fix(engine): treat ffmpegStreamingTimeout as per-frame inactivity, not total render time (#901)

## Summary

- Convert `streamingEncoder.ts`'s safety timer from a total-render hard cap to a per-frame inactivity timeout
- Reset the timer only on `accepted === true` writes — buffered writes don't count as consumer progress
- Update the `ffmpegStreamingTimeout` config doc to reflect the new semantics

## The bug

The timer was set once at spawn and fired SIGTERM unconditionally at `ffmpegStreamingTimeout` ms — turning a "FFmpeg is hung" guard into a hard cap on total render duration. Slow-but-progressing captures (CI runner under load, large compositions, slower compositor paths after [#838](https://github.com/heygen-com/hyperframes/pull/838)'s always-clip change) regularly exceeded the 600s default and were killed mid-encode. The symptom surfaced as:

```
Streaming encode failed: FFmpeg exited with code 255
video:NNNkB audio:0kB ...
[libx264 @ ...] frame I:3  Avg QP:12.91  size: 73263
[libx264 @ ...] frame P:431 Avg QP:14.72  size: 31633
...
[libx264 @ ...] kb/s:7661.05
Exiting normally, received signal 15.
```

libx264 had encoded most frames cleanly; SIGTERM arrived during the encode, libx264 printed its end-of-encode stats, and Node observed a non-zero exit. The `audio:0kB` in stderr is incidental — `streamingEncoder` is video-only; audio is muxed later in `assembleStage`.

Downstream reproduction: `style-13-prod` fails deterministically in `heygen-com/hyperframes-internal` CI after bumping `@hyperframes/producer` from 0.6.7 → 0.6.10. Bisects to #838 widening the SDR capture path at dpr=1 — same composition shape, slower per-frame, total render now crosses 600s.

## The fix

Convert the timer to a heartbeat: each `writeFrame` that goes through to the kernel pipe (i.e. `stdin.write` returns `true`) resets it. Only true hangs (no successful frame write for the timeout window) trip SIGTERM now; "slow but progressing" renders are unbounded.

Crucially, the heartbeat does **not** reset on `accepted === false`. A `false` return means Node had to buffer the write because FFmpeg hasn't drained the pipe yet — that's not proof of consumer progress, just proof we produced. Without this distinction, a hung FFmpeg with a live Chrome would queue frames into Node's writable buffer indefinitely (no backpressure path back to the capture loop) and grow until OOM. In steady state with a slow-but-alive FFmpeg, writes alternate between `true` and `false` as the buffer drains and refills; the `true`s are enough to keep the heartbeat ticking.

Renames are intentionally avoided — `ffmpegStreamingTimeout` keeps its name and `600_000` default; only the semantics changed. The config doc spells out the new behavior so downstream consumers know what 600s now means.

## Test plan

- [x] **Slow-but-progressing capture** (`accepted=true`): 9× `writeFrame` at 900ms intervals (under the 1000ms threshold) — encoder stays alive through 8.1s. Stall past the threshold — SIGTERM fires.
- [x] **Stalled FFmpeg with live producer** (`accepted=false`): override `stdin.write` to return false; pump 9× `writeFrame` at 900ms intervals. SIGTERM still fires inside the 1000ms window — buffered writes don't keep the heartbeat alive.
- [x] Existing 33 tests in `streamingEncoder.test.ts` still pass
- [x] Lint (`oxlint`) + format (`oxfmt --check`) clean
- [ ] CI regression suite

🤖 Generated with [Claude Code](https://claude.com/claude-code)

packages/engine/src/config.ts

@@ -94,7 +94,12 @@ export interface EngineConfig {
   ffmpegEncodeTimeout: number;
   /** Timeout for FFmpeg mux/faststart processes (ms). Default: 300_000 */
   ffmpegProcessTimeout: number;
-  /** Timeout for FFmpeg streaming encode (ms). Default: 600_000 */
+  /**
+   * Inactivity timeout for FFmpeg streaming encode (ms). The timer resets on
+   * every successful `writeFrame` call, so this caps the duration of a
+   * single "no frame arrived" gap (capture hang, dead Chrome), not the total
+   * render time. Default: 600_000 (10 minutes without any frame = dead).
+   */
   ffmpegStreamingTimeout: number;
 
