fix: 优化docker 镜像

crawler/config.py

@@ -39,11 +39,37 @@ RSS_FEEDS = [
     "https://www.aljazeera.com/xml/rss/middleeast.xml",
 ]
 
-# 关键词过滤：至少匹配一个才会入库
+# 关键词过滤：至少匹配一个才会入库（与地图区域对应：伊拉克/叙利亚/海湾/红海/地中海等）
 KEYWORDS = [
-    "iran", "iranian", "tehran", "以色列", "israel",
-    "usa", "us ", "american", "美军", "美国",
-    "middle east", "中东", "persian gulf", "波斯湾",
+    # 伊朗
+    "iran", "iranian", "tehran", "德黑兰", "bushehr", "布什尔", "abbas", "阿巴斯",
+    # 以色列 / 巴勒斯坦
+    "israel", "以色列", "hamas", "gaza", "加沙", "hezbollah", "真主党",
+    # 美国
+    "usa", "us ", "american", "美军", "美国", "pentagon",
+    # 区域（地图覆盖）
+    "middle east", "中东", "persian gulf", "波斯湾", "gulf of oman", "阿曼湾",
+    "arabian sea", "阿拉伯海", "red sea", "红海", "mediterranean", "地中海",
+    "strait of hormuz", "霍尔木兹",
+    # 伊拉克 / 叙利亚
+    "iraq", "伊拉克", "baghdad", "巴格达", "erbil", "埃尔比勒", "basra", "巴士拉",
+    "syria", "叙利亚", "damascus", "大马士革", "deir", "代尔祖尔",
+    # 海湾国家
+    "saudi", "沙特", "riyadh", "利雅得", "qatar", "卡塔尔", "doha", "多哈",
+    "uae", "emirates", "阿联酋", "dubai", "迪拜", "abu dhabi",
+    "bahrain", "巴林", "kuwait", "科威特", "oman", "阿曼", "yemen", "也门",
+    # 约旦 / 土耳其 / 埃及 / 吉布提 / 黎巴嫩
+    "jordan", "约旦", "amman", "安曼",
+    "lebanon", "黎巴嫩",
+    "turkey", "土耳其", "incirlik", "因吉尔利克",
+    "egypt", "埃及", "cairo", "开罗", "sinai", "西奈",
+    "djibouti", "吉布提",
+    # 军事 / 基地
+    "al-asad", "al asad", "阿萨德", "al udeid", "乌代德", "incirlik",
     "strike", "attack", "military", "missile", "核", "nuclear",
-    "carrier", "航母", "houthi", "胡塞", "hamas",
+    "carrier", "航母", "drone", "uav", "无人机", "retaliation", "报复",
+    "base", "基地", "troops", "troop", "soldier", "personnel",
+    # 胡塞 / 武装 / 军力
+    "houthi", "胡塞", "houthis",
+    "idf", "irgc", "革命卫队", "qassem soleimani", "苏莱曼尼",
 ]

