@torrent-tv/proxy 2.74.1 → 2.76.0

package/test/two-viewers-one-picture.test.js

@@ -0,0 +1,347 @@
+/**
+ * @file Two viewers of one picture, each with their own soundtrack.
+ *
+ * Measured 2026-09-03 (`research/two-viewers-one-file-2026-09-03.md`): two
+ * browsers on one copied file got two picture sessions with byte-identical
+ * output, because the key carried the soundtrack a picture without sound does
+ * not have. Once they share one picture, everything about the sound that used
+ * to be a field of the session has to be a fact about a viewer — otherwise they
+ * switch each other's soundtrack off, once per segment, for the whole film.
+ */
+
+import test from "node:test";
+import { fakeProcess as fakeEncoder, startRunOn } from "./helpers/encode-run.js";
+
+/**
+ * Whether anything of this session is encoding.
+ *
+ * A session holds a SET of runs, so the question is about the set and not about
+ * a field: a run told to stop is not encoding, whatever its process is still
+ * doing about the signal.
+ *
+ * @param {object} session
+ * @returns {boolean}
+ */
+function encoding(session) {
+  return [...(session?.runs ?? [])].some((run) => run.isAlive);
+}
+import assert from "node:assert/strict";
+import { SourceFile } from "../services/source/SourceFile.js";
+import { Timeline } from "../services/output/Timeline.js";
+import { mkdtemp, rm } from "node:fs/promises";
+import os from "node:os";
+import path from "node:path";
+import { audioRenditionKey, HlsSessionManager } from "../services/hls-session-manager.js";
+import { viewerOf } from "../services/viewer/Viewer.js";
+import { fmp4Format } from "../services/segment-formats/fmp4.js";
+import { Output } from "../services/output/Output.js";
+
+const BASE_ID = "aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee";
+const SEGMENT_SECONDS = 4;
+const FIRST = "viewer-one";
+const SECOND = "viewer-two";
+
+/**
+ * A session shaped like a live one, without the ffmpeg run behind it.
+ *
+ * @param {{ id: string, dirPath: string, audioTrackIndex?: number, transcodeAudio?: boolean }} params
+ * @returns {object}
+ */
+function fakeSession({ id, dirPath, audioTrackIndex = 0, transcodeAudio = true }) {
+  return {
+    id,
+    dirPath,
+    // Where this file is cut, held by the file. A fixture that stated it
+    // on the session was describing what production no longer does.
+    timeline: new Timeline({
+      boundaries: Array.from({ length: 101 }, (_, index) => index * SEGMENT_SECONDS),
+      cutGrid: "keyframe"
+    }),
+    state: "ready",
+    file: new SourceFile({ sourceKey: "torrent:abc", fileIndex: 0, name: "video.mkv" }).learn({ width: 1920, height: 1080 }),
+    // An ordinary session reads its own file, and its sound is inside it. The
+    // three differ only for a soundtrack shipped as a file of its own.
+    get inputFile() { return this.file; },
+    get audioFile() { return this.file; },
+    startedAt: Date.now(),
+    createEntryMs: Date.now(),
+    lastAccessedAt: Date.now(),
+    ffmpeg: null,
+    lastError: "",
+    consumers: new Set(),
+    viewers: new Map(),
+
+    segmentFormat: fmp4Format,
+    transcodeVideo: false,
+    transcodeAudio,
+    audioTrackIndex,
+    audioSourceTrackIndex: audioTrackIndex,
+    // The shape this output is encoded AS, decided once for the output.
+    output: new Output({
+      encodeWidth: 0,
+      encodeHeight: 0,
+      outputFps: 24,
+      softwarePreset: null,
+      applyTonemap: false
+    }),
+    encodeRunGeneration: 0,
+    encodeStartIndex: 0,
+    waitEpoch: 0,
+    useSyntheticPlaylist: true,
+    playlistText: "#EXTM3U\n",
+    segmentCount: 100,
+    progress: { state: "running", processedSeconds: 0, startPositionSeconds: 0, speed: "1.0x" }
+  };
+}
+
+
+/**
+ * A base picture serving two viewers, with its audio published separately and
+ * every rendition created by a stub instead of an encoder.
