perf(downloads): de-prioritize worker transcodes + idle-back-off polls to stop LAN slowdown

The download/edit child (yt-dlp + ffmpeg) shares one LAN box with the API,
Postgres, and Plex playback. A CPU re-encode pegged every core for ~30s and
starved web requests + playback until it finished. Full-scope S4 mitigations:

- Child self-de-prioritizes at startup (inherited by every ffmpeg/yt-dlp
  descendant): CPU nice increment + best-effort idle ionice class + an
  OMP_NUM_THREADS budget, all config-gated and independently disable-able.
- Cap the browser-safe re-encode's encoder threads (`-threads`) so a transcode
  reserves a core for foreground work instead of claiming all of them.
- Throttle the retention GC's whole-tree orphan sweep (yield IO after each
  batch of files) so its rglob can't monopolize disk IO in the API process.
- Idle-back-off the Download Center (2s→15s when nothing queued/running) and
  UsageBar (2s→15s when running_jobs is 0) polls; both snap back to 2s on the
  next fetch after an enqueue (invalidateDownloads refetches immediately).

Tests: encode_thread_cap (pin/reserve-core/edge), -threads flag presence,
_deprioritize (nice/ionice/OMP branches), GC sweep throttle batching.

backend/app/config.py

@@ -186,6 +186,21 @@ class Settings(BaseSettings):
     download_supervisor_poll_seconds: int = 3
     # Grace between SIGTERM and SIGKILL when stopping a child (matches run_ffmpeg's escalation).
     download_kill_grace_seconds: int = 5
+    # --- worker-child resource de-prioritization ---
+    # The download/edit child (yt-dlp + ffmpeg) shares ONE LAN box with the API, Postgres, and Plex
+    # playback. Without a priority handicap a CPU re-encode pegs every core for ~30s and starves web
+    # requests + playback until it finishes. The child lowers its OWN priority at startup — inherited
+    # by every ffmpeg/yt-dlp descendant — so any foreground work always wins. Each knob is best-effort
+    # and independently disable-able with 0.
+    download_child_nice: int = 10  # CPU nice increment applied to the child; 0 = don't renice
+    download_child_ionice_class: int = 3  # best-effort IO class (3 = idle); 0 = leave IO untouched
+    # Cap the browser-safe re-encode's encoder threads so a transcode can't claim every core. 0 =
+    # reserve one core for foreground work (max(1, cpu_count - 1)); >0 pins exactly that many.
+    download_ffmpeg_threads: int = 0
+    # Throttle the retention GC's whole-tree orphan file sweep so its rglob can't monopolize disk IO
+    # in the API process: sleep `pause_ms` after every `batch` files touched. pause_ms=0 disables.
+    download_gc_sweep_pause_ms: int = 20
+    download_gc_sweep_batch: int = 400
     # Total cache size cap across ALL users (bytes). When >0 and exceeded, the GC evicts the
     # least-recently-accessed ready assets until under it. 0 = unlimited (disk is the limit).
     download_total_max_bytes: int = 0

backend/app/downloads/edit.py

@@ -25,6 +25,7 @@ import hashlib
 import json
 import logging
 import math
+import os
 import queue
 import shutil
 import subprocess
@@ -32,9 +33,21 @@ import threading
 import time
 from pathlib import Path
 
+from app.config import settings
+
 log = logging.getLogger("siftlode.edit")
 
 
+def encode_thread_cap() -> int:
+    """How many encoder threads a browser-safe re-encode may use. `download_ffmpeg_threads` when the
+    admin pinned one (>0), else reserve a core for foreground work (max(1, cpu_count - 1)). Shared by
+    the ffmpeg `-threads` flag and the child's `OMP_NUM_THREADS` so both honour the same budget."""
+    n = settings.download_ffmpeg_threads
+    if n > 0:
+        return n
+    return max(1, (os.cpu_count() or 2) - 1)
+
+
 class EditAborted(Exception):
     """Raised by run_ffmpeg when the job was paused/canceled mid-encode."""
 
