@torrent-tv/proxy 2.9.37 → 2.9.38

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -1,3 +1,7 @@
1
+ ## 2.9.38
2
+
3
+ - **New**: Adaptive bitrate for thin viewer links (OpenSpec change `adaptive-bitrate`). Field evidence (iPhone on cellular): uncapped complex scenes produced 4 s segments of ~18 Mbit/s against a 1–6 Mbit/s link — 45 s prebuffer and a draining buffer. Two parts. (a) Software encodes are now constrained-CRF: `-maxrate`/`-bufsize` per resolution rung (1080p→5000K, 720p→2800K, 480p→1400K, 360p→800K, 240p→400K nominal; ×1.3/×1.5 — webtor's production multipliers), so peaks stay bounded. (b) New data-channel route `POST /api/transcode-sessions/:id/net-report` accepts the browser's measured link throughput + buffered seconds; the realtime-budget loop gains a second downshift trigger — a FRESH report showing the usable link (×0.8 safety) sustainedly (15 s) below the observed produced bitrate while the viewer's buffer is low (<10 s) steps the encode one rung down via the existing machinery (shared 30 s cooldown, step cap, no upswitch). Log reason `viewer-link-bound` distinguishes it from CPU downshifts. Manual-quality sessions are exempt (no budget ladder); old clients that never report simply keep today's behaviour plus the caps.
4
+
1
5
  ## 2.9.37
2
6
 
3
7
  - **Fix**: Scrubbing (server-side seek) no longer hangs the player. A far segment request restarts ffmpeg at that position; native players (notably iOS HLS) issue a burst of scattered far requests after a scrub (observed: `367 → 732 → 369 → 368 → 370`, tens of seconds apart), and the old fixed 4 s cooldown only suppressed restarts within 4 s of the last — so each scattered request restarted ffmpeg and it ping-ponged between positions, producing nothing and stalling playback. Far requests are now **debounced**: the target index is recorded and a short settle timer armed (1.2 s quiet period, 2.5 s hard cap from the burst's first request); further far requests re-arm it and update the target to the latest index; when it settles, ffmpeg restarts once at that index. "Last index wins" self-corrects — a wrong target costs at most one extra settle, never the old infinite loop. The settle timer is cleared on session disposal. (OpenSpec change `seek-debounce`.)
@@ -0,0 +1,92 @@
1
+ # Design: Adaptive bitrate (proxy)
2
+
3
+ Read before coding: `services/hwaccel.js` (`chooseSoftwareEncodeSettings`,
4
+ the software descriptor's `buildVideoArgs`), `services/hls-session-manager.js`
5
+ (the `BUDGET_*` constants, `#chooseEncodeBudget`, the budget interval check
6
+ `#enforceRealtimeBudget`, `#classifyTranscodeBound`, `#applyBudgetDownshift`,
7
+ session fields around `budgetLadder`), `routes/api/transcode-sessions/`
8
+ (route file conventions), `server.js` (route wiring, deps injection).
9
+
10
+ ## (a) Caps — constrained CRF
11
+
12
+ Nominal H.264 rates by rung height (nearest rung wins for odd heights):
13
+
14
+ RUNG_NOMINAL_KBPS = { 1080: 5000, 720: 2800, 480: 1400, 360: 800, 240: 400 }
15
+ CAP_MAXRATE_FACTOR = 1.3
16
+ CAP_BUFSIZE_FACTOR = 1.5
17
+
18
+ In the software descriptor's video args (where `-crf`/`-preset` are emitted),
19
+ add for encode height H (the actual encode height after budget/manual
20
+ selection, not the source height):
21
+
22
+ -maxrate <round(nominal(H) * 1.3)>k
23
+ -bufsize <round(nominal(H) * 1.5)>k
24
+
25
+ CRF stays the quality driver; the caps only bound peaks (standard
26
+ constrained-CRF). Do NOT touch hardware descriptors in this change.
