8 Commits
Author SHA1 Message Date
Worph d7b395b774 CI: push to ipcr-push-85-17-246-67.ip.nsl.sh (the box's cert no longer covers nip.io) [skip ci] 2026-11-04 14:31:45 +00:00
Worph f3ab7f0d17 sign every write as an MMUR2 claim (POST /api/claims)
Build and Push Docker Image / build (push) Successful in 1m9s
2026-10-25 17:37:03 +00:00
Manielle bae3038230 CI accepts four-part tags (v1.2.3.1) for fixes
Build and Push Docker Image / build (push) Successful in 1m12s
2026-10-25 14:22:18 +01:00
Worph c1eb196448 gitea workflow: push to IPCR's direct host (Cloudflare 100 MB cap)
Build and Push Docker Image / build (push) Successful in 1m17s
2026-10-13 10:03:21 +00:00
Worph 4f4b2100d2 gitea workflow: publish to IPCR (metadec.eth) instead of ghcr
Build and Push Docker Image / build (push) Failing after 1m14s
2026-10-12 18:31:39 +00:00
Worph 50772312e6 advertise over beacon v2 as metamesh.enrich/<id> so meta-sort's Plugins page can find it (Scan -> Add)
Build and Push Docker Image / build (push) Failing after 14s
2026-10-04 17:19:21 +00:00
Worph d1cef26c3a gitattributes
Build and Push Docker Image / build (push) Failing after 13s
2026-09-13 15:04:01 +00:00
Worph 5c8385c59b align plugin with new architecture
Build and Push Docker Image / build (push) Failing after 13s
2026-09-13 11:53:05 +02:00
9 changed files with 1123 additions and 109 deletions
+6
View File
@@ -0,0 +1,6 @@
# FIX CRLF always when developer has not set
# Linux/Mac: git config --global core.autocrlf input
# Windows: git config --global core.autocrlf true
# Auto detect text files and perform LF normalization
* text=auto
* eol=lf
+85
View File
@@ -0,0 +1,85 @@
# Gitea twin of .github/workflows/docker.yml. Gitea runs only .gitea/workflows/ once this
# folder exists, and GitHub never reads it.
#
# Publishes to IPCR on watch.nsl.sh instead of ghcr.io:
# push ipcr-push-85-17-246-67.ip.nsl.sh/metadec.eth/<repo>:<tag> docker login as metadec.eth with the
# IPCR_PUSH_TOKEN secret (the metadec.eth namespace token from the IPCR dashboard)
# pull ipcr-watch.nsl.sh/ipns/metadec.eth/metadec.eth/<repo>:<tag> anonymous
# The push goes to ipcr-push-85-17-246-67.ip.nsl.sh, the box's own address, not ipcr-push-watch.nsl.sh: Cloudflare fronts
# *.watch.nsl.sh and refuses any request over 100 MB (413), which a large image layer is.
# That host's certificate is signed by the box's private Mesh Router CA, given to buildkit from
# the IPCR_CA_PEM repo variable. docker login would make the job's Docker daemon check the
# certificate without that CA, so the login is written straight into ~/.docker/config.json.
# The runner's job image (node:22-bookworm) has no docker CLI, so the static one is
# installed first; buildx installs itself. No `type=gha` cache: the runner has none.
name: Build and Push Docker Image
on:
push:
branches: [main, master]
tags: ['v*']
pull_request:
branches: [main, master]
env:
REGISTRY: ipcr-push-85-17-246-67.ip.nsl.sh
IMAGE_NAME: metadec.eth/${{ github.event.repository.name }}
jobs:
build:
runs-on: ubuntu-latest
permissions:
contents: read
packages: write
steps:
- name: Checkout repository
uses: actions/checkout@v4
- name: Install Docker CLI
run: curl -fsSL https://download.docker.com/linux/static/stable/x86_64/docker-28.5.1.tgz | tar -xz -C /usr/local/bin --strip-components=1 docker/docker
- name: Trust the Mesh Router CA
run: printf '%s\n' "$IPCR_CA_PEM" > /tmp/ipcr-ca.pem
env:
IPCR_CA_PEM: ${{ vars.IPCR_CA_PEM }}
- name: Set up Docker Buildx
uses: docker/setup-buildx-action@v3
with:
version: latest
buildkitd-config-inline: |
[registry."${{ env.REGISTRY }}"]
ca = ["/tmp/ipcr-ca.pem"]
- name: Log in to IPCR
if: github.event_name != 'pull_request'
env:
IPCR_PUSH_TOKEN: ${{ secrets.IPCR_PUSH_TOKEN }}
run: |
mkdir -p ~/.docker
printf '{"auths":{"%s":{"auth":"%s"}}}\n' "$REGISTRY" "$(printf 'metadec.eth:%s' "$IPCR_PUSH_TOKEN" | base64 -w0)" > ~/.docker/config.json
- name: Extract metadata
id: meta
uses: docker/metadata-action@v5
with:
images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
tags: |
type=ref,event=branch
type=ref,event=pr
type=semver,pattern={{version}}
# A fix to release 1.2.3 is tagged v1.2.3.1: four numbers are not semver, so
# the pattern above makes no tag for it; this one does (image tag 1.2.3.1).
type=match,pattern=v(\d+\.\d+\.\d+\.\d+),group=1
type=semver,pattern={{major}}.{{minor}}
type=sha
- name: Build and push
uses: docker/build-push-action@v5
with:
context: .
push: ${{ github.event_name != 'pull_request' }}
tags: ${{ steps.meta.outputs.tags }}
labels: ${{ steps.meta.outputs.labels }}
+3
View File
@@ -42,6 +42,9 @@ jobs:
type=ref,event=branch
type=ref,event=pr
type=semver,pattern={{version}}
# A fix to release 1.2.3 is tagged v1.2.3.1: four numbers are not semver, so
# the pattern above makes no tag for it; this one does (image tag 1.2.3.1).
type=match,pattern=v(\d+\.\d+\.\d+\.\d+),group=1
type=semver,pattern={{major}}.{{minor}}
type=sha
+37 -7
View File
@@ -38,14 +38,28 @@ in `src/plugin.ts`.
## How a frame is chosen
1. Candidate offsets at 20 / 45 / 70 % of `fileinfo/duration`, clamped 2s away
from both ends (the head is logos and black, the tail is credits).
2. At each offset, one ffmpeg run with **input seeking** (`-ss` before `-i`, so
it Range-seeks over WebDAV instead of decoding from zero) and
`thumbnail=60`, which picks the most representative of the next 60 frames.
3. The encoded JPEG is measured (mean + stddev of a 32×32 grayscale reduction).
1. **Record first, file second.** `fileinfo/duration` and the stream table
from the record; if either is missing, one header-only `ffprobe` reads it
from the file (a record with no stream data at all has been seen on the dev
stack). meta-sort treats a `failed` or
`skipped` ffmpeg dependency as complete, so a missing duration is a normal
case, not a bug. Only if the probe fails too does it fall back to a blind
10s then 0s — and that fallback lands on distributor idents and opening
credits, so it is a last resort.