+ *
+ * @returns {Promise<{ manager: HlsSessionManager, base: object, dirPath: string, renditions: Map<string, object> }>}
+ */
+async function pictureWithTwoViewers() {
+  const dirPath = await mkdtemp(path.join(os.tmpdir(), "two-viewers-"));
+  const manager = new HlsSessionManager({
+    enabled: true,
+    ffmpegBin: "ffmpeg",
+    localBindHost: "127.0.0.1",
+    localPort: 9090
+  });
+  const base = fakeSession({ id: BASE_ID, dirPath });
+  base.audioSeparate = true;
+  base.consumers = new Set([FIRST, SECOND]);
+  // Both viewers are watching the picture, which is what keeps their choices
+  // alive; a viewer whose head has expired holds no encoder.
+  viewerOf(base, FIRST).head = { segment: 3, seconds: 12, at: Date.now() };
+  viewerOf(base, SECOND).head = { segment: 3, seconds: 12, at: Date.now() };
+  viewerOf(base, FIRST).audio = { trackIndex: 0, transcode: true };
+  viewerOf(base, SECOND).audio = { trackIndex: 1, transcode: true };
+  manager.sessionsById.set(BASE_ID, base);
+  manager.getCachedAudioTracks = () => [
+    { index: 0, language: "rus", title: "Дубляж", isDefault: true, fileIndex: 0, sourceTrackIndex: 0 },
+    { index: 1, language: "eng", title: "", isDefault: false, fileIndex: 0, sourceTrackIndex: 1 }
+  ];
+  manager.getCachedMediaInfo = () => ({ height: 1080, width: 1920, durationSeconds: 400 });
+
+  /** @type {Map<string, object>} */
+  const renditions = new Map();
+  manager.createOrGetSession = async (params) => {
+    const key = audioRenditionKey(params.audioTrackIndex, params.transcodeAudio);
+    const existing = renditions.get(key);
+    if (existing) {
+      return existing;
+    }
+    const rendition = fakeSession({
+      id: `rendition-${key}`,
+      dirPath,
+      audioTrackIndex: params.audioTrackIndex,
+      transcodeAudio: params.transcodeAudio
+    });
+    rendition.audioOnly = true;
+    startRunOn(rendition, { process: fakeEncoder() });
+    manager.sessionsById.set(rendition.id, rendition);
+    renditions.set(key, rendition);
+    return rendition;
+  };
+  return { manager, base, dirPath, renditions };
+}
+
+test("one viewer fetching their soundtrack does not stop the other viewer's", async (t) => {
+  const { manager, renditions, dirPath } = await pictureWithTwoViewers();
+  t.after(async () => {
+    await manager.disposeAll();
+    await rm(dirPath, { recursive: true, force: true });
+  });
+
+  const first = await manager.resolveAudioRenditionFile(BASE_ID, 0, "segment-00003.mp4", FIRST);
+  const second = await manager.resolveAudioRenditionFile(BASE_ID, 1, "segment-00003.mp4", SECOND);
+
+  assert.notEqual(first.sessionId, second.sessionId, "two soundtracks are two encodes");
+  for (const [key, rendition] of renditions) {
+    assert.ok([...rendition.runs][0]?.process, `the encoder of ${key} is still running`);
+    assert.deepEqual([...rendition.runs][0].process.signals, [], `nothing signalled ${key}`);
+  }
+
+  // And it holds under the traffic that actually happens: they alternate.
+  await manager.resolveAudioRenditionFile(BASE_ID, 0, "segment-00004.mp4", FIRST);
+  await manager.resolveAudioRenditionFile(BASE_ID, 1, "segment-00004.mp4", SECOND);
+  for (const [key, rendition] of renditions) {
+    assert.deepEqual([...rendition.runs][0]?.process?.signals ?? [], [], `nothing signalled ${key} on the second round`);
+  }
+});
+
+test("a soundtrack nobody is listening to any more is stopped", async (t) => {
+  const { manager, base, renditions, dirPath } = await pictureWithTwoViewers();
+  t.after(async () => {
+    await manager.disposeAll();
+    await rm(dirPath, { recursive: true, force: true });
+  });
+  // One viewer only, so what they leave is left for nobody. This is the case
+  // the stop exists for: an encoder AND a reader holding pieces of the torrent.
