fix: RSS matching, proxy pool, parser, healthcheck and log issues (#1103 #1104 #1105 #1106 #1108 #1114 #1117)

- match_bangumi_in_list: also accept a torrent whose rebuilt title equals the stored title_raw (#1103, #1114) - download_bangumi / collect_season: drop items whose season or type conflicts with the bangumi (#1105) - request_url: swap in a fresh shared client on PoolTimeout without closing the old one (#1104) - classic parser: Unicode Roman numeral season, strip fullwidth dash wrapper (#1108) - Docker HEALTHCHECK: healthcheck.py reads the configured WebUI port and probes IPv4 then IPv6 loopback (#1106) - scheduler: non-positive interval falls back to 1s; deploy docs cap container logs at 10 MB x 3 (#1117) Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KU9GqKzULrNb7MN5T9TYix
2026-11-04 06:38:05 +08:00 · 2026-09-19 17:58:17 +02:00
parent 5cc2fabc64
commit 37fa6b4ea5
25 changed files with 284 additions and 5 deletions
--- a/4
+++ b/4
@@ -63,7 +63,9 @@ VOLUME ["/app/config", "/app/data"]

 # Liveness probe against the unauthenticated /health route; give the app
 # room to run migrations on first boot before failures count.
+# healthcheck.py reads the configured WebUI port and probes IPv4 then IPv6
+# loopback (#1106).
 HEALTHCHECK --interval=30s --timeout=5s --start-period=30s --retries=3 \
-    CMD wget -qO- http://localhost:7892/health || exit 1
+    CMD python /app/healthcheck.py

 ENTRYPOINT ["tini", "-g", "--", "/entrypoint.sh"]
--- a/backend/src/healthcheck.py
+++ b/backend/src/healthcheck.py
@@ -0,0 +1,40 @@
+"""Docker HEALTHCHECK：探测 /health，退出码 0 表示健康。
+
+端口取自 config.json（WebUI 可改，AB_WEBUI_PORT 只在首次启动时写入），
+不能写死 7892。先连 IPv4 回环再连 IPv6：busybox wget 把 localhost 解析成
+::1，而默认只监听 0.0.0.0（#1106）；IPV6/HOST=::1 时只有 ::1 可达。
+只用标准库，避免每 30 秒导入整个应用。
+"""
+
+import json
+import sys
+import urllib.request
+from pathlib import Path
+
+CONFIG_PATH = Path("config/config.json")
+DEFAULT_PORT = 7892
+
+
+def webui_port(config_path: Path = CONFIG_PATH) -> int:
+    try:
+        config = json.loads(config_path.read_text(encoding="utf-8"))
+        return int(config["program"]["webui_port"])
+    except (OSError, ValueError, KeyError, TypeError):
+        return DEFAULT_PORT
+
+
+def main() -> int:
+    port = webui_port()
+    # 忽略 HTTP_PROXY 等环境变量：回环探测不能走代理
+    opener = urllib.request.build_opener(urllib.request.ProxyHandler({}))
+    for host in ("127.0.0.1", "[::1]"):
+        try:
+            with opener.open(f"http://{host}:{port}/health", timeout=2):
+                return 0
+        except OSError:
+            continue
+    return 1
+
+
+if __name__ == "__main__":
+    sys.exit(main())
--- a/backend/src/module/core/scheduler.py
+++ b/backend/src/module/core/scheduler.py
@@ -70,7 +70,9 @@ class PeriodicTask:
                    raise
                except Exception as e:
                    logger.error("%s tick failed: %s", self._name, e)
-                if await self._wait(self._interval()):
+                # 间隔 <= 0（配置为 0）会让循环空转、日志暴涨（#1117），退回 1 秒
+                interval = self._interval()
+                if await self._wait(interval if interval > 0 else 1):
                    return
        except asyncio.CancelledError:
            pass
--- a/backend/src/module/database/bangumi.py
+++ b/backend/src/module/database/bangumi.py
@@ -175,13 +175,18 @@ def match_bangumi_in_list(
            or persistence_target(release) is not PersistenceTarget.BANGUMI
        ):
            return None
+    # title_raw 由解析器重组（剔除中间的集数、拼接 " / " 分隔的别名），未必是
+    # 种子名的子串（#1103、#1114）；与本种子重组出的标题完全相等也算命中。
+    release_title = release and (
+        release.title_en or release.title_zh or release.title_jp
+    )
    best_match: Optional[Bangumi] = None
    best_rank = (-1, -1, -1)
    for bangumi in bangumi_list:
        if bangumi.deleted or not _release_matches_bangumi(release, bangumi):
            continue
        for pattern in _all_title_patterns(bangumi):
-            if pattern not in torrent_name:
+            if pattern not in torrent_name and pattern != release_title:
                continue
            group_match = int(
                release is not None
@@ -217,6 +222,16 @@ def _release_matches_bangumi(release: "ParsedRelease | None", bangumi: Bangumi)
    return release.season is None or bangumi.season == release.season