27
+
28
+ ## (b) Net report intake
29
+
30
+ Route: `POST /api/transcode-sessions/:id/net-report` (data-channel path, so
31
+ add the prefix to the allowlist regex in `data-channel-handler.js` if not
32
+ covered by the existing `/api/` rule — verify). Body:
33
+
34
+ { linkMbps: number, bufferedAheadSec: number }
35
+
36
+ Validation: both finite numbers, `linkMbps > 0`, else 400. Unknown session →
37
+ 404. On success 204. Handler stores on the session:
38
+
39
+ session.netReport = { linkMbps, bufferedAheadSec, at: Date.now() }
40
+
41
+ ## (b) Budget-loop trigger
42
+
43
+ Constants (near the other BUDGET_ constants):
44
+
45
+ LINK_REPORT_FRESH_MS = 30_000 // ignore stale reports
46
+ LINK_SAFETY = 0.8 // usable share of reported link
47
+ LINK_SLOW_WINDOW_MS = 15_000 // sustained deficit before acting
48
+ LINK_LOW_BUFFER_SEC = 10 // only act while the viewer runs dry
49
+
50
+ In the periodic budget check, for each active software-transcode session,
51
+ alongside the CPU check:
52
+
53
+ report fresh (now - at < LINK_REPORT_FRESH_MS)?
54
+ manualQuality not set?
55
+ observed = observed stream bitrate, Mbit/s — recent produced segment
56
+ bytes / segment duration (reuse/extend whatever the CPU path
57
+ reads; a rolling average over the last ~5 segments)
58
+ deficit = report.linkMbps * LINK_SAFETY < observed
59
+ if deficit AND report.bufferedAheadSec < LINK_LOW_BUFFER_SEC:
60
+ accumulate slow-time (same pattern as the CPU slow window)
61
+ if slow ≥ LINK_SLOW_WINDOW_MS → #applyBudgetDownshift(session,
62
+ reason "link") // existing cooldown + step cap + floor apply
63
+ else: reset the link slow window
64
+
65
+ `#applyBudgetDownshift` is reused as-is except the log line carries the
66
+ reason: `budget downshift (link) …` vs the current CPU wording, so the
67
+ client-log pipeline can tell them apart in the field.
68
+
69
+ No upswitch in v1 (mirrors the CPU budget's conservatism). A downshifted
70
+ session stays down until re-opened.
71
+
72
+ ## Interactions
73
+
74
+ - Caps make `observed` honest: without (a) a complex scene can spike far
75
+ above nominal and flap the trigger; with caps observed ≤ maxrate.
76
+ - CPU trigger and link trigger share the cooldown inside
77
+ `#applyBudgetDownshift` — they cannot double-fire.
78
+ - `manualQuality` set → link trigger skipped entirely (user pinned quality
79
+ explicitly; the proxy respects it, matching the manual-quality contract).
80
+ - Old browsers never POST reports → trigger never fires → behaviour
81
+ identical to today plus caps.
82
+
83
+ ## Verification
84
+
85
+ - Unit: nominal table lookup (exact rungs + nearest for odd heights); ffmpeg
86
+ arg assembly contains the caps; trigger logic with a fake clock (fresh vs
87
+ stale report, deficit accumulating to a downshift, buffer-high suppresses,
88
+ manualQuality suppresses).
89
+ - `node --check` on touched files.
90
+ - Field (after the server change ships): cellular session shows
91
+ `budget downshift (link)` and the stream settles at a rung whose bitrate
92
+ fits the link; segment sizes bounded (~maxrate × segDur / 8 max).
@@ -0,0 +1,56 @@
1
+ # Proposal: Adaptive bitrate for thin viewer links (proxy side)
2
+
3
+ ## Why
4
+
5
+ Field evidence (iPhone on cellular, 2026-07-10, session `14cc1017`): encode
6
+ bitrate is unbounded — transcoded segments reached 9 MB per 4 s (~18 Mbit/s)
7
+ against a measured 1–5.8 Mbit/s cellular link. Result: 45 s prebuffer, buffer
8
+ draining during playback (`bottleneck delta=-7.1s`), "было так себе". The
9
+ realtime budget protects the PROXY's CPU but nothing protects the VIEWER's
10
+ link: quality never adapts to how fast the viewer can actually download.
11
+
12
+ ## What Changes
13
+
14
+ Two parts, one release:
15
+
16
+ - **(a) Bitrate caps (constrained CRF).** Software video encodes gain
17
+ `-maxrate`/`-bufsize` sized per resolution rung from a nominal-rate table
18
+ (H.264 ladder: 1080p→5000K, 720p→2800K, 480p→1400K, 360p→800K, 240p→400K;
19
+ nearest rung by encode height). Multipliers from webtor's production
20
+ ladder: `maxrate = 1.3×` nominal, `bufsize = 1.5×` nominal. CRF/preset
21
+ selection stays — the caps only bound the peaks that killed the cellular
22
+ session. Applies to the software (libx264) path; hardware encoders keep
23
+ their current args (follow-up — their rate-control flags differ and any
24
+ change must pass the strict startup test on real hardware).