@@ -510,11 +523,12 @@ def browser_transcode_flags(
 
 
 def build_browser_safe_cmd(
-    src: Path, dest: Path, reencode_video: bool, reencode_audio: bool
+    src: Path, dest: Path, reencode_video: bool, reencode_audio: bool, threads: int | None = None
 ) -> list[str]:
     """ffmpeg command that lands `src` as a browser-safe mp4 (faststart): re-encode only the
     stream(s) that need it (libx264 / aac), stream-copy the rest, always into mp4. `-progress
-    pipe:1` streams machine-readable progress for run_ffmpeg."""
+    pipe:1` streams machine-readable progress for run_ffmpeg. `threads` caps the encoder so a
+    re-encode can't claim every core on the shared box (see the worker-child de-prioritization)."""
     cmd = [
         "ffmpeg",
         "-y",
@@ -534,6 +548,8 @@ def build_browser_safe_cmd(
         else ["-c:v", "copy"]
     )
     cmd += ["-c:a", "aac", "-b:a", "192k"] if reencode_audio else ["-c:a", "copy"]
+    if threads and threads > 0:
+        cmd += ["-threads", str(threads)]
     cmd += ["-movflags", "+faststart", str(dest)]
     return cmd
 

backend/app/downloads/gc.py

@@ -86,11 +86,25 @@ def _sweep_orphan_files(db: Session, download_root: str) -> int:
 
     cutoff = time.time() - _ORPHAN_MIN_AGE_S
     removed = 0
+    # Throttle the whole-tree walk: this runs in the API process, and an unpaced rglob+stat over a
+    # large download root monopolizes disk IO long enough to make web requests + playback stutter.
+    # Yield the IO channel after every `batch` files touched so foreground work interleaves.
+    pause_s = max(0, settings.download_gc_sweep_pause_ms) / 1000.0
+    batch = max(1, settings.download_gc_sweep_batch)
+    touched = 0
+
+    def _throttle() -> None:
+        nonlocal touched
+        touched += 1
+        if pause_s and touched % batch == 0:
+            time.sleep(pause_s)
+
     for entry in root.iterdir():
         if entry.name.startswith("."):  # .edits / .plex-* / .staging / .storyboards + any dotfile
             continue
         files = [entry] if entry.is_file() else [p for p in entry.rglob("*") if p.is_file()]
         for f in files:
+            _throttle()
             try:
                 if f.resolve() in keep or f.stat().st_mtime > cutoff:
                     continue

backend/app/downloads/pipeline.py

@@ -381,7 +381,9 @@ def _ensure_browser_playable(job_id: int, produced: Path, spec: dict, info: dict
     # re-encode still shows real progress instead of a frozen 0%.
     total_s = int(info["duration"]) if info.get("duration") else editmod.probe_duration(produced)
     editmod.run_ffmpeg(
-        editmod.build_browser_safe_cmd(produced, dest, reenc_v, reenc_a),
+        editmod.build_browser_safe_cmd(
+            produced, dest, reenc_v, reenc_a, threads=editmod.encode_thread_cap()
+        ),
         total_s=total_s,
         on_progress=_ffmpeg_progress_writer(job_id, "processing"),
         should_cancel=_cancel_poll(job_id),

backend/app/downloads/runner.py

@@ -12,16 +12,48 @@ progress hook), then exits 0. An unexpected crash exits non-zero and the parent
 """
 
 import logging
+import os
+import shutil
+import subprocess
 import sys
 
+from app.config import settings
 from app.db import SessionLocal
+from app.downloads import edit as editmod
 from app.models import DownloadJob
 
 log = logging.getLogger("siftlode.worker.runner")
 
 
+def _deprioritize() -> None:
+    """Lower THIS child's CPU + IO priority so its yt-dlp/ffmpeg work — every descendant inherits the
+    handicap — can never starve the co-located API, Postgres, or Plex playback. Run once at startup,
+    before any heavy work. Every step is best-effort: a failed renice must never fail a download."""
+    nice_inc = settings.download_child_nice
+    if nice_inc > 0:
+        try:
+            os.nice(nice_inc)  # relative increment; the child starts at the worker's nice (0)
+        except OSError as exc:
+            log.info("could not renice download child: %s", str(exc)[:120])
+    cls = settings.download_child_ionice_class
+    if cls > 0 and shutil.which("ionice"):  # util-linux; absent on some minimal images
+        try:
+            subprocess.run(
+                ["ionice", "-c", str(cls), "-p", str(os.getpid())],
+                capture_output=True,
+                timeout=5,
+                check=False,
+            )
+        except (OSError, subprocess.SubprocessError) as exc:
+            log.info("could not ionice download child: %s", str(exc)[:120])
+    # Cap OpenMP-parallel ffmpeg filters to the same budget as the encoder threads, so a filter
+    # graph can't fan out across every core behind ffmpeg's own `-threads` cap.
+    os.environ.setdefault("OMP_NUM_THREADS", str(editmod.encode_thread_cap()))
+
+
 def main() -> None:
     logging.basicConfig(level=logging.INFO)
+    _deprioritize()
     if len(sys.argv) != 2:
         print("usage: python -m app.downloads.runner <job_id>", file=sys.stderr)
         raise SystemExit(2)