+def release_fits_bangumi(torrent_name: str, bangumi: Bangumi) -> bool:
+    """种子的季度/类型与番剧不冲突。整订阅下载（subscribe/collect）不做标题
+    匹配，搜索类 RSS 会混入其他季，需用此过滤（#1105）。"""
+    from module.parser.analyser.selector import parse_configured_release_title
+
+    return _release_matches_bangumi(
+        parse_configured_release_title(torrent_name), bangumi
+    )
+
+
 class BangumiDatabase:
    def __init__(self, session: AsyncSession):
        self.session = session
--- a/backend/src/module/manager/collector.py
+++ b/backend/src/module/manager/collector.py
@@ -1,6 +1,7 @@
 import logging

 from module.database import Database
+from module.database.bangumi import release_fits_bangumi
 from module.downloader import AddResult, DownloadClient
 from module.models import Bangumi, ResponseModel
 from module.network import RequestContent
@@ -53,6 +54,7 @@ class SeasonCollector:
                torrents = await req.get_torrents(
                    link, bangumi.filter.replace(",", "|")
                )
+            torrents = [t for t in torrents if release_fits_bangumi(t.name, bangumi)]
        async with Database() as db:
            # bangumi 必须先落库拿到 id：add_torrent 用它打 ab:<id> 标签，
            # 种子行也要用它关联 bangumi_id——否则种子会被记成孤儿，
--- a/backend/src/module/network/request_url.py
+++ b/backend/src/module/network/request_url.py
@@ -88,6 +88,17 @@ async def get_shared_client() -> httpx.AsyncClient:
    return _shared_client


+async def _replace_stale_client(stale: httpx.AsyncClient) -> httpx.AsyncClient:
+    """#1104: httpcore 在 HTTP 代理 CONNECT 成功、TLS 握手失败时泄漏隧道连接，
+    连接池被占满后永久 PoolTimeout。换一个新 client 给后续请求；旧的不 aclose
+    （其他协程可能仍在使用，见 dc786a29），交给 GC 回收。"""
+    global _shared_client
+    # ponytail: 旧连接要等 GC 才释放；httpcore 修复 encode/httpcore#1087 发布后升级依赖并删除此函数
+    if _shared_client is stale:
+        _shared_client = None
+    return await get_shared_client()
+
+
 async def reset_shared_client():
    """关闭并清除共享客户端，下次请求时自动创建新连接池。"""
    global _shared_client, _shared_client_proxy_key
@@ -156,6 +167,8 @@ class RequestURL:
                logger.warning(
                    f"Request error for {url}: {type(e).__name__}. Retry {try_time + 1}/{retry}"
                )
+                if isinstance(e, httpx.PoolTimeout):
+                    self._client = await _replace_stale_client(self._client)
                try_time += 1
                if try_time >= retry:
                    break
@@ -191,8 +204,10 @@ class RequestURL:
                    e.response.text[:200],
                )
                break
-            except httpx.RequestError:
+            except httpx.RequestError as e:
                logger.warning(f"Cannot connect to {url}. Wait for 5 seconds.")
+                if isinstance(e, httpx.PoolTimeout):
+                    self._client = await _replace_stale_client(self._client)
                try_time += 1
                if try_time >= retry:
                    break
--- a/backend/src/module/parser/analyser/tokenizer/classic.py
+++ b/backend/src/module/parser/analyser/tokenizer/classic.py
@@ -43,6 +43,8 @@ _SEASON_EPISODE_WORDS = re.compile(
 _SEASON = re.compile(r"(?<!\w)(S\d{1,2}|Season\s+\d{1,2})(?!\w)", re.I)
 _ORDINAL_SEASON = re.compile(r"\b(\d{1,2})(?:st|nd|rd|th)\s+Season\b", re.I)
 _CHINESE_SEASON = re.compile(r"第([零〇一二两三四五六七八九十百\d]+)[季期]")
+# Unicode 罗马数字 Ⅰ-Ⅻ（U+2160-216B）作季度标记，如 "Clevatess Ⅱ"（#1108）
+_ROMAN_SEASON = re.compile(r"(?<![A-Za-z])[Ⅰ-Ⅻ](?![A-Za-z])")
 _EXPLICIT_EPISODE = re.compile(
    r"(?<!\w)(?:Episode|EP?\.?|#)\s*[-_. ]?\s*" r"(\d{1,4}(?:\.\d+)?)(?:v(\d+))?(?!\w)",
    re.I,
@@ -536,6 +538,12 @@ def _extract_numbers(working: list[_WorkingSegment], state: _State) -> None:
            state.season_raw = match.group(0)
            state.evidence.append("season")