   // ── HDR ──────────────────────────────────────────────────────────────

packages/engine/src/services/streamingEncoder.test.ts

@@ -572,4 +572,71 @@ describe("spawnStreamingEncoder lifecycle and cleanup", () => {
     controller.abort();
     expect(proc.kill).not.toHaveBeenCalled();
   });
+
+  it("inactivity timeout fires only after a no-frame gap exceeds ffmpegStreamingTimeout", async () => {
+    vi.useFakeTimers();
+    try {
+      const { spawn, calls } = createSpawnSpy();
+      vi.resetModules();
+      vi.doMock("child_process", () => ({ spawn }));
+
+      const { spawnStreamingEncoder } = await import("./streamingEncoder.js");
+      const dir = mkdtempSync(join(tmpdir(), "se-heartbeat-"));
+      const encoder = await spawnStreamingEncoder(join(dir, "out.mp4"), baseOptions, undefined, {
+        ffmpegStreamingTimeout: 1000,
+      });
+
+      const proc = calls[0]!.proc;
+
+      // Frames every 900ms — under the 1000ms inactivity threshold — should
+      // keep resetting the timer. After 9× 900ms = 8.1s of "slow but
+      // progressing" capture the encoder must still be alive. The old total-
+      // render timeout would have fired SIGTERM at ~1000ms.
+      for (let i = 0; i < 9; i++) {
+        encoder.writeFrame(Buffer.from([i]));
+        vi.advanceTimersByTime(900);
+      }
+      expect(proc.kill).not.toHaveBeenCalled();
+
+      // Now stall — no writeFrame for longer than the threshold. SIGTERM fires.
+      vi.advanceTimersByTime(1100);
+      expect(proc.kill).toHaveBeenCalledWith("SIGTERM");
+    } finally {
+      vi.useRealTimers();
+    }
+  });
+
+  it("inactivity timeout still fires when stdin is backpressured (stalled ffmpeg, live producer)", async () => {
+    vi.useFakeTimers();
+    try {
+      // Simulate the FFmpeg-hangs-but-Chrome-keeps-producing case: stdin.write
+      // always returns false (Node has to buffer because ffmpeg isn't draining
+      // the pipe). The heartbeat must NOT reset on those buffered writes —
+      // otherwise a hung ffmpeg with a steady frame producer would never
+      // SIGTERM and we'd grow Node's stdin buffer until OOM.
+      const { spawn, calls } = createSpawnSpy();
+      vi.resetModules();
+      vi.doMock("child_process", () => ({ spawn }));
+
+      const { spawnStreamingEncoder } = await import("./streamingEncoder.js");
+      const dir = mkdtempSync(join(tmpdir(), "se-backpressure-"));
+      const encoder = await spawnStreamingEncoder(join(dir, "out.mp4"), baseOptions, undefined, {
+        ffmpegStreamingTimeout: 1000,
+      });
+
+      const proc = calls[0]!.proc;
+      proc.stdin.write = (_chunk: Buffer) => false;
+
+      // Pump 9 frames at 900ms intervals — all returning false. The reset
+      // should NOT fire (every write was buffered, not accepted), so the
+      // 1000ms timer (last reset on spawn) elapses near the start.
+      for (let i = 0; i < 9; i++) {
+        encoder.writeFrame(Buffer.from([i]));
+        vi.advanceTimersByTime(900);
+      }
+      expect(proc.kill).toHaveBeenCalledWith("SIGTERM");
+    } finally {
+      vi.useRealTimers();
+    }
+  });
 });

packages/engine/src/services/streamingEncoder.ts

@@ -416,13 +416,26 @@ export async function spawnStreamingEncoder(
     }
   }
 
-  // Timeout safety
+  // Inactivity timeout: fires only when no frame has been written for
+  // `ffmpegStreamingTimeout` ms. A slow-but-progressing capture (e.g. a CI
+  // runner under load) keeps resetting the timer on each writeFrame, so total
+  // wall-clock render time is unbounded — only a true hang (Chrome dead,
+  // capture stuck, no frames arriving) trips SIGTERM. The 600s default was
+  // previously a total-render cap, which intermittently killed legitimate
+  // slow renders mid-encode (FFmpeg got SIGTERM after most frames were sent;
+  // libx264 printed its summary and exited 255, observable as
+  // "Streaming encode failed: FFmpeg exited with code 255" with audio:0kB).
   const streamingTimeout = config?.ffmpegStreamingTimeout ?? DEFAULT_CONFIG.ffmpegStreamingTimeout;
-  const timer = setTimeout(() => {
+  let timer: NodeJS.Timeout | null = null;
-    if (exitStatus === "running") {
+  const resetTimer = () => {
-      ffmpeg.kill("SIGTERM");
+    if (timer) clearTimeout(timer);
-    }
+    timer = setTimeout(() => {
-  }, streamingTimeout);
+      if (exitStatus === "running") {
+        ffmpeg.kill("SIGTERM");
+      }
+    }, streamingTimeout);
+  };
+  resetTimer();
 
   const encoder: StreamingEncoder = {
     writeFrame: (buffer: Buffer): boolean => {
@@ -433,11 +446,20 @@ export async function spawnStreamingEncoder(
       // provided buffer and drain it asynchronously. The HDR path's compositor
       // reuses pre-allocated transOutput/normalCanvas buffers across frames,
       // so without this copy the pipe would read partially-overwritten data
-      // and flicker. The SDR path doesn't invoke writeFrame at all (it pipes
+      // and flicker.
-      // PNG files via encodeFramesFromDir), so the memcpy here is HDR-only
-      // and justified by correctness.
       const copy = Buffer.from(buffer);
-      return ffmpeg.stdin.write(copy);
+      const accepted = ffmpeg.stdin.write(copy);
+      // Reset inactivity timer ONLY on `accepted === true`. `true` means the
+      // write went through to the kernel pipe without buffering in Node —
+      // proof FFmpeg is actually consuming. `false` means Node's writable
+      // stream had to buffer (FFmpeg hasn't drained the pipe yet); we deliberately
+      // don't reset on `false` so a hung FFmpeg with a still-producing Chrome
+      // can't keep us alive forever while Node's stdin buffer grows to OOM. In
+      // steady state with a slower-but-alive FFmpeg, writes alternate between
+      // true and false as the buffer drains and refills; the trues are enough
+      // to keep the heartbeat ticking.
+      if (accepted) resetTimer();
+      return accepted;
     },
 
     close: async (): Promise<StreamingEncoderResult> => {
@@ -455,7 +477,10 @@ export async function spawnStreamingEncoder(
       // repeated calls. If you change this method, preserve idempotency or
       // a regression here will silently double-close ffmpeg and produce
       // harder-to-trace errors at the orchestrator layer.
-      clearTimeout(timer);
+      if (timer) {
+        clearTimeout(timer);
+        timer = null;
+      }
       if (signal) signal.removeEventListener("abort", onAbort);
 
       const stdin = ffmpeg.stdin;