backend/tests/test_download_deprioritize.py

@@ -0,0 +1,103 @@
+"""Worker-child resource de-prioritization (S4): the encoder thread cap, the `-threads` flag on the
+browser-safe re-encode, and the child's startup renice/ionice/OMP handicap. These keep a CPU-heavy
+transcode from pegging every core on the box it shares with the API, Postgres, and Plex playback."""
+
+import subprocess
+
+import pytest
+
+from app.config import settings
+from app.downloads import edit as editmod
+from app.downloads import runner
+
+
+class TestEncodeThreadCap:
+    def test_pinned_setting_wins(self, monkeypatch):
+        monkeypatch.setattr(settings, "download_ffmpeg_threads", 3)
+        assert editmod.encode_thread_cap() == 3
+
+    def test_zero_reserves_one_core(self, monkeypatch):
+        monkeypatch.setattr(settings, "download_ffmpeg_threads", 0)
+        monkeypatch.setattr(editmod.os, "cpu_count", lambda: 8)
+        assert editmod.encode_thread_cap() == 7
+
+    def test_single_core_box_never_caps_below_one(self, monkeypatch):
+        monkeypatch.setattr(settings, "download_ffmpeg_threads", 0)
+        monkeypatch.setattr(editmod.os, "cpu_count", lambda: 1)
+        assert editmod.encode_thread_cap() == 1
+
+    def test_unknown_cpu_count_falls_back(self, monkeypatch):
+        monkeypatch.setattr(settings, "download_ffmpeg_threads", 0)
+        monkeypatch.setattr(editmod.os, "cpu_count", lambda: None)
+        assert editmod.encode_thread_cap() == 1  # max(1, 2 - 1)
+
+
+class TestBrowserSafeThreads:
+    def test_threads_flag_added_when_capped(self):
+        cmd = editmod.build_browser_safe_cmd(
+            "/in.mkv", "/out.mp4", reencode_video=True, reencode_audio=False, threads=2
+        )
+        assert "-threads" in cmd
+        assert cmd[cmd.index("-threads") + 1] == "2"
+        # The cap precedes the output file, so it binds the encoder (not input decoding).
+        assert cmd.index("-threads") < cmd.index("/out.mp4")
+
+    @pytest.mark.parametrize("threads", [0, None])
+    def test_no_threads_flag_when_uncapped(self, threads):
+        cmd = editmod.build_browser_safe_cmd(
+            "/in.mkv", "/out.mp4", reencode_video=True, reencode_audio=False, threads=threads
+        )
+        assert "-threads" not in cmd
+
+
+class TestDeprioritize:
+    def test_renices_by_configured_increment(self, monkeypatch):
+        monkeypatch.setattr(settings, "download_child_nice", 12)
+        monkeypatch.setattr(settings, "download_child_ionice_class", 0)  # skip ionice branch
+        calls = []
+        monkeypatch.setattr(runner.os, "nice", lambda n: calls.append(n))
+        monkeypatch.setattr(runner.os, "environ", {})
+        runner._deprioritize()
+        assert calls == [12]
+
+    def test_nice_zero_does_not_renice(self, monkeypatch):
+        monkeypatch.setattr(settings, "download_child_nice", 0)
+        monkeypatch.setattr(settings, "download_child_ionice_class", 0)
+        calls = []
+        monkeypatch.setattr(runner.os, "nice", lambda n: calls.append(n))
+        monkeypatch.setattr(runner.os, "environ", {})
+        runner._deprioritize()
+        assert calls == []
+
+    def test_sets_omp_budget_to_thread_cap(self, monkeypatch):
+        monkeypatch.setattr(settings, "download_child_nice", 0)
+        monkeypatch.setattr(settings, "download_child_ionice_class", 0)
+        monkeypatch.setattr(settings, "download_ffmpeg_threads", 4)
+        env: dict = {}
+        monkeypatch.setattr(runner.os, "environ", env)
+        runner._deprioritize()
+        assert env["OMP_NUM_THREADS"] == "4"
+
+    def test_ionice_invoked_when_binary_present(self, monkeypatch):
+        monkeypatch.setattr(settings, "download_child_nice", 0)
+        monkeypatch.setattr(settings, "download_child_ionice_class", 3)
+        monkeypatch.setattr(runner.os, "environ", {})
+        monkeypatch.setattr(runner.shutil, "which", lambda name: "/usr/bin/ionice")
+        ran = []
+        monkeypatch.setattr(
+            runner.subprocess,
+            "run",
+            lambda cmd, **kw: ran.append(cmd) or subprocess.CompletedProcess(cmd, 0),
+        )
+        runner._deprioritize()
+        assert ran and ran[0][:3] == ["ionice", "-c", "3"]
+
+    def test_ionice_skipped_when_binary_absent(self, monkeypatch):
+        monkeypatch.setattr(settings, "download_child_nice", 0)
+        monkeypatch.setattr(settings, "download_child_ionice_class", 3)
+        monkeypatch.setattr(runner.os, "environ", {})
+        monkeypatch.setattr(runner.shutil, "which", lambda name: None)
+        ran = []
+        monkeypatch.setattr(runner.subprocess, "run", lambda cmd, **kw: ran.append(cmd))
+        runner._deprioritize()
+        assert ran == []

backend/tests/test_gc.py

@@ -5,6 +5,8 @@ import os
 import time
 from pathlib import Path
 
+from app.config import settings
+from app.downloads import gc as gcmod
 from app.downloads.gc import _sweep_orphan_files
 