25
+ - **(b) Viewer-link downshift trigger.** A new data-channel route lets the
26
+ browser report its measured link state (`linkMbps` — rolling median of
27
+ per-segment transfer throughput; `bufferedAheadSec`) every ~10 s. The
28
+ existing realtime-budget loop gains a SECOND downshift trigger: when
29
+ reports are fresh, `manualQuality` is not set, the viewer's link is
30
+ sustainedly slower than the produced stream (`linkMbps × 0.8 <` observed
31
+ segment bitrate for ≥ 15 s) AND the viewer's buffer is low (< 10 s), step
32
+ one rung down via the existing `#applyBudgetDownshift` (same cooldown,
33
+ step cap, no upswitch v1). Distinct log reason (`reason=link`) so field
34
+ logs distinguish CPU-bound from link-bound downshifts.
35
+
36
+ Missing/stale reports change nothing (old clients keep working — the trigger
37
+ simply never fires). Manual quality pins win: when `manualQuality` is set the
38
+ trigger is skipped.
39
+
40
+ ## Capabilities
41
+
42
+ ### Modified Capabilities
43
+
44
+ - `transcode-quality`: encodes are bitrate-capped per rung; the budget loop
45
+ adapts to the viewer's link, not only to the proxy's CPU.
46
+
47
+ ## Impact
48
+
49
+ - `services/hwaccel.js` — nominal-rate table + caps in the software
50
+ descriptor's video args.
51
+ - `services/hls-session-manager.js` — accept/store net reports per session;
52
+ link trigger in the budget check; log line.
53
+ - `routes/api/transcode-sessions/net-report/post.js` (new) + `server.js`
54
+ wiring — report intake.
55
+ - Proxy release + ha-addon bump. The server-side reporter is a separate
56
+ change (`server/viewer-net-report`); release proxy+addon FIRST.
@@ -0,0 +1,51 @@
1
+ # adaptive-bitrate — delta spec (proxy)
2
+
3
+ ## ADDED Requirements
4
+
5
+ ### Requirement: Software encodes are bitrate-capped per resolution rung
6
+
7
+ Software video encodes SHALL carry `-maxrate`/`-bufsize` derived from a
8
+ per-rung nominal-rate table (1.3× / 1.5× of the rung's nominal), keeping the
9
+ existing CRF/preset quality selection (constrained CRF). The cap follows the
10
+ ACTUAL encode height after budget or manual selection.
11
+
12
+ #### Scenario: Complex scene on a capped encode
13
+ - **WHEN** a transcoded scene would spike far above the rung's nominal rate
14
+ - **THEN** the produced segments stay bounded by the cap instead of reaching
15
+ multi-megabyte sizes that a thin viewer link cannot download in time
16
+
17
+ ### Requirement: The proxy accepts viewer link reports
18
+
19
+ A data-channel route SHALL accept periodic reports
20
+ `{ linkMbps, bufferedAheadSec }` for an active transcode session and store
21
+ the latest report with its arrival time. Invalid bodies are rejected;
22
+ unknown sessions get 404; missing reports are not an error condition.
23
+
24
+ #### Scenario: Old client
25
+ - **WHEN** a browser never sends net reports
26
+ - **THEN** the session behaves exactly as before this change (plus caps)
27
+
28
+ ### Requirement: The budget loop downshifts on a sustained link deficit
29
+
30
+ When a fresh report shows the viewer's usable link (reported × safety
31
+ margin) sustainedly below the observed produced bitrate AND the viewer's
32
+ buffer is low, the existing budget downshift SHALL step the encode one rung
33
+ down (same cooldown, step cap and floor as the CPU trigger; no upswitch).
34
+ The trigger SHALL be skipped when `manualQuality` is set, when reports are
35
+ stale, or when the viewer's buffer is comfortable. The downshift log line
36
+ SHALL name the reason (`link`) distinctly from the CPU reason.