+        for match in _consume_all(work, _ROMAN_SEASON):
+            work.consume(match)
+            state.season = ord(match.group()) - 0x215F
+            state.season_raw = match.group()
+            state.evidence.append("season")
+
        for match in _consume_all(work, _ORDINAL_SEASON):
            if state.season is None:
                work.consume(match)
@@ -876,6 +884,8 @@ def _clean_fragment(text: str) -> str:

 def _clean_title(text: str) -> str:
    text = _clean_fragment(text)
+    # 全角破折号包裹整段标题："－魔獸之王…－" → "魔獸之王…"（#1108）
+    text = re.sub(r"^－(.+)－$", r"\1", text).strip()
    return re.sub(r"_+", " ", text).strip()


--- a/backend/src/module/rss/engine.py
+++ b/backend/src/module/rss/engine.py
@@ -8,7 +8,11 @@ from urllib.parse import urlparse

 from module.conf import settings
 from module.database import Database
-from module.database.bangumi import _groups_are_similar, match_bangumi_in_list
+from module.database.bangumi import (
+    _groups_are_similar,
+    match_bangumi_in_list,
+    release_fits_bangumi,
+)
 from module.downloader import AddResult, DownloadClient
 from module.models import Bangumi, Movie, ResponseModel, RSSItem, Torrent
 from module.network import RequestContent
@@ -462,6 +466,7 @@ class RSSEngine:
            torrents = await req.get_torrents(
                bangumi.rss_link, bangumi.filter.replace(",", "|")
            )
+            torrents = [t for t in torrents if release_fits_bangumi(t.name, bangumi)]
            if torrents:
                async with DownloadClient() as client:
                    result = await client.add_torrent(torrents, bangumi)
--- a/backend/src/test/test_database.py
+++ b/backend/src/test/test_database.py
@@ -579,6 +579,37 @@ def test_numeric_mixed_collection_does_not_fall_through_to_title_matching(
    assert match_bangumi_in_list(raw, [bangumi]) is None