 
@@ -69,3 +71,38 @@ def test_sweep_removes_orphans_but_keeps_owned_system_and_fresh(tmp_path: Path):
     assert not (root / "OldChan").exists()
     assert sysfile.exists()  # system tree untouched
     assert fresh.exists()  # too new to sweep
+
+
+def test_sweep_throttles_after_each_batch(tmp_path: Path, monkeypatch):
+    """The whole-tree walk yields the IO channel after every `batch` files so it can't monopolize
+    disk IO in the API process. batch=2 over 5 touched files → 2 pauses (at file 2 and 4)."""
+    root = tmp_path
+    d = root / "Chan"
+    d.mkdir()
+    for i in range(5):
+        (d / f"orphan_{i}.dat").write_bytes(b"x")
+    for f in d.iterdir():
+        _aged(f, 5)  # old enough that every file is "touched" by the walk
+
+    monkeypatch.setattr(settings, "download_gc_sweep_pause_ms", 10)
+    monkeypatch.setattr(settings, "download_gc_sweep_batch", 2)
+    pauses = []
+    monkeypatch.setattr(gcmod.time, "sleep", lambda s: pauses.append(s))
+
+    _sweep_orphan_files(_StubDB([]), str(root))
+
+    assert pauses == [0.01, 0.01]  # 5 files, batch 2 → yield at #2 and #4
+
+
+def test_sweep_no_throttle_when_pause_disabled(tmp_path: Path, monkeypatch):
+    (tmp_path / "Chan").mkdir()
+    f = tmp_path / "Chan" / "orphan.dat"
+    f.write_bytes(b"x")
+    _aged(f, 5)
+    monkeypatch.setattr(settings, "download_gc_sweep_pause_ms", 0)
+    pauses = []
+    monkeypatch.setattr(gcmod.time, "sleep", lambda s: pauses.append(s))
+
+    _sweep_orphan_files(_StubDB([]), str(tmp_path))
+
+    assert pauses == []

frontend/src/components/DownloadCenter.tsx

@@ -51,7 +51,13 @@ export default function DownloadCenter() {
   const [editJob, setEditJob] = useState<DownloadJob | null>(null);
   const [errorJob, setErrorJob] = useState<DownloadJob | null>(null);
 
-  const jobsQ = useLiveQuery(qk.downloads(), api.downloads, { intervalMs: 2000 });
+  // Poll fast (2s) while work is actually in flight; back off to 15s once the queue is settled
+  // (only done/error/paused rows left, which don't advance on their own) so an idle Download Center
+  // isn't hitting the API — and adding to the shared box's load — every 2s for nothing.
+  const jobsQ = useLiveQuery(qk.downloads(), api.downloads, {
+    intervalMs: (data) =>
+      data?.some((j) => j.status === "queued" || j.status === "running") ? 2000 : 15000,
+  });
   const sharedQ = useQuery({
     queryKey: qk.downloadsShared(),
     queryFn: api.downloadsShared,

frontend/src/components/downloadRow.tsx

@@ -278,8 +278,12 @@ export function UsageBar() {
   const { t } = useTranslation();
   // Poll live (like the library list) so the footprint updates the moment the worker finishes an
   // edit/download — enqueue only creates a queued job (0 bytes), the size lands asynchronously, so
-  // a one-shot fetch at enqueue time would show a stale 0 B until a manual reload.
-  const usage = useLiveQuery(qk.downloadUsage(), api.downloadUsage, { intervalMs: 2000 });
+  // a one-shot fetch at enqueue time would show a stale 0 B until a manual reload. The footprint
+  // only moves while a job is in flight, so poll fast (2s) then and idle-back-off to 15s otherwise
+  // (running_jobs counts queued+running — it flips back to 2s on the first poll after a new enqueue).
+  const usage = useLiveQuery(qk.downloadUsage(), api.downloadUsage, {
+    intervalMs: (u) => (u?.running_jobs ? 2000 : 15000),
+  });
   const u = usage.data;
   if (!u) return null;
   const pct =