+  base.consumers = new Set([FIRST]);
+  base.viewers.delete(SECOND);
+  base.viewers.delete(SECOND);
+
+  await manager.resolveAudioRenditionFile(BASE_ID, 0, "segment-00003.mp4", FIRST);
+  await manager.resolveAudioRenditionFile(BASE_ID, 1, "segment-00004.mp4", FIRST);
+
+  const left = renditions.get(audioRenditionKey(0, true));
+  const moved = renditions.get(audioRenditionKey(1, true));
+  assert.equal(encoding(left), false, "the track the viewer left is not encoding for anybody");
+  assert.ok(encoding(moved), "the track they moved to is");
+});
+
+test("each viewer's browser decides for itself whether its soundtrack is re-encoded", async (t) => {
+  const { manager, base, renditions, dirPath } = await pictureWithTwoViewers();
+  t.after(async () => {
+    await manager.disposeAll();
+    await rm(dirPath, { recursive: true, force: true });
+  });
+  // The same track, two browsers: one can decode it as it stands, the other
+  // cannot. Answering both from the session's own flag would leave the second
+  // viewer with silence.
+  viewerOf(base, FIRST).audio = { trackIndex: 0, transcode: false };
+  viewerOf(base, SECOND).audio = { trackIndex: 0, transcode: true };
+
+  const copied = await manager.resolveAudioRenditionFile(BASE_ID, 0, "segment-00003.mp4", FIRST);
+  const encoded = await manager.resolveAudioRenditionFile(BASE_ID, 0, "segment-00003.mp4", SECOND);
+
+  assert.notEqual(copied.sessionId, encoded.sessionId);
+  assert.equal(renditions.get(audioRenditionKey(0, false)).transcodeAudio, false);
+  assert.equal(renditions.get(audioRenditionKey(0, true)).transcodeAudio, true);
+  // Both are wanted, so neither is stopped.
+  assert.ok(encoding(renditions.get(audioRenditionKey(0, false))));
+  assert.ok(encoding(renditions.get(audioRenditionKey(0, true))));
+});
+
+test("the master marks each viewer's own soundtrack as the default one", async (t) => {
+  const { manager, dirPath } = await pictureWithTwoViewers();
+  t.after(async () => {
+    await manager.disposeAll();
+    await rm(dirPath, { recursive: true, force: true });
+  });
+
+  const forFirst = manager.buildMasterPlaylist(BASE_ID, FIRST);
+  const forSecond = manager.buildMasterPlaylist(BASE_ID, SECOND);
+
+  const defaultsOf = (master) =>
+    [...master.matchAll(/^#EXT-X-MEDIA:.*?NAME="([^"]+)".*?DEFAULT=(YES|NO)/gm)]
+      .filter((match) => match[2] === "YES")
+      .map((match) => match[1]);
+  assert.deepEqual(defaultsOf(forFirst), ["Дубляж"]);
+  assert.deepEqual(defaultsOf(forSecond).length, 1);
+  assert.notDeepEqual(defaultsOf(forFirst), defaultsOf(forSecond));
+});
+
+test("one viewer changing quality does not take the other off their step", async (t) => {
+  const { manager, base, dirPath } = await pictureWithTwoViewers();
+  t.after(async () => {
+    await manager.disposeAll();
+    await rm(dirPath, { recursive: true, force: true });
+  });
+  /** @type {Map<number, object>} */
+  const variants = new Map();
+  manager.createOrGetSession = async (params) => {
+    const height = params.targetHeight;
+    const existing = variants.get(height);
+    if (existing) {
+      return existing;
+    }
+    const variant = fakeSession({ id: `variant-${height}`, dirPath });
+    variant.transcodeVideo = true;
+    variant.output.encodeHeight = height;
+    variant.variantHeight = height;
+    variant.isStep = true;
+    variant.file = base.file;
+    startRunOn(variant, { process: fakeEncoder() });
+    manager.sessionsById.set(variant.id, variant);
+    variants.set(height, variant);
+    return variant;
+  };
+
+  await manager.resolveVariantFile(BASE_ID, 720, "segment-00003.mp4", FIRST);
+  await manager.resolveVariantFile(BASE_ID, 540, "segment-00003.mp4", SECOND);
+  // Both viewers go on watching their own step, which is what a player does
+  // every few seconds.
+  await manager.resolveVariantFile(BASE_ID, 720, "segment-00004.mp4", FIRST);
+  await manager.resolveVariantFile(BASE_ID, 540, "segment-00004.mp4", SECOND);
+
+  assert.ok(encoding(variants.get(720)), "the first viewer's step is still encoding");
+  assert.ok(encoding(variants.get(540)), "and so is the second viewer's");
+
+  // Now the first viewer steps down. Theirs is left for nobody and stops; the
+  // other viewer's is untouched.