2. Candidate offsets at 20 / 45 / 70 % of the duration, clamped 2s away from
both ends (the head is logos and black, the tail is credits).
3. At each offset, one ffmpeg run with **input seeking** (`-ss` before `-i`, so
it Range-seeks over WebDAV instead of decoding from zero), scaled down
**before** `thumbnail=60` picks the most representative of the next 60
frames. That order matters: `thumbnail` buffers all 60 frames, and at full
1080p a single grab peaked at 409 MiB versus 196 MiB scaled-first.
4. The encoded JPEG is measured (mean + stddev of a 32×32 grayscale reduction).
Black, blown-out and flat frames are rejected and the next offset is tried.
4. All candidates rejected → `skipped`, no field written.
5. All candidates rejected → `skipped`, no field written.
Extractions run **one at a time** by default (`maxConcurrent`). meta-sort's
background queue sends several tasks at once, and four concurrent 1080p grabs
OOM-killed the 512 MiB container; with one CPU, parallel decodes buy nothing.
The primary video stream is the largest one that **looks like moving video** —
not `0:v:0`, and not simply the largest. ffprobe reports attached cover art as
@@ -68,6 +82,7 @@ curl -k -X PUT https://metasort-dev.localhost:8180/api/plugins/still-extractor/c
| `maxWidth` | `640` | Frame is scaled down to at most this width |
| `jpegQuality` | `4` | ffmpeg `-q:v`; lower is better quality |
| `frameTimeoutMs` | `120000` | Hard kill for one ffmpeg grab |
| `maxConcurrent` | `1` | Extractions allowed to run at once in this instance |
## Development
@@ -80,3 +95,18 @@ docker build -t metamesh-plugin-still-extractor:main .
Depends on `file-info` (for `fileType`) and `ffmpeg` (for `fileinfo/duration`
and the `stream/{n}` table). Runs on the **background** queue.
### What meta-sort actually sends
⚠ The `existingMeta` in a `/process` request is the **nested document form** —
`fileinfo: { duration }`, `stream: [...]`, `cids: [...]` — not the flat
`fileinfo/duration` / `stream/{n}` keys you see in meta-core's flat record.
Reading only the flat key made every task fall back to 10s/0s on the dev stack.
`durationFromMeta` and `selectPrimaryVideoStream` accept both shapes; keep it
that way.
A replay or rescan queues several tasks for the same file. Each reads the
record at dispatch, so the plugin re-checks `still` in meta-core once it holds
an extraction slot, instead of recomputing the same frame for every duplicate.
A file that yields no usable frame is remembered in memory for an hour, so its
queued duplicates skip instead of decoding it again; a later rescan retries.
+188
View File
@@ -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();
}
});
}
+3
View File
@@ -19,6 +19,7 @@
import Fastify from 'fastify';
import type { HealthResponse, ProcessRequest, ProcessResponse, CallbackPayload, ConfigureRequest, ConfigureResponse } from './types.js';
import { manifest, process as processFile, configure } from './plugin.js';
import { startBeacon, stopBeacon } from './beacon-advertise.js';
const app = Fastify({ logger: true });
let ready = false;
@@ -70,11 +71,13 @@ 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(() => {
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);
+343 -56
View File
@@ -1,93 +1,380 @@
/**
* meta-core API Client for writing metadata
* Fails gracefully if meta-core is unavailable (logs warning, returns without throwing)
* meta-core client: signed claim writes (MMUR2) + plain reads.
*
* MIRRORED byte-for-byte across every `metamesh-plugin-*` TypeScript plugin and
* meta-sort-core (`src/kv/meta-core-client.ts`) — guarded by
* `scripts/check-mirrors.sh`. Edit one copy, then copy it everywhere.
*
* Every write is a vote signed by THIS container's key
* (docs/implementation/metacore-claims-model.md §3, §9):
*
* bytes_to_sign = "MMUR2\n" || canonical CBOR [cid, uid, key, value, ts]
* sig = ECDSA-secp256k1(SHA-256(bytes)), compact r||s, low-S
*
* and is posted to `POST {baseUrl}/api/claims` — meta-core directly, or
* meta-sort's pass-through relay when the plugin runs under meta-sort. The
* encoding is pinned by meta-core's `internal/claim/testdata/claim-vectors.json`
* (vendored as `test/claim-vectors.json`).
*
* Key: `METAMESH_SIGNING_KEY` (64 hex chars) or `METAMESH_SIGNING_KEY_FILE`
* (read; generated and written 0600 if missing). With neither, writes are
* refused (logged, return false) — no silent throwaway identities.
*
* Fails gracefully if meta-core is unavailable: logs and returns, never throws.