+@pytest.mark.parametrize(
+    ("title_raw", "raw"),
+    (
+        # #1103: 集数从标题中间被剔除，title_raw 不再是种子名的子串
+        (
+            "攻壳机动队「攻壳机动队 THE GHOST IN THE SHELL」 日英双语-多国字幕",
+            "攻壳机动队「攻壳机动队 THE GHOST IN THE SHELL」 S01E03 1080p 日英双语-多国字幕",
+        ),
+        # #1114: " / " 分隔的多个英文名被拼接成一个 title_raw
+        (
+            "20 Seiki Denki Mokuroku Nijusseiki Denki Mokuroku",
+            "[喵萌奶茶屋&LoliHouse] 二十世纪电气目录 / 20 Seiki Denki Mokuroku / "
+            "Nijusseiki Denki Mokuroku - 07 [WebRip 1080p HEVC-10bit AAC][简繁日内封字幕]",
+        ),
+    ),
+)
+def test_match_bangumi_in_list_reconstructed_title_raw_matches(
+    title_raw: str, raw: str, monkeypatch
+) -> None:
+    monkeypatch.setattr(settings.rss_parser, "engine", "classic")
+    bangumi = Bangumi(
+        official_title=title_raw,
+        title_raw=title_raw,
+        rss_link="rss",
+        season=1,
+        episode_type="episode",
+    )
+
+    assert match_bangumi_in_list(raw, [bangumi]) is bangumi
+
+
 def test_titleless_mixed_collection_does_not_bypass_typed_matching(monkeypatch) -> None:
    monkeypatch.setattr(settings.rss_parser, "engine", "tokenizer")
    bangumi = Bangumi(
--- a/backend/src/test/test_healthcheck.py
+++ b/backend/src/test/test_healthcheck.py
@@ -0,0 +1,17 @@
+"""Docker HEALTHCHECK 端口解析（#1106）。"""
+
+import json
+from pathlib import Path
+
+import healthcheck
+
+
+def test_webui_port_reads_configured_port(tmp_path: Path) -> None:
+    config = tmp_path / "config.json"
+    config.write_text(json.dumps({"program": {"webui_port": 8080}}))
+
+    assert healthcheck.webui_port(config) == 8080
+
+
+def test_webui_port_defaults_when_config_missing(tmp_path: Path) -> None:
+    assert healthcheck.webui_port(tmp_path / "config.json") == 7892
--- a/backend/src/test/test_raw_parser.py
+++ b/backend/src/test/test_raw_parser.py
@@ -514,3 +514,26 @@ class TestIssue1025NoGroupPrefix:
        # Before the fix all three title fields were None and title_parser
        # raised "Cannot extract title_raw". At least one must now be set.
        assert any([info.title_en, info.title_zh, info.title_jp])
+
+
+class TestIssue1108RomanNumeralSeason:
+    """Issue #1108: Unicode 罗马数字（Ⅱ）标记季度，全角破折号包裹中文标题。"""
+
+    CONTENT = (
+        "[ANi] Clevatess Ⅱ－魔獸之王與虛假的勇者傳承－ - 05 "
+        "[1080P][Baha][WEB-DL][AAC AVC][CHT].mp4"
+    )
+
+    def _parse(self):
+        info = raw_parser(self.CONTENT)
+        assert info is not None
+        return info
+
+    def test_roman_numeral_is_season(self):
+        assert self._parse().season == 2
+
+    def test_zh_title_strips_numeral_and_fullwidth_dashes(self):
+        assert self._parse().title_zh == "魔獸之王與虛假的勇者傳承"
+
+    def test_en_title_kept(self):
+        assert self._parse().title_en == "Clevatess"
--- a/backend/src/test/test_request_url.py
+++ b/backend/src/test/test_request_url.py
@@ -117,6 +117,42 @@ class TestRetryWithReset:
        assert result.status_code == 200


+class TestPoolTimeoutRecovery:
+    """#1104: httpcore 在 HTTP 代理 CONNECT 成功、TLS 握手失败时泄漏隧道连接，
+    共享连接池被占满后永久 PoolTimeout。遇到 PoolTimeout 时应换新 client。"""
+
+    async def _get_with_poisoned_pool(self):
+        import httpx
+
+        import module.network.request_url as mod
+
+        old, new = AsyncMock(), AsyncMock()
+        old.get.side_effect = httpx.PoolTimeout("pool full")
+        resp = MagicMock(status_code=200, raise_for_status=MagicMock())
+        new.get.return_value = resp
+        mod._shared_client = old
+        mod._shared_client_proxy_key = mod._proxy_config_key()
+        try:
+            with (
+                patch("module.network.request_url.asyncio.sleep", new=AsyncMock()),
+                patch("module.network.request_url.httpx.AsyncClient", return_value=new),
+            ):
+                async with RequestURL() as req:
+                    result = await req.get_url("https://example.com/x", retry=2)
+        finally:
+            mod._shared_client = None
+            mod._shared_client_proxy_key = None
+        return old, resp, result
+
+    async def test_get_url_pool_timeout_retries_on_fresh_client(self):
+        _, resp, result = await self._get_with_poisoned_pool()
+        assert result is resp
+
+    async def test_get_url_pool_timeout_does_not_close_old_client(self):
+        old, _, _ = await self._get_with_poisoned_pool()
+        old.aclose.assert_not_called()
+
+
 class TestReentrantContextManager:
    """Regression for the notification-dispatch concurrency restore.