37
+
38
+ #### Scenario: Cellular viewer, stream too fat
39
+ - **WHEN** reports show 3 Mbit/s usable link against 6 Mbit/s produced
40
+ bitrate for longer than the slow window, with a draining buffer
41
+ - **THEN** the encode steps down one rung and the log shows a link-reason
42
+ downshift
43
+
44
+ #### Scenario: Slow link but full buffer
45
+ - **WHEN** the link is slower than the stream but the viewer's buffer stays
46
+ comfortable (e.g. paused playback filling ahead)
47
+ - **THEN** no downshift happens
48
+
49
+ #### Scenario: Manual quality pinned
50
+ - **WHEN** the viewer picked a forced quality
51
+ - **THEN** link reports never trigger a downshift
@@ -0,0 +1,40 @@
1
+ # Tasks: Adaptive bitrate (proxy)
2
+
3
+ ## 1. Caps
4
+
5
+ - [x] 1.1 `hwaccel.js`: `RUNG_NOMINAL_KBPS` table + nearest-rung lookup;
6
+ software descriptor's video args emit `-maxrate`/`-bufsize`
7
+ (1.3× / 1.5× nominal for the actual encode height). Hardware
8
+ descriptors untouched.
9
+
10
+ ## 2. Net report intake
11
+
12
+ - [x] 2.1 `routes/api/transcode-sessions/net-report/post.js`: validate
13
+ `{ linkMbps, bufferedAheadSec }`, store
14
+ `session.netReport = { linkMbps, bufferedAheadSec, at }`, 204/400/404.
15
+ - [x] 2.2 Wire in `server.js`; verify the data-channel path allowlist covers
16
+ the route.
17
+
18
+ ## 3. Budget trigger
19
+
20
+ - [x] 3.1 Constants `LINK_REPORT_FRESH_MS`, `LINK_SAFETY`,
21
+ `LINK_SLOW_WINDOW_MS`, `LINK_LOW_BUFFER_SEC`.
22
+ - [x] 3.2 Observed produced bitrate (rolling, last ~5 segments) available to
23
+ the budget check.
24
+ - [x] 3.3 Link deficit accumulation + `#applyBudgetDownshift(session,
25
+ reason "link")`; skip on manualQuality/stale/comfortable buffer; log
26
+ reason distinctly.
27
+
28
+ ## 4. Verification
29
+
30
+ - [x] 4.1 Unit: table lookup, arg assembly, trigger logic (fake clock):
31
+ deficit→downshift, stale report→no-op, high buffer→no-op,
32
+ manualQuality→no-op.
33
+ - [x] 4.2 `node --check` on touched files.
34
+
35
+ ## 5. Release
36
+
37
+ - [ ] 5.1 CHANGELOG (next patch) + `npm run patch` (user OTP), then addon
38
+ bump + push + HA update. Release BEFORE the server reporter change.
39
+ - [ ] 5.2 Field: cellular Poirot/Mavka run shows `budget downshift (link)`
40
+ and bounded segment sizes; correlate via client-log pipeline.
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@torrent-tv/proxy",
3
- "version": "2.9.37",
3
+ "version": "2.9.38",
4
4
  "description": "Torrent proxy client that exposes webseed-like HTTP stream endpoint.",
5
5
  "license": "GPL-3.0-or-later",
6
6
  "publishConfig": {
@@ -0,0 +1,31 @@
1
+ /**
2
+ * Accept a viewer link report for a transcode session (adaptive bitrate).
3
+ * The browser measures its own data-channel throughput per segment fetch and
4
+ * posts a rolling median + its buffered seconds; the session manager's budget
5
+ * loop uses the latest report as the link-deficit downshift trigger.
6
+ *
7
+ * POST /api/transcode-sessions/:sessionId/net-report
8
+ * Body: { linkMbps: number, bufferedAheadSec: number }
9
+ *
10
+ * Best-effort telemetry: invalid body → 400, unknown session → 404, ok → 204.