+  await manager.resolveVariantFile(BASE_ID, 480, "segment-00005.mp4", FIRST);
+
+  assert.equal(encoding(variants.get(720)), false, "the step nobody is on stops");
+  assert.ok(encoding(variants.get(540)), "the step the other viewer is watching does not");
+  assert.ok(encoding(variants.get(480)), "and the one they moved to is encoding");
+});
+
+test("a step somebody is watching is never withdrawn from the offer", async (t) => {
+  const { manager, base, dirPath } = await pictureWithTwoViewers();
+  t.after(async () => {
+    await manager.disposeAll();
+    await rm(dirPath, { recursive: true, force: true });
+  });
+  // A host that can re-encode 240p and nothing above it — the shape of the
+  // field case of 2026-08-15.
+  base.output.encodeHeight = 1080;
+  base.variantHeight = 1080;
+  manager.softwarePresetBenchmark = [{ preset: "ultrafast", pixelsPerSec: 12_000_000 }];
+  manager.decodeCostModel = { pixelTerm: 0.00793, bitrateTerm: 0, constantTerm: 0 };
+  // 1080p24 at 8 Mbit/s, stated as the file's own facts — what decoding costs
+  // is derived from them.
+  base.file.learn({ width: 1920, height: 1080, fps: 24, bitrateKbps: 8000 });
+
+  const variant = fakeSession({ id: "variant-720", dirPath });
+  variant.transcodeVideo = true;
+  variant.output.encodeHeight = 720;
+  variant.variantHeight = 720;
+  // One file, two sessions of it.
+  variant.file = base.file;
+  variant.isStep = true;
+  variant.file = base.file;
+  manager.sessionsById.set(variant.id, variant);
+  base.file.stepHeights.set(720, 720);
+
+  // Nobody on it: measured below realtime, it is withdrawn. This half is the
+  // control — without it the other half proves nothing.
+  const withoutAViewer = manager.offeredHeights(base);
+  assert.ok(
+    !withoutAViewer.includes(720),
+    `a step nobody is on and that cannot keep up is withdrawn: ${withoutAViewer.join(" ")}`
+  );
+
+  viewerOf(base, SECOND).activeVariantId = variant.id;
+  const withAViewer = manager.offeredHeights(base);
+
+  assert.ok(
+    withAViewer.includes(720),
+    `a step on somebody's screen stays offered, whatever it is measured at: ${withAViewer.join(" ")}`
+  );
+});
+
+test("a viewer whose picture has gone quiet holds no soundtrack encoder", async (t) => {
+  const { manager, base, renditions, dirPath } = await pictureWithTwoViewers();
+  t.after(async () => {
+    await manager.disposeAll();
+    await rm(dirPath, { recursive: true, force: true });
+  });
+  await manager.resolveAudioRenditionFile(BASE_ID, 0, "segment-00003.mp4", FIRST);
+  await manager.resolveAudioRenditionFile(BASE_ID, 1, "segment-00003.mp4", SECOND);
+  // The second viewer's tab is gone. Nothing releases the session when a
+  // channel closes (roadmap item 54), so what expires is their head on the
+  // picture — and with it their claim on an encoder.
+  base.viewers.delete(SECOND);
+
+  await manager.resolveAudioRenditionFile(BASE_ID, 1, "segment-00004.mp4", FIRST);
+
+  assert.equal(
+    encoding(renditions.get(audioRenditionKey(0, true))),
+    false,
+    "the first viewer moved on, so their old track is stopped"
+  );
+});

package/test/video-facts.test.js

@@ -0,0 +1,102 @@
+/**
+ * @file The picture's facts, from the two readings that state them.
+ *
+ * Audio and subtitles have had this reconciliation since their flags were first
+ * read from the file. Video never did: every figure the encode is planned from
+ * came from ffmpeg's `-i` banner alone, and the `VideoTrack` the container
+ * declares was read and then used for nothing but a line in the log — so a file
+ * that states its bit depth or its HDR signalling, against a probe that does
+ * not print them, disagreed with nobody watching.