*/
import * as crypto from 'crypto';
import * as fs from 'fs';
import * as path from 'path';
// ---------------------------------------------------------------- encoding --
const DOMAIN = Buffer.from('MMUR2\n', 'utf8');
const B58 = '123456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefghijkmnopqrstuvwxyz';
const N = BigInt('0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFEBAAEDCE6AF48A03BBFD25E8CD0364141');
const HALF_N = N >> 1n;
/** A claim value: text (public), bytes (private envelope) or null (retraction). */
export type ClaimValue = string | Uint8Array | null;
export interface SignedClaim {
cid: string;
uid: string;
key: string;
value: ClaimValue;
ts: number;
sig: Uint8Array;
}
function head(major: number, n: number | bigint): Buffer {
const m = major << 5;
const v = BigInt(n);
if (v < 24n) return Buffer.from([m | Number(v)]);
if (v <= 0xffn) return Buffer.from([m | 24, Number(v)]);
if (v <= 0xffffn) {
const b = Buffer.alloc(3);
b[0] = m | 25;
b.writeUInt16BE(Number(v), 1);
return b;
}
if (v <= 0xffffffffn) {
const b = Buffer.alloc(5);
b[0] = m | 26;
b.writeUInt32BE(Number(v), 1);
return b;
}
const b = Buffer.alloc(9);
b[0] = m | 27;
b.writeBigUInt64BE(v, 1);
return b;
}
function text(s: string): Buffer {
const raw = Buffer.from(s, 'utf8');
return Buffer.concat([head(3, raw.length), raw]);
}
/** The exact bytes an MMUR2 signature covers. */
export function signedBytes(cid: string, uid: string, key: string, value: ClaimValue, ts: number): Buffer {
let v: Buffer;
if (value === null) v = Buffer.from([0xf6]);
else if (typeof value === 'string') v = text(value);
else v = Buffer.concat([head(2, value.length), Buffer.from(value)]);
return Buffer.concat([DOMAIN, head(4, 5), text(cid), text(uid), text(key), v, head(0, ts)]);
}
function base58Encode(bytes: Uint8Array): string {
let n = BigInt('0x' + (Buffer.from(bytes).toString('hex') || '0'));
let out = '';
while (n > 0n) {
out = B58[Number(n % 58n)] + out;
n /= 58n;
}
for (const b of bytes) {
if (b !== 0) break;
out = '1' + out;
}
return out;
}
function base58Decode(s: string): Buffer {
let n = 0n;
for (const ch of s) {
const i = B58.indexOf(ch);
if (i < 0) throw new Error(`invalid base58 character ${ch}`);
n = n * 58n + BigInt(i);
}
let hex = n === 0n ? '' : n.toString(16);
if (hex.length % 2) hex = '0' + hex;
let zeros = 0;
while (zeros < s.length && s[zeros] === '1') zeros++;
return Buffer.concat([Buffer.alloc(zeros), Buffer.from(hex, 'hex')]);
}
function b64url(b: Buffer): string {
return b.toString('base64url');
}
function publicKeyFromCompressed(compressed: Buffer): crypto.KeyObject {
const full = crypto.ECDH.convertKey(compressed, 'secp256k1', undefined, undefined, 'uncompressed') as Buffer;
return crypto.createPublicKey({
key: { kty: 'EC', crv: 'secp256k1', x: b64url(full.subarray(1, 33)), y: b64url(full.subarray(33, 65)) },
format: 'jwk',
});
}
/** Rewrite a compact signature to its low-S form (MMUR2 refuses high-S). */
function lowS(sig: Buffer): Buffer {
const s = BigInt('0x' + sig.subarray(32).toString('hex'));
if (s <= HALF_N) return sig;
const fixed = Buffer.from((N - s).toString(16).padStart(64, '0'), 'hex');
return Buffer.concat([sig.subarray(0, 32), fixed]);
}
/** True when the claim's signature is a valid low-S signature by its uid. */
export function verifyClaim(c: SignedClaim): boolean {
try {
const sig = Buffer.from(c.sig);
if (sig.length !== 64) return false;
if (BigInt('0x' + sig.subarray(32).toString('hex')) > HALF_N) return false;
if (!c.uid.startsWith('z')) return false;
const pub = base58Decode(c.uid.slice(1));
if (pub.length !== 33) return false;
return crypto.verify(
'sha256',
signedBytes(c.cid, c.uid, c.key, c.value, c.ts),
{ key: publicKeyFromCompressed(pub), dsaEncoding: 'ieee-p1363' },
sig,
);
} catch {
return false;
}
}
// ----------------------------------------------------------------- signer --
export class ClaimSigner {
readonly uid: string;
private readonly key: crypto.KeyObject;
private lastTs = 0;
constructor(privateKeyHex: string) {
const hex = privateKeyHex.trim().replace(/^0x/i, '');
if (!/^[0-9a-fA-F]{64}$/.test(hex)) throw new Error('signing key must be 64 hex characters');
const d = Buffer.from(hex, 'hex');
const ecdh = crypto.createECDH('secp256k1');
ecdh.setPrivateKey(d);
const full = ecdh.getPublicKey(null, 'uncompressed');
this.uid = 'z' + base58Encode(ecdh.getPublicKey(null, 'compressed'));
this.key = crypto.createPrivateKey({
key: {
kty: 'EC', crv: 'secp256k1', d: b64url(d),
x: b64url(full.subarray(1, 33)), y: b64url(full.subarray(33, 65)),
},
format: 'jwk',
});
}
/** Milliseconds, strictly increasing within this process (ts is the version). */
nextTs(): number {
this.lastTs = Math.max(Date.now(), this.lastTs + 1);
return this.lastTs;
}
sign(cid: string, key: string, value: ClaimValue, ts: number = this.nextTs()): SignedClaim {
const raw = crypto.sign('sha256', signedBytes(cid, this.uid, key, value, ts), {
key: this.key,
dsaEncoding: 'ieee-p1363',
});
return { cid, uid: this.uid, key, value, ts, sig: lowS(raw) };
}
}
let cachedSigner: ClaimSigner | null | undefined;
/**
* The process signer from the environment, or null when no key is configured.
* Read lazily on first write; `resetSignerForTests()` forgets it.
*/
export function loadSigner(): ClaimSigner | null {
if (cachedSigner !== undefined) return cachedSigner;
const inline = process.env.METAMESH_SIGNING_KEY;
const file = process.env.METAMESH_SIGNING_KEY_FILE;
try {
if (inline) {
cachedSigner = new ClaimSigner(inline);
} else if (file) {
if (!fs.existsSync(file)) {
fs.mkdirSync(path.dirname(file), { recursive: true });
fs.writeFileSync(file, crypto.randomBytes(32).toString('hex') + '\n', { mode: 0o600 });
}
cachedSigner = new ClaimSigner(fs.readFileSync(file, 'utf8'));
} else {
cachedSigner = null;
}
} catch (error) {
console.warn(`[MetaCoreClient] Signing key unusable: ${error instanceof Error ? error.message : String(error)}`);
cachedSigner = null;
}
if (cachedSigner) console.log(`[MetaCoreClient] Signing claims as ${cachedSigner.uid}`);
else console.warn('[MetaCoreClient] No METAMESH_SIGNING_KEY / METAMESH_SIGNING_KEY_FILE — metadata writes are refused');
return cachedSigner;
}
export function resetSignerForTests(): void {
cachedSigner = undefined;
}
/** The §3.4 JSON form. */
export function claimToJSON(c: SignedClaim): Record<string, unknown> {
let value: unknown = null;
if (typeof c.value === 'string') value = c.value;
else if (c.value) value = { b64: Buffer.from(c.value).toString('base64') };
return { cid: c.cid, uid: c.uid, key: c.key, value, ts: c.ts, sig: Buffer.from(c.sig).toString('base64') };
}
// ----------------------------------------------------------------- client --
export class MetaCoreClient {
constructor(private baseUrl: string) {}
private async safeFetch(url: string, options: RequestInit): Promise<Response | null> {
try {
return await fetch(url, { ...options, signal: AbortSignal.timeout(5000) });
return await fetch(url, { ...options, signal: AbortSignal.timeout(10000) });
} catch (error) {
console.warn(`[MetaCoreClient] Warning: meta-core unavailable at ${this.baseUrl}`);
console.warn(`[MetaCoreClient] Warning: meta-core unavailable at ${this.baseUrl}: ${error instanceof Error ? error.message : String(error)}`);
return null;
}
}
async setProperty(hashId: string, key: string, value: string): Promise<void> {
const response = await this.safeFetch(`${this.baseUrl}/meta/${hashId}/${key}`, {
method: 'PUT',
/**
* Sign and post claims about `cid`. True when every claim was accepted (or
* was already stored: `unchanged` / `stale` are not failures).