11
+ *
12
+ * @param {import("fastify").FastifyRequest} req
13
+ * @param {import("fastify").FastifyReply} reply
14
+ * @param {{ hlsSessionManager: import("../../../../services/hls-session-manager.js").HlsSessionManager }} deps
15
+ * @returns {Promise<void>}
16
+ */
17
+ export async function handleApiTranscodeSessionNetReportPost(req, reply, { hlsSessionManager }) {
18
+ const sessionId = typeof req.params.sessionId === "string" ? req.params.sessionId : "";
19
+ const body = req.body && typeof req.body === "object" && !Array.isArray(req.body) ? req.body : {};
20
+ const linkMbps = Number(body.linkMbps);
21
+ const bufferedAheadSec = Number(body.bufferedAheadSec);
22
+ if (!sessionId || !Number.isFinite(linkMbps) || linkMbps <= 0 || !Number.isFinite(bufferedAheadSec) || bufferedAheadSec < 0) {
23
+ return reply.code(400).send({ error: "linkMbps (>0) and bufferedAheadSec (>=0) are required." });
24
+ }
25
+
26
+ const recorded = hlsSessionManager.recordNetReport(sessionId, { linkMbps, bufferedAheadSec });
27
+ if (!recorded) {
28
+ return reply.code(404).send({ error: "Transcode session was not found." });
29
+ }
30
+ return reply.code(204).send();
31
+ }
package/server.js CHANGED
@@ -24,6 +24,7 @@ import { handleApiSubtitlesGet } from "./routes/api/subtitles/get.js";
24
24
  import { handleApiTranscodeSessionsPost } from "./routes/api/transcode-sessions/post.js";
25
25
  import { handleApiTranscodeSessionsProgressGet } from "./routes/api/transcode-sessions/progress/get.js";
26
26
  import { handleApiTranscodeSessionReleasePost } from "./routes/api/transcode-sessions/release/post.js";
27
+ import { handleApiTranscodeSessionNetReportPost } from "./routes/api/transcode-sessions/net-report/post.js";
27
28
  import { handleStreamGet } from "./routes/stream/get.js";
28
29
  import { handleTranscodeSessionFileGet } from "./routes/transcode/session-file/get.js";
29
30
  import { createSourceRegistry } from "./store/source-registry.js";
@@ -186,6 +187,9 @@ export async function startProxyServer({ host, port, transcodeAudio, ffmpegBin,
186
187
  app.get("/api/transcode-sessions/:sessionId/progress", async (req, reply) =>
187
188
  handleApiTranscodeSessionsProgressGet(req, reply, { hlsSessionManager })
188
189
  );
190
+ app.post("/api/transcode-sessions/:sessionId/net-report", async (req, reply) =>
191
+ handleApiTranscodeSessionNetReportPost(req, reply, { hlsSessionManager })
192
+ );
189
193
  app.get("/transcode/:sessionId/:fileName", async (req, reply) =>
190
194
  handleTranscodeSessionFileGet(req, reply, { hlsSessionManager })
191
195
  );
@@ -8,7 +8,7 @@
8
8
  */
9
9
 
10
10
  import { createReadStream } from "node:fs";
11
- import { access, mkdir, readdir, readFile, rm } from "node:fs/promises";
11
+ import { access, mkdir, readdir, readFile, rm, stat } from "node:fs/promises";
12
12
  import { Readable } from "node:stream";
13
13
  import os from "node:os";
14
14
  import path from "node:path";
@@ -84,6 +84,24 @@ const BUDGET_MAX_DOWNSHIFTS = 3;
84
84
  // multiple of the source's average byte rate. Below it (and not yet fully
85
85
  // downloaded), a low speed is download-bound, not CPU-bound → do NOT downscale.
86
86
  const BUDGET_DOWNLOAD_OK_FACTOR = 1.0;
87
+ // Viewer-link adaptation (adaptive bitrate, part b). The browser reports its
88
+ // measured data-channel throughput + buffered seconds every ~10 s; when a
89
+ // FRESH report shows the usable link (reported × safety margin) sustainedly
90
+ // below the observed produced bitrate AND the viewer's buffer is low, the
91
+ // budget loop steps the encode one rung down — same machinery, cooldown and
92
+ // floor as the CPU trigger. Manual-quality sessions are inherently exempt
93
+ // (their budgetLadder is null).
94
+ const LINK_REPORT_FRESH_MS = 30_000;
95
+ // Usable share of the reported link (protocol overhead + measurement noise).
96
+ const LINK_SAFETY = 0.8;
97
+ // Deficit must persist this long before acting (absorbs one slow segment).
98
+ const LINK_SLOW_WINDOW_MS = 15_000;
99
+ // Only act while the viewer is actually running dry; a comfortable buffer
100
+ // (e.g. paused playback filling ahead) suppresses the trigger.
