goon/scripts/backfill_xvideos_performers.py

"""Backfill performerów dla browse-scraped scen xvideos z 0 performerami.

Kontekst: `_MODEL_RE` w xvideos_browse.py przez pewien czas nie pasował do markupu
xvideos (nazwa modela w zagnieżdżonym `<span class="name">`, nie jako tekst anchora)
→ część scen wpadła z 0 performerów (bug-report 2026-06-07 "czemu nie ma aktorów,
są na stronie: Rebecca Johnson"). Forward-fix poprawił parser; ten skrypt domyka
zaległe sceny re-fetchem strony + dowiązaniem `/models/` przez ten sam resolver co
ingest (więc nazwy mergują z canonical po name_normalized).

Sample 2026-06-08: 54% zero-perf scen ma realny /models/ na stronie (~1.4 perf/scenę),
46% to amatorskie uploady bez modela (nie ruszamy). 0 fetch-failów (VPS niezablokowany).

Użycie (w kontenerze worker):
  python scripts/backfill_xvideos_performers.py [LIMIT] [--commit] [--workers N] [--sleep S]
Bez --commit = dry-run (tylko liczy yield). LIMIT pusty/0 = wszystkie zero-perf sceny.
"""
from __future__ import annotations

import logging
import sys
import threading
import time
from concurrent.futures import ThreadPoolExecutor

from sqlalchemy import select, text

from app.connectors.direct_scrapers.xvideos_browse import XVideosBrowseScraper
from app.db import session_scope
from app.extractors import browser_get
from app.models.scene import ScenePerformer
from app.models.source import Source
from app.normalize.scenes import normalize_performer
from app.resolve.performer_resolver import resolve_performer

logging.basicConfig(level=logging.WARNING, format="%(asctime)s %(message)s")
log = logging.getLogger("backfill_xvideos")

_scraper = XVideosBrowseScraper()
_lock = threading.Lock()


def _args() -> tuple[int, bool, int, float]:
    limit = 0
    commit = "--commit" in sys.argv
    workers = 3
    sleep = 0.3
    for i, a in enumerate(sys.argv[1:], 1):
        if a.isdigit():
            limit = int(a)
        elif a == "--workers" and i < len(sys.argv) - 1:
            workers = int(sys.argv[i + 1])
        elif a == "--sleep" and i < len(sys.argv) - 1:
            sleep = float(sys.argv[i + 1])
    return limit, commit, workers, sleep


def _candidates(limit: int) -> list[tuple]:
    sql = (
        "SELECT s.id, ps.page_url FROM scenes s "
        "JOIN playback_sources ps ON ps.scene_id=s.id AND ps.origin='tube:xvideoscom' "
        "WHERE NOT EXISTS (SELECT 1 FROM scene_performers sp WHERE sp.scene_id=s.id) "
        "ORDER BY s.created_at DESC"
    )
    if limit:
        sql += f" LIMIT {limit}"
    with session_scope() as s:
        return list(s.execute(text(sql)).all())


def _source_id() -> "uuid.UUID":  # noqa: F821
    with session_scope() as s:
        return s.execute(select(Source.id).where(Source.name == "tube-scraper")).scalar_one()


_stats = {"gain": 0, "perf": 0, "nomodel": 0, "fail": 0, "done": 0}


def _process(row, *, commit: bool, src_id, sleep: float) -> None:
    scene_id, page_url = row
    try:
        html = browser_get(page_url, timeout=20).text
    except Exception:
        with _lock:
            _stats["fail"] += 1
        return
    rs = _scraper._parse_detail(page_url, html)
    perfs = rs.performers if rs else []
    if not perfs:
        with _lock:
            _stats["nomodel"] += 1; _stats["done"] += 1
        return
    if commit:
        with session_scope() as s:
            # re-check: ktoś mógł w międzyczasie dowiązać
            still_zero = not s.execute(
                select(ScenePerformer.scene_id).where(ScenePerformer.scene_id == scene_id).limit(1)
            ).first()
            if still_zero:
                for pos, rp in enumerate(perfs):
                    perf = resolve_performer(s, norm=normalize_performer(rp), source_id=src_id)
                    exists = s.execute(
                        select(ScenePerformer).where(
                            ScenePerformer.scene_id == scene_id,
                            ScenePerformer.performer_id == perf.id,
                        )
                    ).first()
                    if not exists:
                        s.add(ScenePerformer(scene_id=scene_id, performer_id=perf.id, position=pos))
    with _lock:
        _stats["gain"] += 1; _stats["perf"] += len(perfs); _stats["done"] += 1
    if sleep:
        time.sleep(sleep)


def main() -> None:
    limit, commit, workers, sleep = _args()
    rows = _candidates(limit)
    src_id = _source_id() if commit else None
    print(f"candidates={len(rows)} commit={commit} workers={workers} sleep={sleep}", flush=True)
    with ThreadPoolExecutor(max_workers=workers) as ex:
        futs = [ex.submit(_process, r, commit=commit, src_id=src_id, sleep=sleep) for r in rows]
        last = 0
        for _ in futs:
            pass
        # progress poll
        while any(not f.done() for f in futs):
            time.sleep(15)
            d = _stats["done"]
            if d != last:
                print(f"  progress done={d}/{len(rows)} gain={_stats['gain']} perf={_stats['perf']} nomodel={_stats['nomodel']} fail={_stats['fail']}", flush=True)
                last = d
    print(f"DONE gain={_stats['gain']} perf={_stats['perf']} nomodel={_stats['nomodel']} fail={_stats['fail']}", flush=True)


if __name__ == "__main__":
    main()