*/
async writeClaims(cid: string, values: Array<[string, ClaimValue]>): Promise<boolean> {
if (values.length === 0) return true;
const signer = loadSigner();
if (!signer) return false;
const claims = values.map(([key, value]) => claimToJSON(signer.sign(cid, key, value)));
const response = await this.safeFetch(`${this.baseUrl}/api/claims`, {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ value }),
body: JSON.stringify(claims),
});
if (response && !response.ok) {
console.warn(`[MetaCoreClient] Failed to set property: ${response.status}`);
if (!response) return false;
if (!response.ok) {
console.warn(`[MetaCoreClient] Claims for ${cid} rejected: ${response.status}`);
return false;
}
const statuses = await response.json().catch(() => null) as Array<{ status?: string; reason?: string }> | null;
const bad = (statuses ?? []).filter((s) => s.status === 'bad_signature' || s.status === 'bad_request');
if (bad.length > 0) {
console.warn(`[MetaCoreClient] ${bad.length} claim(s) for ${cid} refused: ${bad[0].status} ${bad[0].reason ?? ''}`);
return false;
}
return true;
}
/** One text claim. */
async setProperty(hashId: string, key: string, value: string): Promise<boolean> {
return this.writeClaims(hashId, [[key, String(value)]]);
}
/** One text claim per entry, in a single batch. */
async mergeMetadata(hashId: string, metadata: Record<string, string>): Promise<boolean> {
return this.writeClaims(hashId, Object.entries(metadata).map(([k, v]) => [k, String(v)]));
}
/**
* Retract this signer's own vote for `key`. Other signers' votes for the
* same key are untouched (there are no negative votes).
*/
async deleteProperty(hashId: string, key: string): Promise<boolean> {
return this.writeClaims(hashId, [[key, null]]);
}
/**
* Append a member to a legacy csv-set field (`genres`, `studio`, `tags`,
* `announceList`, …) — the comma-joined shape readers still parse (the
* METADATA_KEYS §14.12 key-set migration is out of scope for phase 1).
*
* Under claims a signer owns only its OWN vote, so this reads this
* signer's current claim for (cid, key) from
* `GET /api/meta/{cid}/claims?key=<key>` (other signers' claims are
* ignored), adds the member if absent and signs the new full value as one
* text claim. Members are kept raw (no escaping), same as the old
* meta-sort `_add` proxy. A member already present writes nothing.
*/
async addToSet(hashId: string, key: string, value: string): Promise<boolean> {
const member = String(value);
if (!member) return true;
const signer = loadSigner();
if (!signer) return false;
const current = await this.ownClaimValue(hashId, key, signer.uid);
if (current === undefined) return false; // could not read: never clobber our own set
const members = current ? current.split(',').map((v) => v.trim()).filter(Boolean) : [];
if (members.includes(member)) return true;
return this.writeClaims(hashId, [[key, [...members, member].join(',')]]);
}
/**
* This signer's newest text claim for (cid, key): the value, `null` when it
* has none, `undefined` when the claims could not be read.
*/
private async ownClaimValue(cid: string, key: string, uid: string): Promise<string | null | undefined> {
const url = `${this.baseUrl}/api/meta/${encodeURIComponent(cid)}/claims?key=${encodeURIComponent(key)}`;
const response = await this.safeFetch(url, { method: 'GET' });
if (!response) return undefined;
if (response.status === 404) return null;
if (!response.ok) {
console.warn(`[MetaCoreClient] Failed to read claims for ${cid}/${key}: ${response.status}`);
return undefined;
}
const body = await response.json().catch(() => null) as unknown;
const list = (Array.isArray(body) ? body : (body as { claims?: unknown })?.claims) as
Array<{ uid?: string; key?: string; value?: unknown; ts?: number }> | undefined;
let best: { value: string; ts: number } | null = null;
for (const c of Array.isArray(list) ? list : []) {
if (c.uid !== uid || c.key !== key || typeof c.value !== 'string') continue;
const ts = Number(c.ts) || 0;
if (!best || ts > best.ts) best = { value: c.value, ts };
}
return best ? best.value : null;
}
async getProperty(hashId: string, key: string): Promise<string | null> {
const response = await this.safeFetch(`${this.baseUrl}/meta/${hashId}/${key}`, { method: 'GET' });
if (!response || response.status === 404) return null;
if (!response.ok) return null;
const data = await response.json() as { value?: string };
return data.value ?? null;
if (!response || !response.ok) return null;
const data = await response.json() as { value?: unknown };
return data.value === undefined || data.value === null ? null : String(data.value);
}
async mergeMetadata(hashId: string, metadata: Record<string, string>): Promise<void> {
const response = await this.safeFetch(`${this.baseUrl}/meta/${hashId}`, {
method: 'PATCH',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify(metadata),
});
if (response && !response.ok) {
console.warn(`[MetaCoreClient] Failed to merge metadata: ${response.status}`);
}
}
async deleteProperty(hashId: string, key: string): Promise<void> {
const response = await this.safeFetch(`${this.baseUrl}/meta/${hashId}/${key}`, { method: 'DELETE' });
if (response && !response.ok && response.status !== 404) {
console.warn(`[MetaCoreClient] Failed to delete property: ${response.status}`);
}
}
async addToSet(hashId: string, key: string, value: string): Promise<void> {
const response = await this.safeFetch(`${this.baseUrl}/meta/${hashId}/_add/${key}`, {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ value }),
});
if (response && !response.ok) {
console.warn(`[MetaCoreClient] Failed to add to set: ${response.status}`);
}
}
async getMetadata(hashId: string): Promise<Record<string, string>> {
/**
* Whole merged record, in whatever form the endpoint serves (meta-sort's
* /meta is nested, meta-core's is flat) — callers flatten. `null` when the
* call itself failed, `{}` when the record does not exist.
*/
async getMetadata(hashId: string): Promise<Record<string, unknown> | null> {
const response = await this.safeFetch(`${this.baseUrl}/meta/${hashId}`, { method: 'GET' });
if (!response || response.status === 404) return {};
if (!response.ok) return {};
const data = await response.json() as { metadata?: Record<string, string> };
if (!response) return null;
if (response.status === 404) return {};
if (!response.ok) return null;
const data = await response.json() as { metadata?: Record<string, unknown> };
return data.metadata ?? {};
}
/**
* Compute the CID (Content Identifier) for a file
* Uses the meta-core /file/cid API endpoint
* Compute the CID of a file under FILES_PATH via meta-core's `POST /file/cid`.