101
+ const LINK_LOW_BUFFER_SEC = 10;
102
+ // Observed produced bitrate: average over this many recently completed
103
+ // segments (the newest file on disk may still be written and is excluded).
104
+ const LINK_OBSERVED_SEGMENTS = 5;
87
105
  const MICROSECONDS_PER_SECOND = 1_000_000;
88
106
  const PROGRESS_LOG_INTERVAL_MS = 5_000;
89
107
  // Read segment files in large blocks so the body is delivered to the data
@@ -873,6 +891,10 @@ export class HlsSessionManager {
873
891
  budgetDownshifts: 0,
874
892
  budgetSlowSince: 0,
875
893
  budgetLastActionAt: 0,
894
+ // Latest viewer link report ({ linkMbps, bufferedAheadSec, at }) and the
895
+ // link-deficit slow window (mirrors budgetSlowSince for the CPU path).
896
+ netReport: null,
897
+ linkSlowSince: 0,
876
898
  sourceWidth,
877
899
  sourceHeight,
878
900
  // Container start time (seconds); subtracted on the copy path so the
@@ -1087,6 +1109,110 @@ export class HlsSessionManager {
1087
1109
  *
1088
1110
  * @returns {Promise<void>}
1089
1111
  */
1112
+ /**
1113
+ * Record the latest viewer link report for a session (adaptive bitrate).
1114
+ * Returns false for an unknown/disposed session.
1115
+ *
1116
+ * @param {string} sessionId
1117
+ * @param {{ linkMbps: number, bufferedAheadSec: number }} report
1118
+ * @returns {boolean}
1119
+ */
1120
+ recordNetReport(sessionId, { linkMbps, bufferedAheadSec }) {
1121
+ const session = this.sessionsById.get(sessionId);
1122
+ if (!session || session.state === "disposed") {
1123
+ return false;
1124
+ }
1125
+ session.netReport = { linkMbps, bufferedAheadSec, at: Date.now() };
1126
+ return true;
1127
+ }
1128
+
1129
+ /**
1130
+ * Observed produced bitrate (Mbit/s) averaged over the last few COMPLETED
1131
+ * segment files (the newest file may still be being written and is
1132
+ * excluded). Transcode sessions only — their segment grid is uniform, so
1133
+ * bytes / (count × segDur) is exact. Returns null when there is not enough
1134
+ * material to measure.
1135
+ *
1136
+ * @param {HlsSession} session
1137
+ * @returns {Promise<number | null>}
1138
+ */
1139
+ async #observedStreamMbps(session) {
1140
+ let names;
1141
+ try {
1142
+ names = await readdir(session.dirPath);
1143
+ } catch {
1144
+ return null;
1145
+ }
1146
+ const indices = [];
1147
+ for (const name of names) {
1148
+ const match = /^segment-(\d{5})\.ts$/.exec(name);
1149
+ if (match) {
1150
+ indices.push(parseInt(match[1], 10));
1151
+ }
1152
+ }
1153
+ if (indices.length < 3) {
1154
+ return null; // need ≥2 completed segments after dropping the newest
1155
+ }
1156
+ indices.sort((a, b) => a - b);
1157
+ const completed = indices.slice(0, -1).slice(-LINK_OBSERVED_SEGMENTS);
1158
+ let bytes = 0;
1159
+ try {
1160
+ for (const index of completed) {
1161
+ const st = await stat(path.join(session.dirPath, `segment-${String(index).padStart(5, "0")}.ts`));
1162
+ bytes += st.size;
1163
+ }
1164
+ } catch {
1165
+ return null; // a segment vanished mid-measure (seek-restart cleanup)
1166
+ }
1167
+ return (bytes * 8) / (completed.length * this.segmentDurationSec) / 1e6;
1168
+ }
1169
+
1170
+ /**
1171
+ * Viewer-link deficit check for one session (adaptive bitrate, part b).
1172
+ * Mirrors the CPU slow-window pattern; shares the action cooldown and the
1173
+ * downshift machinery. Returns true when a downshift was applied this tick.