--- a/backend/src/test/test_rss_engine_new.py
+++ b/backend/src/test/test_rss_engine_new.py
@@ -683,6 +683,36 @@ class TestDownloadBangumi:
        assert resp.status_code == 502
        assert await rss_engine.db.torrent.search_all() == []

+    async def test_skips_torrents_of_another_season(self, rss_engine):
+        """#1105: 搜索 RSS 同时返回第一季和第二季，第二季不能按第一季下载。"""
+        bangumi = make_bangumi(
+            official_title="无职转生", rss_link="https://example.com/rss", filter=""
+        )
+        s1 = Torrent(
+            name="[沸班亚马制作组] 无职转生～到了异世界就拿出真本事～ - 01 [2160p]",
+            url="https://example.com/s1e01.torrent",
+        )
+        s2 = Torrent(
+            name="[沸班亚马制作组] 无职转生 第二季 ～到了异世界就拿出真本事～ - 01 [2160p]",
+            url="https://example.com/s2e01.torrent",
+        )
+        with (
+            patch("module.rss.engine.RequestContent") as MockReq,
+            patch("module.rss.engine.DownloadClient") as MockClient,
+        ):
+            req = AsyncMock()
+            req.get_torrents = AsyncMock(return_value=[s1, s2])
+            MockReq.return_value.__aenter__ = AsyncMock(return_value=req)
+            MockReq.return_value.__aexit__ = AsyncMock(return_value=False)
+            client = AsyncMock()
+            client.add_torrent = AsyncMock(return_value=AddResult.ADDED)
+            MockClient.return_value.__aenter__ = AsyncMock(return_value=client)
+            MockClient.return_value.__aexit__ = AsyncMock(return_value=False)
+
+            await rss_engine.download_bangumi(bangumi)
+
+        client.add_torrent.assert_awaited_once_with([s1], bangumi)
+

 # ---------------------------------------------------------------------------
 # add_rss
--- a/backend/src/test/test_scheduler.py
+++ b/backend/src/test/test_scheduler.py
@@ -27,6 +27,20 @@ class TestPeriodicTask:
        assert task.running is False
        assert counter["n"] >= 2

+    async def test_nonpositive_interval_does_not_spin(self):
+        """#1117: rss_time/rename_time 配成 0 时循环空转、日志暴涨；应退回 1 秒间隔。"""
+        counter = {"n": 0}
+
+        async def body():
+            counter["n"] += 1
+
+        task = PeriodicTask("t", run=body, interval=lambda: 0)
+        task.start()
+        await asyncio.sleep(0.2)
+        await task.stop()
+
+        assert counter["n"] == 1
+
    async def test_stop_is_idempotent(self):
        """Calling stop() twice (or before start) does not raise."""
        task = PeriodicTask("t", run=lambda: asyncio.sleep(0), interval=lambda: 0.01)
--- a/docs/deploy/docker-cli.md
+++ b/docs/deploy/docker-cli.md
@@ -40,6 +40,8 @@ docker run -d \
  --network=bridge \
  --dns=8.8.8.8 \
  --restart unless-stopped \
+  --log-opt max-size=10m \
+  --log-opt max-file=3 \
  ghcr.io/estrellaxd/auto_bangumi:latest
 ```

--- a/docs/deploy/docker-compose.md
+++ b/docs/deploy/docker-compose.md
@@ -41,6 +41,11 @@ services:
    ports:
      - "7892:7892"
    restart: unless-stopped
+    logging:
+      driver: json-file
+      options:
+        max-size: "10m"
+        max-file: "3"
    dns:
      - 8.8.8.8
    network_mode: bridge
--- a/docs/deploy/quick-start.md
+++ b/docs/deploy/quick-start.md
@@ -42,6 +42,8 @@ docker run -d \
  --network=bridge \
  --dns=8.8.8.8 \
  --restart unless-stopped \
+  --log-opt max-size=10m \
+  --log-opt max-file=3 \
  ghcr.io/estrellaxd/auto_bangumi:latest
 ```

