta, 'languages', code)) {
+            writes[`languages/${code}`] = 'true';
+        }
     }
     return writes;
 }

test/language.test.ts

@@ -164,18 +164,47 @@ describe('audioLanguagesFromMeta', () => {
 });
 
 describe('languageWrites', () => {
-    it('skips members already present in the flat or nested form', () => {
-        expect(languageWrites(nestedDoc({ 'languages/jpn': 'true' }))).toEqual({ 'languages/eng': 'true' });
-        expect(languageWrites(nestedDoc({ languages: { eng: 'true', jpn: true } }))).toEqual({});
+    it('writes each audio code to BOTH the split and the union (METADATA_KEYS §9 rule #7)', () => {
+        expect(languageWrites(nestedDoc())).toEqual({
+            'audioLanguages/jpn': 'true',
+            'languages/jpn': 'true',
+            'audioLanguages/eng': 'true',
+            'languages/eng': 'true',
+        });
+    });
+
+    it('skips members already present in the flat or nested form, per field', () => {
+        // The union is already there; the split is not — the record still owes
+        // the split, and vice versa. The two are tracked independently.
+        expect(languageWrites(nestedDoc({ 'languages/jpn': 'true' }))).toEqual({
+            'audioLanguages/jpn': 'true',
+            'audioLanguages/eng': 'true',
+            'languages/eng': 'true',
+        });
+        expect(languageWrites(nestedDoc({ languages: { eng: 'true', jpn: true } }))).toEqual({
+            'audioLanguages/jpn': 'true',
+            'audioLanguages/eng': 'true',
+        });
+        expect(
+            languageWrites(nestedDoc({ audioLanguages: { eng: 'true', jpn: 'true' }, languages: { eng: 'true', jpn: 'true' } })),
+        ).toEqual({});
     });
 });
 
 describe('process', () => {
-    it('writes the audio languages/* key-set from the nested payload in one merge', async () => {
+    it('writes the audio split AND the union from the nested payload in one merge', async () => {
         const { callback, writer } = await run(nestedDoc());
         expect(callback?.status).toBe('completed');
         expect(writer.calls).toEqual([
-            { hashId: 'cid-1', metadata: { 'languages/jpn': 'true', 'languages/eng': 'true' } },
+            {
+                hashId: 'cid-1',
+                metadata: {
+                    'audioLanguages/jpn': 'true',
+                    'languages/jpn': 'true',
+                    'audioLanguages/eng': 'true',
+                    'languages/eng': 'true',
+                },
+            },
         ]);
     });
 
@@ -183,14 +212,27 @@ describe('process', () => {
         const flat = Object.fromEntries(REAL_STREAMS.map((s, i) => [`stream/${i}`, s]));
         const { callback, writer } = await run(flat);
         expect(callback?.status).toBe('completed');
-        expect(writer.calls[0].metadata).toEqual({ 'languages/jpn': 'true', 'languages/eng': 'true' });
+        expect(writer.calls[0].metadata).toEqual({
+            'audioLanguages/jpn': 'true',
+            'languages/jpn': 'true',
+            'audioLanguages/eng': 'true',
+            'languages/eng': 'true',
+        });
     });
 
-    it('never writes titles/* or any non-languages key, even with originalTitle and title keys', async () => {
+    it('every split member has a matching union member — the record is never left half-written', async () => {
+        const { writer } = await run(nestedDoc());
+        const keys = writer.calls.flatMap((c) => Object.keys(c.metadata));
+        for (const k of keys.filter((k) => k.startsWith('audioLanguages/'))) {
+            expect(keys).toContain(`languages/${k.slice('audioLanguages/'.length)}`);
+        }
+    });
+
+    it('never writes titles/* or any non-language key, even with originalTitle and title keys', async () => {
         const { writer } = await run(nestedDoc({ 'titles/eng/Better Title': 'true', titles: { jpn: { '番組': 'true' } } }));
         const keys = writer.calls.flatMap((c) => Object.keys(c.metadata));
         expect(keys.length).toBeGreaterThan(0);
-        expect(keys.every((k) => k.startsWith('languages/'))).toBe(true);
+        expect(keys.every((k) => k.startsWith('languages/') || k.startsWith('audioLanguages/'))).toBe(true);
         expect(keys.some((k) => k.startsWith('titles'))).toBe(false);
     });
 
@@ -209,10 +251,23 @@ describe('process', () => {
     });
 
     it('is idempotent: a re-run over a record that already has the members writes nothing', async () => {
-        const { writer } = await run(nestedDoc({ 'languages/jpn': 'true', 'languages/eng': 'true' }));
+        const { writer } = await run(nestedDoc({
+            'audioLanguages/jpn': 'true', 'audioLanguages/eng': 'true',
+            'languages/jpn': 'true', 'languages/eng': 'true',
+        }));
         expect(writer.calls).toEqual([]);
     });
 
+    it('backfills the union on a record written before the split existed', async () => {
+        // Pre-1.2.0 records carry only `languages/*`. A re-run must add the
+        // split without disturbing the union that is already correct.
+        const { writer } = await run(nestedDoc({ 'languages/jpn': 'true', 'languages/eng': 'true' }));
+        expect(writer.calls[0].metadata).toEqual({
+            'audioLanguages/jpn': 'true',
+            'audioLanguages/eng': 'true',
+        });
+    });
+
     it('fails the task when the write is rejected', async () => {
         const { callback } = await run(nestedDoc(), new RecordingWriter(false));
         expect(callback?.status).toBe('failed');