* @param filePath - path relative to FILES_PATH (e.g. "poster/Movie (2024)_poster.jpg")
*/
async computeFileCID(filePath: string): Promise<string | null> {
try {
const response = await this.safeFetch(`${this.baseUrl}/file/cid`, {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ path: filePath }),
});
if (!response || !response.ok) return null;
const data = await response.json() as { cid?: string };
return data.cid ?? null;
} catch (error) {
console.debug(`[MetaCoreClient] Failed to compute CID for ${filePath}:`, error);
return null;
}
const data = await response.json().catch(() => null) as { cid?: string } | null;
return data?.cid ?? null;
}
}
+272 -40
View File
@@ -77,6 +77,7 @@ export const manifest: PluginManifest = {
maxWidth: { type: 'number', label: 'Max Width (px)', default: 640 },
jpegQuality: { type: 'number', label: 'JPEG Quality (ffmpeg -q:v, lower is better)', default: 4 },
frameTimeoutMs: { type: 'number', label: 'Per-Frame ffmpeg Timeout (ms)', default: 120000 },
maxConcurrent: { type: 'number', label: 'Max Concurrent Extractions', default: 1 },
},
};
@@ -86,6 +87,20 @@ let seekPercents = [20, 45, 70];
let maxWidth = 640;
let jpegQuality = 4;
let frameTimeoutMs = 120000;
let maxConcurrent = 1;
/**
* A config value as a number, or undefined when it is absent.
*
* ⚠ Not `Number.isFinite(Number(v))`: `Number(null)` and `Number('')` are both
* `0`, so a config payload carrying `"maxWidth": null` would silently set a
* 0-pixel width or a 0 ms timeout.
*/
function configNumber(v: unknown): number | undefined {
if (v === null || v === undefined || v === '') return undefined;
const n = Number(v);
return Number.isFinite(n) ? n : undefined;
}
export function configure(config: Record<string, unknown>): void {
forceRecompute = config.forceRecompute === true;
@@ -96,15 +111,72 @@ export function configure(config: Record<string, unknown>): void {
.filter((p) => Number.isFinite(p) && p > 0 && p < 100);
if (parsed.length > 0) seekPercents = parsed;
}
if (Number.isFinite(Number(config.maxWidth))) maxWidth = Number(config.maxWidth);
if (Number.isFinite(Number(config.jpegQuality))) jpegQuality = Number(config.jpegQuality);
if (Number.isFinite(Number(config.frameTimeoutMs))) frameTimeoutMs = Number(config.frameTimeoutMs);
const width = configNumber(config.maxWidth);
if (width !== undefined && width > 0) maxWidth = width;
const quality = configNumber(config.jpegQuality);
if (quality !== undefined && quality >= 1 && quality <= 31) jpegQuality = quality;
const timeout = configNumber(config.frameTimeoutMs);
if (timeout !== undefined && timeout > 0) frameTimeoutMs = timeout;
const concurrent = configNumber(config.maxConcurrent);
if (concurrent !== undefined && concurrent >= 1) maxConcurrent = Math.floor(concurrent);
console.log(
`[still-extractor] Config: forceRecompute=${forceRecompute}, seekPercents=${seekPercents.join(',')}, ` +
`maxWidth=${maxWidth}, jpegQuality=${jpegQuality}, frameTimeoutMs=${frameTimeoutMs}`
`maxWidth=${maxWidth}, jpegQuality=${jpegQuality}, frameTimeoutMs=${frameTimeoutMs}, maxConcurrent=${maxConcurrent}`
);
}
let activeExtractions = 0;
const extractionWaiters: Array<() => void> = [];
/**
* Run `fn` once an extraction slot is free.
*
* meta-sort's background queue dispatches several tasks at this single
* instance at once, and each ffmpeg grab of a 1080p file peaks near 200 MiB
* even with the scale-first chain. Four at a time OOM-killed the 512 MiB
* container on the dev stack (`OOMKilled=true`, ffmpeg dying on SIGKILL). The
* container gets one CPU, so running them concurrently buys no throughput
* anyway — the cap lives here so it holds whatever the scheduler sends.
*/
export async function withExtractionSlot<T>(fn: () => Promise<T>): Promise<T> {
while (activeExtractions >= maxConcurrent) {
await new Promise<void>((resolve) => extractionWaiters.push(resolve));
}
activeExtractions++;
try {
return await fn();
} finally {
activeExtractions--;
extractionWaiters.shift()?.();
}
}
/**
* Files that produced no usable frame, remembered briefly by cid.
*
* A failed extraction writes no `still`, so the slot-time re-check cannot stop
* the duplicates a replay or rescan queues behind it — on the dev stack a 17s
* title-card test clip was fully decoded a dozen times in a row. An in-memory
* TTL absorbs that burst without inventing a persisted "tried and failed" key,
* and a rescan after the TTL still gets a fresh attempt.
*/
const NO_USABLE_FRAME_TTL_MS = 60 * 60 * 1000;
const noUsableFrameAt = new Map<string, number>();
export function recentlyFailed(cid: string, now = Date.now()): boolean {
const at = noUsableFrameAt.get(cid);
if (at === undefined) return false;
if (now - at > NO_USABLE_FRAME_TTL_MS) {
noUsableFrameAt.delete(cid);
return false;
}
return true;
}
export function rememberNoUsableFrame(cid: string, now = Date.now()): void {
noUsableFrameAt.set(cid, now);
}
/**
* Compute midhash256 CID from a Buffer (matches meta-hash algorithm).
* Copied verbatim from metamesh-plugin-subtitle-extractor.
@@ -185,6 +257,32 @@ export interface VideoStream {
height?: number;
codec?: string;
frameRate?: string;
/** ffprobe's `attached_pic` disposition. Probe path only — the ffmpeg
* plugin's stream table does not carry dispositions. */
attachedPic?: boolean;
}
/**
* `fileinfo/duration` out of whichever shape the payload carries.
*
* ⚠ meta-sort's /process payload is the NESTED document form — `fileinfo:
* { duration }`, `stream: [...]`, `cids: [...]` — not meta-core's flat
* `fileinfo/duration` keys. On the dev stack every task's existingMeta had
* `fileinfo` and `stream`, never `fileinfo/duration`, and reading only the flat
* key sent 16 of 16 stills to the 10s/0s fallback. Accept both, plus a
* JSON-string `fileinfo`.
*/
export function durationFromMeta(existingMeta: Record<string, unknown> | undefined): number | undefined {
if (!existingMeta) return undefined;
let fileinfo: unknown = existingMeta['fileinfo'];
if (typeof fileinfo === 'string') {
try { fileinfo = JSON.parse(fileinfo); } catch { fileinfo = undefined; }
}
const raw = existingMeta['fileinfo/duration']
?? (fileinfo && typeof fileinfo === 'object' ? (fileinfo as Record<string, unknown>)['duration'] : undefined);
if (raw === null || raw === undefined || raw === '') return undefined;
const n = Number(raw);
return Number.isFinite(n) && n > 0 ? n : undefined;
}
/**
@@ -210,19 +308,21 @@ const STILL_IMAGE_CODECS = new Set(['mjpeg', 'png', 'bmp', 'gif', 'webp', 'tiff'
* within the better class; if nothing looks like moving video we fall back to
* the largest of whatever is there rather than giving up.