1174
+ *
1175
+ * @param {HlsSession} session
1176
+ * @param {number} now
1177
+ * @returns {Promise<boolean>}
1178
+ */
1179
+ async #checkLinkBudget(session, now) {
1180
+ const report = session.netReport;
1181
+ if (!report || now - report.at > LINK_REPORT_FRESH_MS) {
1182
+ session.linkSlowSince = 0; // no fresh data — old clients / stopped reporter
1183
+ return false;
1184
+ }
1185
+ if (report.bufferedAheadSec >= LINK_LOW_BUFFER_SEC) {
1186
+ session.linkSlowSince = 0; // viewer is comfortable — nothing to fix
1187
+ return false;
1188
+ }
1189
+ const observed = await this.#observedStreamMbps(session);
1190
+ if (observed === null) {
1191
+ return false; // not enough produced material to compare against
1192
+ }
1193
+ if (report.linkMbps * LINK_SAFETY >= observed) {
1194
+ session.linkSlowSince = 0; // link keeps up
1195
+ return false;
1196
+ }
1197
+ if (session.linkSlowSince === 0) {
1198
+ session.linkSlowSince = now;
1199
+ return false;
1200
+ }
1201
+ if (now - session.linkSlowSince < LINK_SLOW_WINDOW_MS) {
1202
+ return false; // not sustained yet
1203
+ }
1204
+ if (now - session.budgetLastActionAt < BUDGET_ACTION_COOLDOWN_MS) {
1205
+ return false; // let the previous action settle
1206
+ }
1207
+ this.#applyBudgetDownshift(
1208
+ session,
1209
+ `link=${report.linkMbps.toFixed(2)}Mbps stream=${observed.toFixed(2)}Mbps buffer=${report.bufferedAheadSec.toFixed(1)}s`,
1210
+ "link"
1211
+ );
1212
+ session.linkSlowSince = 0;
1213
+ return true;
1214
+ }
1215
+
1090
1216
  async #enforceRealtimeBudget() {
1091
1217
  if (this.videoEncoder?.kind !== "software") {
1092
1218
  return;
@@ -1110,6 +1236,13 @@ export class HlsSessionManager {
1110
1236
  ) {
1111
1237
  continue;
1112
1238
  }
1239
+ // Viewer-link deficit first (adaptive bitrate): independent of encoder
1240
+ // speed — a thin cellular link starves even a faster-than-realtime
1241
+ // encode. When it acts, skip the CPU check this tick (shared cooldown
1242
+ // guards double-firing anyway).
1243
+ if (await this.#checkLinkBudget(session, now)) {
1244
+ continue;
1245
+ }
1113
1246
  const speed = this.#parseSpeed(session.progress?.speed);
1114
1247
  if (speed === null) {
1115
1248
  continue; // no measurement yet
@@ -1143,7 +1276,7 @@ export class HlsSessionManager {
1143
1276
  session.budgetSlowSince = 0; // re-evaluate fresh; don't thrash on this
1144
1277
  continue;
1145
1278
  }
1146
- this.#applyBudgetDownshift(session, speed, bound);
1279
+ this.#applyBudgetDownshift(session, `speed=${speed.toFixed(2)}x`, bound);
1147
1280
  }
1148
1281
  }
1149
1282
 
@@ -1189,11 +1322,11 @@ export class HlsSessionManager {
1189
1322
  * encode at the current segment with the lighter profile.
1190
1323
  *
1191
1324
  * @param {HlsSession} session
1192
- * @param {number} speed - The measured (sub-realtime) speed, for logging.
1193
- * @param {"cpu" | "unknown"} bound
1325
+ * @param {string} reasonText - Measurement summary for the log line.
1326
+ * @param {"cpu" | "unknown" | "link"} bound
1194
1327
  * @returns {void}
1195
1328
  */
1196
- #applyBudgetDownshift(session, speed, bound) {
1329
+ #applyBudgetDownshift(session, reasonText, bound) {
1197
1330
  const nextIndex = session.budgetRungIndex + 1;
1198
1331
  const rung = session.budgetLadder[nextIndex];
1199
1332
  if (!rung) {
@@ -1214,9 +1347,11 @@ export class HlsSessionManager {
1214
1347
  ? session.progress.processedSeconds
1215
1348
  : this.#segmentStartTime(session, head);
1216
1349
  const currentSeg = Math.max(head, this.#segmentIndexForTime(session, processed));
1350
+ const boundLabel =
1351
+ bound === "link" ? "viewer-link-bound" : bound === "unknown" ? "assuming CPU-bound" : "CPU-bound";
1217
1352
  logger.info(
1218
- `[budget] transcode ${session.id} ${bound === "unknown" ? "assuming CPU-bound" : "CPU-bound"} ` +
1219
- `speed=${speed.toFixed(2)}x → downscale to ${rung.width}x${rung.height}/${session.softwarePreset} ` +
1353
+ `[budget] transcode ${session.id} ${boundLabel} ` +
1354
+ `${reasonText} → downscale to ${rung.width}x${rung.height}/${session.softwarePreset} ` +
1220
1355
  `(rung ${nextIndex + 1}/${session.budgetLadder.length}, downshift ${session.budgetDownshifts}/${BUDGET_MAX_DOWNSHIFTS}), ` +
1221
1356
  `restart at segment #${currentSeg} "${session.fileName}"`
1222
1357
  );
@@ -75,6 +75,54 @@ export function chooseOutputFps(sourceFps, cap = MAX_OUTPUT_FPS) {
75
75
  // Software x264 on weak ARM hosts is the transcode bottleneck — use all cores.
