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.
This commit is contained in:
2026-08-28 02:37:56 +02:00
parent b7a6451dba
commit 9ae6c47d6c
9 changed files with 235 additions and 6 deletions
+32
View File
@@ -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)