*
* Accepts both shapes meta-core hands back: the namespaced `stream/{n}` keys
* the ffmpeg plugin writes, and the flattened `stream` array.
* Accepts both shapes: the nested `stream` collection meta-sort's /process
* payload actually carries (array, JSON string, or index-keyed object), and the
* namespaced `stream/{n}` keys meta-core stores.
*/
export function selectPrimaryVideoStream(existingMeta: Record<string, string> | undefined): VideoStream | null {
if (!existingMeta) return null;
const raw: unknown[] = [];
const flattened = existingMeta['stream'] as unknown;
if (flattened) {
const nested = existingMeta['stream'] as unknown;
if (nested) {
try {
const arr = Array.isArray(flattened) ? flattened : JSON.parse(String(flattened));
if (Array.isArray(arr)) raw.push(...arr);
const parsed = typeof nested === 'string' ? JSON.parse(nested) : nested;
if (Array.isArray(parsed)) raw.push(...parsed);
else if (parsed && typeof parsed === 'object') raw.push(...Object.values(parsed as Record<string, unknown>));
} catch {
// fall through to the namespaced form
}
@@ -234,9 +334,7 @@ export function selectPrimaryVideoStream(existingMeta: Record<string, string> |
}
}
const moving: VideoStream[] = [];
const other: VideoStream[] = [];
const candidates: VideoStream[] = [];
for (const entry of raw) {
let stream: {
type?: string; index?: number; width?: number; height?: number;
@@ -249,22 +347,31 @@ export function selectPrimaryVideoStream(existingMeta: Record<string, string> |
}
if (!stream || stream.type !== 'video' || stream.index == null) continue;
const candidate: VideoStream = {
candidates.push({
index: Number(stream.index),
width: stream.width != null ? Number(stream.width) : undefined,
height: stream.height != null ? Number(stream.height) : undefined,
codec: stream.codec,
frameRate: stream.frameRate,
};
const looksMoving =
!!candidate.frameRate && !STILL_IMAGE_CODECS.has((candidate.codec ?? '').toLowerCase());
(looksMoving ? moving : other).push(candidate);
});
}
return pickPrimaryVideoStream(candidates);
}
const area = (s: VideoStream) => (s.width ?? 0) * (s.height ?? 0);
const pool = moving.length > 0 ? moving : other;
return pool.reduce<VideoStream | null>((best, s) => (!best || area(s) > area(best) ? s : best), null);
/**
* The moving picture among video candidates, wherever they came from.
*
* Prefers streams that have a frame rate, are not a still-image codec and are
* not flagged as an attached picture; the largest area wins within that class,
* falling back to the largest of the rest rather than giving up.
*/
export function pickPrimaryVideoStream(candidates: VideoStream[]): VideoStream | null {
const looksMoving = (v: VideoStream) =>
!v.attachedPic && !!v.frameRate && !STILL_IMAGE_CODECS.has((v.codec ?? '').toLowerCase());
const moving = candidates.filter(looksMoving);
const pool = moving.length > 0 ? moving : candidates;
const area = (v: VideoStream) => (v.width ?? 0) * (v.height ?? 0);
return pool.reduce<VideoStream | null>((best, v) => (!best || area(v) > area(best) ? v : best), null);
}
/**
@@ -323,7 +430,7 @@ function run(
cmd: string,
args: string[],
timeoutMs: number
): Promise<{ code: number | null; stdout: Buffer; stderr: string; timedOut: boolean }> {
): Promise<{ code: number | null; signal: NodeJS.Signals | null; stdout: Buffer; stderr: string; timedOut: boolean }> {
return new Promise((resolve) => {
const child = spawn(cmd, args, { stdio: ['ignore', 'pipe', 'pipe'] });
const stdout: Buffer[] = [];
@@ -342,15 +449,74 @@ function run(
});
child.on('error', (err) => {
clearTimeout(timer);
resolve({ code: null, stdout: Buffer.concat(stdout), stderr: stderr + String(err), timedOut });
resolve({ code: null, signal: null, stdout: Buffer.concat(stdout), stderr: stderr + String(err), timedOut });
});
child.on('close', (code) => {
child.on('close', (code, signal) => {
clearTimeout(timer);
resolve({ code, stdout: Buffer.concat(stdout), stderr, timedOut });
resolve({ code, signal, stdout: Buffer.concat(stdout), stderr, timedOut });
});
});
}
export interface MediaProbe {
duration?: number;
videoStreams: VideoStream[];
}
/**
* Duration and video streams read straight from the file.
*
* Both normally arrive from the ffmpeg plugin, but meta-sort marks a dependency
* complete on `failed` and `skipped` callbacks as well as `completed`
* (ContainerPluginScheduler.markPluginCompleted), so a still task can run
* against a record missing either — or both: on the dev stack a midhash record
* had `fileType=video` and no `fileinfo` or `stream` at all. Without a duration
* frame choice falls back to 10s/0s (idents and opening credits); without
* streams there is nothing to map. One header-only ffprobe answers both.
*/
export async function probeMedia(input: string): Promise<MediaProbe> {
const args = ['-v', 'error'];
if (/^https?:\/\//i.test(input)) {
args.push('-rw_timeout', '30000000'); // microseconds
}
args.push('-print_format', 'json', '-show_format', '-show_streams', input);
const result = await run('ffprobe', args, 30000);
if (result.timedOut || result.code !== 0) return { videoStreams: [] };
let parsed: {
format?: { duration?: string };
streams?: Array<{
codec_type?: string; index?: number; codec_name?: string;
width?: number; height?: number; avg_frame_rate?: string;
disposition?: { attached_pic?: number };
}>;
};
try {
parsed = JSON.parse(result.stdout.toString());
} catch {
return { videoStreams: [] };
}
const duration = Number(parsed.format?.duration);
const videoStreams = (parsed.streams ?? [])
.filter((v) => v.codec_type === 'video' && v.index != null)
.map((v): VideoStream => ({
index: Number(v.index),
width: v.width,
height: v.height,
codec: v.codec_name,
frameRate: v.avg_frame_rate && v.avg_frame_rate !== '0/0' ? v.avg_frame_rate : undefined,
attachedPic: v.disposition?.attached_pic === 1,
}));
return { duration: Number.isFinite(duration) && duration > 0 ? duration : undefined, videoStreams };
}
/** Container duration read from the file — see probeMedia. */
export async function probeDuration(input: string): Promise<number | undefined> {
return (await probeMedia(input)).duration;
}
/**
* Grab one frame at `offsetSec` and return the encoded JPEG.