76
76
  const CPU_THREADS = Math.max(1, os.cpus().length);
77
77
 
78
+ // Bitrate caps (constrained CRF). CRF stays the quality driver; -maxrate/
79
+ // -bufsize only bound the peaks. Field evidence (iPhone on cellular,
80
+ // 2026-07-10): uncapped complex scenes produced 4 s segments of ~18 Mbit/s
81
+ // against a 1-6 Mbit/s viewer link — 45 s prebuffer, draining buffer.
82
+ // Nominal H.264 rates per rung height; multipliers from webtor's production
83
+ // ladder (content-transcoder): maxrate = 1.3x nominal, bufsize = 1.5x.
84
+ const RUNG_NOMINAL_KBPS = [
85
+ [1080, 5000],
86
+ [720, 2800],
87
+ [480, 1400],
88
+ [360, 800],
89
+ [240, 400]
90
+ ];
91
+ const CAP_MAXRATE_FACTOR = 1.3;
92
+ const CAP_BUFSIZE_FACTOR = 1.5;
93
+
94
+ /**
95
+ * Nominal kbps for an encode height: nearest rung wins (odd heights snap to
96
+ * the closest standard rung; anything above the top rung uses the top one).
97
+ *
98
+ * @param {number} height
99
+ * @returns {number}
100
+ */
101
+ export function nominalKbpsForHeight(height) {
102
+ const h = Number.isFinite(height) && height > 0 ? height : 720;
103
+ let best = RUNG_NOMINAL_KBPS[0];
104
+ for (const rung of RUNG_NOMINAL_KBPS) {
105
+ if (Math.abs(rung[0] - h) < Math.abs(best[0] - h)) {
106
+ best = rung;
107
+ }
108
+ }
109
+ return best[1];
110
+ }
111
+
112
+ /**
113
+ * `-maxrate`/`-bufsize` args for an encode height (constrained CRF).
114
+ *
115
+ * @param {number} height
116
+ * @returns {string[]}
117
+ */
118
+ function bitrateCapArgs(height) {
119
+ const nominal = nominalKbpsForHeight(height);
120
+ return [
121
+ "-maxrate", `${Math.round(nominal * CAP_MAXRATE_FACTOR)}k`,
122
+ "-bufsize", `${Math.round(nominal * CAP_BUFSIZE_FACTOR)}k`
123
+ ];
124
+ }
125
+
78
126
  // libx264 presets to benchmark, ordered slowest/highest-quality → fastest.
79
127
  const BENCHMARK_PRESETS = ["fast", "faster", "veryfast", "superfast", "ultrafast"];
80
128
  const BENCHMARK_REF_W = 640;
@@ -140,6 +188,11 @@ export function softwareDescriptor() {
140
188
  // faster than realtime); falls back to the static default.
141
189
  "-preset", chosenPreset,
142
190
  "-crf", SOFTWARE_CRF,
191
+ // Constrained CRF: bound peak bitrate per rung so a complex scene
192
+ // cannot produce segments a thin viewer link (cellular) can't
193
+ // download in time. Sized by the TARGET box height (the rung the
194
+ // budget/manual selection chose).
195
+ ...bitrateCapArgs(h),
143
196
  "-threads", String(CPU_THREADS),
144
197
  "-pix_fmt", "yuv420p",
145
198
  // Fixed GOP: a keyframe exactly every (segmentDurationSec × fps) frames,