crawler/db_merge.py

@@ -67,7 +67,7 @@ def merge(extracted: Dict[str, Any], db_path: Optional[str] = None) -> bool:
             )
             if conn.total_changes > 0:
                 updated = True
-        # combat_losses：增量叠加到当前值
+        # combat_losses：增量叠加到当前值，无行则先插入初始行
         if "combat_losses_delta" in extracted:
             for side, delta in extracted["combat_losses_delta"].items():
                 if side not in ("us", "iran"):
@@ -77,13 +77,14 @@ def merge(extracted: Dict[str, Any], db_path: Optional[str] = None) -> bool:
                         "SELECT personnel_killed,personnel_wounded,civilian_killed,civilian_wounded,bases_destroyed,bases_damaged,aircraft,warships,armor,vehicles FROM combat_losses WHERE side = ?",
                         (side,),
                     ).fetchone()
-                    if not row:
-                        continue
-                    cur = {
-                        "personnel_killed": row[0], "personnel_wounded": row[1], "civilian_killed": row[2] or 0,
-                        "civilian_wounded": row[3] or 0, "bases_destroyed": row[4], "bases_damaged": row[5],
-                        "aircraft": row[6], "warships": row[7], "armor": row[8], "vehicles": row[9],
-                    }
+                    cur = {"personnel_killed": 0, "personnel_wounded": 0, "civilian_killed": 0, "civilian_wounded": 0,
+                           "bases_destroyed": 0, "bases_damaged": 0, "aircraft": 0, "warships": 0, "armor": 0, "vehicles": 0}
+                    if row:
+                        cur = {
+                            "personnel_killed": row[0], "personnel_wounded": row[1], "civilian_killed": row[2] or 0,
+                            "civilian_wounded": row[3] or 0, "bases_destroyed": row[4], "bases_damaged": row[5],
+                            "aircraft": row[6], "warships": row[7], "armor": row[8], "vehicles": row[9],
+                        }
                     pk = max(0, (cur["personnel_killed"] or 0) + delta.get("personnel_killed", 0))
                     pw = max(0, (cur["personnel_wounded"] or 0) + delta.get("personnel_wounded", 0))
                     ck = max(0, (cur["civilian_killed"] or 0) + delta.get("civilian_killed", 0))
@@ -95,11 +96,18 @@ def merge(extracted: Dict[str, Any], db_path: Optional[str] = None) -> bool:
                     ar = max(0, (cur["armor"] or 0) + delta.get("armor", 0))
                     vh = max(0, (cur["vehicles"] or 0) + delta.get("vehicles", 0))
                     ts = datetime.utcnow().strftime("%Y-%m-%dT%H:%M:%S.000Z")
-                    conn.execute(
-                        """UPDATE combat_losses SET personnel_killed=?, personnel_wounded=?, civilian_killed=?, civilian_wounded=?,
-                        bases_destroyed=?, bases_damaged=?, aircraft=?, warships=?, armor=?, vehicles=?, updated_at=? WHERE side=?""",
-                        (pk, pw, ck, cw, bd, bm, ac, ws, ar, vh, ts, side),
-                    )
+                    if row:
+                        conn.execute(
+                            """UPDATE combat_losses SET personnel_killed=?, personnel_wounded=?, civilian_killed=?, civilian_wounded=?,
+                            bases_destroyed=?, bases_damaged=?, aircraft=?, warships=?, armor=?, vehicles=?, updated_at=? WHERE side=?""",
+                            (pk, pw, ck, cw, bd, bm, ac, ws, ar, vh, ts, side),
+                        )
+                    else:
+                        conn.execute(
+                            """INSERT OR REPLACE INTO combat_losses (side, personnel_killed, personnel_wounded, civilian_killed, civilian_wounded,
+                            bases_destroyed, bases_damaged, aircraft, warships, armor, vehicles, updated_at) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)""",
+                            (side, pk, pw, ck, cw, bd, bm, ac, ws, ar, vh, ts),
+                        )
                     if conn.total_changes > 0:
                         updated = True
                 except Exception:
@@ -115,6 +123,30 @@ def merge(extracted: Dict[str, Any], db_path: Optional[str] = None) -> bool:
             w = extracted["wall_street"]
             conn.execute("INSERT INTO wall_street_trend (time, value) VALUES (?, ?)", (w["time"], w["value"]))
             updated = True
+        # key_location：更新受袭基地 status/damage_level
+        if "key_location_updates" in extracted:
+            try:
+                for u in extracted["key_location_updates"]:
+                    kw = (u.get("name_keywords") or "").replace("|", " ").split()
+                    side = u.get("side")
+                    status = u.get("status", "attacked")[:20]
+                    dmg = u.get("damage_level", 2)
+                    if not kw or side not in ("us", "iran"):
+                        continue
+                    conditions = " OR ".join(
+                        "(LOWER(name) LIKE ? OR name LIKE ?)" for _ in kw
+                    )
+                    params = [status, dmg, side]
+                    for k in kw:
+                        params.extend([f"%{k}%", f"%{k}%"])
+                    cur = conn.execute(
+                        f"UPDATE key_location SET status=?, damage_level=? WHERE side=? AND ({conditions})",
+                        params,
+                    )
+                    if cur.rowcount > 0:
+                        updated = True
+            except Exception:
+                pass
         if updated:
             conn.execute("INSERT OR REPLACE INTO situation (id, data, updated_at) VALUES (1, '{}', ?)", (datetime.utcnow().strftime("%Y-%m-%dT%H:%M:%S.000Z"),))
         conn.commit()