*
@@ -365,7 +531,9 @@ export async function grabFrame(
offsetSec: number,
outputPath: string
): Promise<Buffer | null> {
const args = ['-y', '-hide_banner', '-loglevel', 'error'];
// -threads caps the decoder's thread pool. Left alone, dav1d/hevc size it
// to the host's cores, and every thread holds its own frame buffers.
const args = ['-y', '-hide_banner', '-loglevel', 'error', '-threads', '2'];
// rw_timeout is a protocol option; only http/tcp accept it, and passing it
// to the file protocol makes ffmpeg bail before it starts.
@@ -380,8 +548,11 @@ export async function grabFrame(
'-an', '-sn', '-dn',
'-frames:v', '1',
// thumbnail=N picks the most representative of the next N frames rather
// than whatever the seek happened to land on.
'-vf', `thumbnail=60,scale='min(iw,${maxWidth})':-2`,
// than whatever the seek happened to land on — and it BUFFERS all N.
// ⚠ Scale first. `thumbnail=60,scale=…` holds 60 full-resolution frames;
// on a 1080p AV1 episode that peaked at 409 MiB, and `scale=…,thumbnail=60`
// at 196 MiB (and 2s instead of 4s) for a byte-for-byte comparable JPEG.
'-vf', `scale='min(iw,${maxWidth})':-2,thumbnail=60`,
'-q:v', String(jpegQuality),
'-f', 'image2',
outputPath
@@ -393,7 +564,12 @@ export async function grabFrame(
return null;
}
if (result.code !== 0 || !existsSync(outputPath)) {
console.warn(`[still-extractor] ffmpeg failed at ${offsetSec}s: ${result.stderr.trim() || `exit ${result.code}`}`);
// A signal with no stderr is the kernel, not ffmpeg: in practice the
// OOM killer. Say so, instead of the old unhelpful "exit null".
const why = result.signal
? `killed by ${result.signal}${result.signal === 'SIGKILL' ? ' (likely out of memory)' : ''}`
: result.stderr.trim() || `exit ${result.code}`;
console.warn(`[still-extractor] ffmpeg failed at ${offsetSec}s: ${why}`);
return null;
}
@@ -497,12 +673,16 @@ export async function process(
const { taskId, cid, filePath, existingMeta } = request;
const metaCore = new MetaCoreClient(request.metaCoreUrl);
const skip = (reason: string) =>
sendCallback({ taskId, status: 'skipped', duration: Date.now() - startTime, reason });
const skip = (reason: string, quiet = false) => {
// Every non-video file in the library comes through here too — logging
// those would bury the skips that matter ("No video stream" on a video).
if (!quiet) console.log(`[still-extractor] ${filePath}: skipped — ${reason}`);
return sendCallback({ taskId, status: 'skipped', duration: Date.now() - startTime, reason });
};
try {
if (existingMeta?.fileType !== 'video') {
await skip('Not a video file');
await skip('Not a video file', true);
return;
}
@@ -514,16 +694,17 @@ export async function process(
return;
}
const stream = selectPrimaryVideoStream(existingMeta);
if (!stream) {
await skip('No video stream');
return;
}
let stream = selectPrimaryVideoStream(existingMeta);
// WebDAV of the very core we're enriching, so the frame bytes and the
// CID we write can never point at different cores.
const webdavClient = await getWebDAVClient(request.metaCoreUrl);
if (!webdavClient) {
if (!stream) {
// Nothing in the record and no way to read the file.
await skip('No video stream');
return;
}
await sendCallback({
taskId,
status: 'failed',
@@ -533,10 +714,61 @@ export async function process(
return;
}
const duration = Number(existingMeta?.['fileinfo/duration']);
const input = webdavClient.toWebDAVUrl(filePath);
let probe: MediaProbe | undefined;
const probeOnce = async () => (probe ??= await probeMedia(input));
const extracted = await extractStill(input, stream.index, Number.isFinite(duration) ? duration : undefined);
let streamSource = '';
if (!stream) {
stream = pickPrimaryVideoStream((await probeOnce()).videoStreams);
streamSource = ' [probed]';
if (!stream) {
await skip('No video stream');
return;
}
}
const recordDuration = durationFromMeta(existingMeta as Record<string, unknown> | undefined);
if (recordDuration === undefined) {
// Name what this task actually received, so an upstream cause of a
// missing duration stays diagnosable even though we recover from it.
const keys = Object.keys(existingMeta ?? {});
console.warn(
`[still-extractor] ${filePath}: no usable fileinfo/duration ` +
`(fileinfo=${JSON.stringify((existingMeta as Record<string, unknown> | undefined)?.['fileinfo'] ?? existingMeta?.['fileinfo/duration'] ?? null).slice(0, 120)}), probing the file; ` +
`existingMeta has ${keys.length} keys: ${keys.slice(0, 40).join(', ')}`
);
}
const duration = recordDuration ?? (await probeOnce()).duration;
const durationSource = recordDuration !== undefined ? 'record' : duration !== undefined ? 'probed' : 'unknown';
console.log(
`[still-extractor] ${filePath}: stream=0:${stream.index}${streamSource} ` +
`${stream.width ?? '?'}x${stream.height ?? '?'} ${stream.codec ?? ''} ` +
`duration=${duration ?? 'unknown'} (${durationSource}) ` +
`offsets=${candidateOffsets(duration).join(',')}`
);
const extracted = await withExtractionSlot(async () => {
// existingMeta was read at dispatch, before this task queued for a slot.
// A replay or rescan queues several tasks per file, so re-check here —
// otherwise every duplicate behind the first recomputes the same still.
if (!forceRecompute && await metaCore.getProperty(cid, 'still')) return 'present' as const;
if (!forceRecompute && recentlyFailed(cid)) return 'recently-failed' as const;
const result = await extractStill(input, stream.index, duration);
if (!result) rememberNoUsableFrame(cid);
return result;
});
if (extracted === 'present') {
await skip('Still already present');
return;
}
if (extracted === 'recently-failed') {
// The first attempt already logged "No usable frame".
await skip('No usable frame (recent attempt, not retried)', true);
return;
}
if (!extracted) {
await skip('No usable frame');
return;
+180
View File
@@ -23,6 +23,14 @@ import {
selectPrimaryVideoStream,
extractStill,
grabFrame,
configure,
withExtractionSlot,
probeDuration,
durationFromMeta,
probeMedia,
pickPrimaryVideoStream,
recentlyFailed,
rememberNoUsableFrame,
} from '../src/plugin.js';
import type { CallbackPayload } from '../src/types.js';
@@ -282,3 +290,175 @@ describe('process - skip logic', () => {
expect(seen[0].reason).toBe('No video stream');
});
});
describe('withExtractionSlot', () => {
const measurePeak = async (jobs: number) => {
let active = 0;
let peak = 0;
const job = () => withExtractionSlot(async () => {
active++;
peak = Math.max(peak, active);
await new Promise((r) => setTimeout(r, 20));
active--;
});
await Promise.all(Array.from({ length: jobs }, job));
return peak;
};
it('never runs more extractions than maxConcurrent', async () => {
// The dev-stack OOM: four concurrent 1080p grabs in a 512 MiB container.