--- a/docs/en/deploy/docker-cli.md
+++ b/docs/en/deploy/docker-cli.md
@@ -40,6 +40,8 @@ docker run -d \
  --network=bridge \
  --dns=8.8.8.8 \
  --restart unless-stopped \
+  --log-opt max-size=10m \
+  --log-opt max-file=3 \
  ghcr.io/estrellaxd/auto_bangumi:latest
 ```

--- a/docs/en/deploy/docker-compose.md
+++ b/docs/en/deploy/docker-compose.md
@@ -41,6 +41,11 @@ services:
    ports:
      - "7892:7892"
    restart: unless-stopped
+    logging:
+      driver: json-file
+      options:
+        max-size: "10m"
+        max-file: "3"
    dns:
      - 8.8.8.8
    network_mode: bridge
--- a/docs/en/deploy/quick-start.md
+++ b/docs/en/deploy/quick-start.md
@@ -42,6 +42,8 @@ docker run -d \
  --network=bridge \
  --dns=8.8.8.8 \
  --restart unless-stopped \
+  --log-opt max-size=10m \
+  --log-opt max-file=3 \
  ghcr.io/estrellaxd/auto_bangumi:latest
 ```

--- a/docs/ja/deploy/docker-cli.md
+++ b/docs/ja/deploy/docker-cli.md
@@ -40,6 +40,8 @@ docker run -d \
  --network=bridge \
  --dns=8.8.8.8 \
  --restart unless-stopped \
+  --log-opt max-size=10m \
+  --log-opt max-file=3 \
  ghcr.io/estrellaxd/auto_bangumi:latest
 ```

--- a/docs/ja/deploy/docker-compose.md
+++ b/docs/ja/deploy/docker-compose.md
@@ -41,6 +41,11 @@ services:
    ports:
      - "7892:7892"
    restart: unless-stopped
+    logging:
+      driver: json-file
+      options:
+        max-size: "10m"
+        max-file: "3"
    dns:
      - 8.8.8.8
    network_mode: bridge
--- a/docs/ja/deploy/quick-start.md
+++ b/docs/ja/deploy/quick-start.md
@@ -42,6 +42,8 @@ docker run -d \
  --network=bridge \
  --dns=8.8.8.8 \
  --restart unless-stopped \
+  --log-opt max-size=10m \
+  --log-opt max-file=3 \
  ghcr.io/estrellaxd/auto_bangumi:latest
 ```

--- a/docs/resource/docker-compose/AutoBangumi/docker-compose.yml
+++ b/docs/resource/docker-compose/AutoBangumi/docker-compose.yml
@@ -11,6 +11,11 @@ services:
    # 应用内“一键更新”依赖此重启策略：更新后容器会自行退出，由 Docker 重启
    # 以应用更新，因此必须保留 restart: unless-stopped。
    restart: unless-stopped
+    logging:
+      driver: json-file
+      options:
+        max-size: "10m"
+        max-file: "3"
    dns:
      - 223.5.5.5
    environment:
--- a/docs/resource/docker-compose/qBittorrent+AutoBangumi/docker-compose.yml
+++ b/docs/resource/docker-compose/qBittorrent+AutoBangumi/docker-compose.yml
@@ -28,6 +28,11 @@ services:
    # 应用内“一键更新”依赖此重启策略：更新后容器会自行退出，由 Docker 重启
    # 以应用更新，因此必须保留 restart: unless-stopped。
    restart: unless-stopped
+    logging:
+      driver: json-file
+      options:
+        max-size: "10m"
+        max-file: "3"
    environment:
      - TZ=Asia/Shanghai
      - PGID=$(id -g)