+ *
+ * What each check pins is WHICH reading answers, and why: the size and the
+ * frame rate are what the decoder will produce, so the probe answers; the bit
+ * depth and the HDR signalling are the file saying how its samples are to be
+ * read, so the container answers.
+ */
+
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import { Container } from "../services/container/Container.js";
+
+const banner = (fields) => ({ width: null, height: null, fps: null, isHdr: false, bitDepth: null, ...fields });
+
+test("with no container reading the probe answers for everything", () => {
+  const facts = Container.mergeVideoFacts(
+    banner({ width: 1920, height: 1080, fps: 23.976, isHdr: true, bitDepth: 10 }),
+    null
+  );
+  assert.equal(facts.width, 1920);
+  assert.equal(facts.height, 1080);
+  assert.equal(facts.fps, 23.976);
+  assert.equal(facts.isHdr, true);
+  assert.equal(facts.bitDepth, 10);
+  assert.deepEqual(facts.disagreements, []);
+});
+
+test("the size and the frame rate are the probe's, because that is what will be decoded", () => {
+  const facts = Container.mergeVideoFacts(
+    banner({ width: 1920, height: 1080, fps: 24 }),
+    { width: 1440, height: 1080, fps: 25, displayWidth: 1920, displayHeight: 1080 }
+  );
+  assert.equal(facts.width, 1920, "the ladder and the scale filter are sized to the decoded frame");
+  assert.equal(facts.height, 1080);
+  assert.equal(facts.fps, 24);
+});
+
+test("the bit depth and the HDR signalling are the file's, because the file states them", () => {
+  // A ten-bit HEVC whose probe printed neither: this is the case that decides
+  // whether tone mapping runs and how the decode cost is priced.
+  const facts = Container.mergeVideoFacts(
+    banner({ width: 3840, height: 2160, fps: 24 }),
+    { width: 3840, height: 2160, bitDepth: 10, isHdr: true }
+  );
+  assert.equal(facts.bitDepth, 10);
+  assert.equal(facts.isHdr, true);
+  assert.deepEqual(facts.disagreements, [], "one side saying nothing is not a disagreement");
+});
+
+test("a field only the probe states is still answered", () => {
+  const facts = Container.mergeVideoFacts(
+    banner({ width: 1280, height: 720, fps: 30, bitDepth: 8 }),
+    { width: null, height: null, fps: null, bitDepth: null }
+  );
+  assert.equal(facts.width, 1280);
+  assert.equal(facts.bitDepth, 8);
+});
+
+test("a real disagreement is reported, with both values and which is which", () => {
+  const facts = Container.mergeVideoFacts(
+    banner({ width: 1920, height: 1080, bitDepth: 8, isHdr: false }),
+    { width: 1280, height: 720, bitDepth: 10, isHdr: true }
+  );
+  assert.equal(facts.disagreements.length, 3);
+  assert.ok(facts.disagreements.some((line) => /width 1280 in the container against 1920 in the probe/.test(line)));
+  assert.ok(facts.disagreements.some((line) => /bit depth 10 in the container against 8 in the probe/.test(line)));
+  // HDR is not among them, and cannot be: see `mergeVideoFacts`.
+  assert.ok(!facts.disagreements.some((line) => /HDR/.test(line)));
+  // And the rule still decides: the probe for the size, the file for the rest.
+  assert.equal(facts.width, 1920);
+  assert.equal(facts.bitDepth, 10);
+  assert.equal(facts.isHdr, true);
+});
+
+test("the display size is the container's alone — the probe has no such field", () => {
+  const facts = Container.mergeVideoFacts(
+    banner({ width: 1440, height: 1080 }),
+    { width: 1440, height: 1080, displayWidth: 1920, displayHeight: 1080 }
+  );
+  assert.equal(facts.displayWidth, 1920);
+  assert.equal(facts.displayHeight, 1080);
+});
+
+test("zero and nonsense are not values", () => {
+  const facts = Container.mergeVideoFacts(
+    banner({ width: 0, height: 0, fps: 0, bitDepth: 0 }),
+    { width: 1920, height: 1080, fps: 24, bitDepth: 8 }
+  );
+  assert.equal(facts.width, 1920, "a probe that printed nothing does not outrank a file that speaks");
+  assert.equal(facts.fps, 24);
+  assert.equal(facts.bitDepth, 8);
+  assert.deepEqual(facts.disagreements, []);
+});