configure({ maxConcurrent: 1 });
expect(await measurePeak(4)).toBe(1);
});
it('allows parallelism when configured, and releases every slot', async () => {
configure({ maxConcurrent: 3 });
expect(await measurePeak(6)).toBe(3);
configure({ maxConcurrent: 1 });
expect(await measurePeak(3)).toBe(1);
});
it('releases the slot when the job throws', async () => {
configure({ maxConcurrent: 1 });
await expect(withExtractionSlot(async () => { throw new Error('boom'); })).rejects.toThrow('boom');
expect(await measurePeak(2)).toBe(1); // would hang forever if the slot leaked
});
});
describe('configure', () => {
it.skipIf(!ffmpegAvailable || !fixturesReady)(
'treats null numeric values as absent instead of zero',
async () => {
// Number(null) === 0: before the fix this set a 0 px width and a 0 ms
// timeout, killing every grab on the spot.
configure({ maxWidth: null, jpegQuality: null, frameTimeoutMs: null, maxConcurrent: null });
mkdirSync(TEMP, { recursive: true });
const result = await extractStill(fixture('color.mp4'), 0, 10, TEMP);
expect(result).not.toBeNull();
}
);
});
describe.skipIf(!ffmpegAvailable || !fixturesReady)('probeDuration', () => {
it('reads the duration from the file header', async () => {
const d = await probeDuration(fixture('color.mp4'));
expect(d).toBeGreaterThan(9);
expect(d).toBeLessThan(11);
});
it('returns undefined for an unreadable input instead of throwing', async () => {
expect(await probeDuration(fixture('does-not-exist.mkv'))).toBeUndefined();
});
it('turns a record with no duration into real candidate offsets, not the 10s/0s fallback', async () => {
const d = await probeDuration(fixture('black-head.mp4'));
expect(d).toBeDefined();
expect(candidateOffsets(d)).not.toEqual([10, 0]);
// and the 20% candidate lands past the 6s black head
expect(candidateOffsets(d)[0]).toBeGreaterThanOrEqual(4);
});
});
describe('durationFromMeta', () => {
it('reads the nested document form meta-sort actually sends', () => {
expect(durationFromMeta({ fileType: 'video', fileinfo: { duration: '1433.089', formatName: 'matroska,webm' } }))
.toBeCloseTo(1433.089);
});
it('reads a numeric nested duration, the exact shape metasort-app/meta returns', () => {
// Captured from the dev stack: fileinfo.duration is a JSON number there, not a string.
expect(durationFromMeta({ fileinfo: { duration: 1433.089, formatName: 'matroska,webm' } })).toBe(1433.089);
});
it('reads the flat meta-core key', () => {
expect(durationFromMeta({ 'fileinfo/duration': '1380.031' })).toBeCloseTo(1380.031);
});
it('reads a JSON-string fileinfo', () => {
expect(durationFromMeta({ fileinfo: JSON.stringify({ duration: '42.5' }) })).toBe(42.5);
});
it('treats missing, zero and junk values as absent', () => {
expect(durationFromMeta(undefined)).toBeUndefined();
expect(durationFromMeta({})).toBeUndefined();
expect(durationFromMeta({ fileinfo: {} })).toBeUndefined();
expect(durationFromMeta({ fileinfo: { duration: '0' } })).toBeUndefined();
expect(durationFromMeta({ fileinfo: { duration: 'N/A' } })).toBeUndefined();
expect(durationFromMeta({ fileinfo: 'not json' })).toBeUndefined();
});
});
describe('selectPrimaryVideoStream — nested payload shapes', () => {
it('reads a nested stream object keyed by index', () => {
const picked = selectPrimaryVideoStream({
stream: {
'0': JSON.stringify({ type: 'audio', index: 0 }),
'1': JSON.stringify({ type: 'video', index: 1, width: 1280, height: 720, codec: 'h264', frameRate: '24/1' }),
},
} as unknown as Record<string, string>);
expect(picked?.index).toBe(1);
});
it('reads a nested stream array of objects', () => {
const picked = selectPrimaryVideoStream({
stream: [
{ type: 'video', index: 0, width: 1920, height: 1080, codec: 'av1', frameRate: '24000/1001' },
{ type: 'audio', index: 1 },
],
} as unknown as Record<string, string>);
expect(picked?.index).toBe(0);
});
});
describe('pickPrimaryVideoStream', () => {
it('never prefers an attached picture, even one with a frame rate', () => {
const picked = pickPrimaryVideoStream([
{ index: 0, width: 640, height: 480, codec: 'h264', frameRate: '24/1' },
{ index: 1, width: 1000, height: 1500, codec: 'h264', frameRate: '24/1', attachedPic: true },
]);
expect(picked?.index).toBe(0);
});
it('returns null with no candidates', () => {
expect(pickPrimaryVideoStream([])).toBeNull();
});
});
describe.skipIf(!ffmpegAvailable || !fixturesReady)('probeMedia', () => {
it('reads the duration and video stream of a plain video', async () => {
const probe = await probeMedia(fixture('color.mp4'));
expect(probe.duration).toBeGreaterThan(9);
expect(pickPrimaryVideoStream(probe.videoStreams)?.width).toBe(320);
});
it('flags attached cover art and never picks it', async () => {
const probe = await probeMedia(fixture('cover-art.mp4'));
expect(probe.videoStreams.some((v) => v.attachedPic)).toBe(true);
const primary = pickPrimaryVideoStream(probe.videoStreams);
expect(primary?.attachedPic).toBe(false);
expect(primary?.width).toBe(320);
});
it('finds no video stream in audio-only input', async () => {
expect(pickPrimaryVideoStream((await probeMedia(fixture('audio-only.mp3'))).videoStreams)).toBeNull();
});
it('returns an empty probe for unreadable input instead of throwing', async () => {
expect(await probeMedia(fixture('does-not-exist.mkv'))).toEqual({ videoStreams: [] });
});
});
describe('no-usable-frame memory', () => {
it('remembers a failed cid for an hour, then forgets it', () => {
const t0 = 1_000_000;
expect(recentlyFailed('bafk-fail', t0)).toBe(false);
rememberNoUsableFrame('bafk-fail', t0);
expect(recentlyFailed('bafk-fail', t0 + 59 * 60 * 1000)).toBe(true);
expect(recentlyFailed('bafk-fail', t0 + 61 * 60 * 1000)).toBe(false);
});
it('is keyed per cid', () => {
rememberNoUsableFrame('bafk-a', 5);
expect(recentlyFailed('bafk-b', 5)).toBe(false);
});
});