crawler/extractor_ai.py

@@ -27,11 +27,16 @@ def _call_ollama_extract(text: str, timeout: int = 10) -> Optional[Dict[str, Any
 - summary: 1-2句中文事实，≤80字
 - category: deployment|alert|intel|diplomatic|other
 - severity: low|medium|high|critical
-- us_personnel_killed, iran_personnel_killed 等：仅当新闻明确提及具体数字时填写
+- 战损（仅当新闻明确提及数字时填写，格式 us_XXX / iran_XXX）:
+  us_personnel_killed, iran_personnel_killed, us_personnel_wounded, iran_personnel_wounded,
+  us_civilian_killed, iran_civilian_killed, us_civilian_wounded, iran_civilian_wounded,
+  us_bases_destroyed, iran_bases_destroyed, us_bases_damaged, iran_bases_damaged,
+  us_aircraft, iran_aircraft, us_warships, iran_warships, us_armor, iran_armor, us_vehicles, iran_vehicles
 - retaliation_sentiment: 0-100，仅当新闻涉及伊朗报复情绪时
 - wall_street_value: 0-100，仅当新闻涉及美股/市场反应时
+- key_location_updates: 当新闻提及具体基地/地点遭袭时，数组项 { "name_keywords": "asad|阿萨德|assad", "side": "us", "status": "attacked", "damage_level": 1-3 }
 
-原文：{str(text)[:500]}
+原文：{str(text)[:800]}
 
 直接输出 JSON，不要解释："""
         r = requests.post(
@@ -97,4 +102,17 @@ def extract_from_news(text: str, timestamp: Optional[str] = None) -> Dict[str, A
         v = parsed["wall_street_value"]
         if isinstance(v, (int, float)) and 0 <= v <= 100:
             out["wall_street"] = {"time": ts, "value": int(v)}
+    # key_location_updates：受袭基地
+    if "key_location_updates" in parsed and isinstance(parsed["key_location_updates"], list):
+        valid = []
+        for u in parsed["key_location_updates"]:
+            if isinstance(u, dict) and u.get("name_keywords") and u.get("side") in ("us", "iran"):
+                valid.append({
+                    "name_keywords": str(u["name_keywords"]),
+                    "side": u["side"],
+                    "status": str(u.get("status", "attacked"))[:20],
+                    "damage_level": min(3, max(1, int(u["damage_level"]))) if isinstance(u.get("damage_level"), (int, float)) else 2,
+                })
+        if valid:
+            out["key_location_updates"] = valid
     return out

crawler/extractor_rules.py

@@ -24,18 +24,64 @@ def extract_from_news(text: str, timestamp: Optional[str] = None) -> Dict[str, A
     t = (text or "").lower()
 
     loss_us, loss_ir = {}, {}
-    v = _first_int(t, r"(?:us|american|u\.?s\.?)[\s\w]*(?:say|report)[\s\w]*(\d+)[\s\w]*(?:troop|soldier|killed|dead)")
+
+    # 美军人员伤亡
+    v = _first_int(t, r"(?:us|american|u\.?s\.?)[\s\w]*(?:say|report)[\s\w]*(\d+)[\s\w]*(?:troop|soldier|military)[\s\w]*(?:killed|dead)")
     if v is not None:
         loss_us["personnel_killed"] = v
-    v = _first_int(t, r"(\d+)[\s\w]*(?:us|american)[\s\w]*(?:troop|soldier|killed|dead)")
+    v = _first_int(t, r"(\d+)[\s\w]*(?:us|american)[\s\w]*(?:troop|soldier|military)[\s\w]*(?:killed|dead)")
     if v is not None:
         loss_us["personnel_killed"] = v
-    v = _first_int(t, r"(?:iran|iranian)[\s\w]*(?:say|report)[\s\w]*(\d+)[\s\w]*(?:troop|soldier|killed|dead)")
+    v = _first_int(t, r"(?:us|american)[\s\w]*(\d+)[\s\w]*(?:wounded|injured)")
+    if v is not None:
+        loss_us["personnel_wounded"] = v
+
+    # 伊朗人员伤亡
+    v = _first_int(t, r"(?:iran|iranian)[\s\w]*(?:say|report)[\s\w]*(\d+)[\s\w]*(?:troop|soldier|guard|killed|dead)")
     if v is not None:
         loss_ir["personnel_killed"] = v
-    v = _first_int(t, r"(\d+)[\s\w]*(?:iranian|iran)[\s\w]*(?:troop|soldier|killed|dead)")
+    v = _first_int(t, r"(\d+)[\s\w]*(?:iranian|iran)[\s\w]*(?:troop|soldier|guard|killed|dead)")
     if v is not None:
         loss_ir["personnel_killed"] = v
+    v = _first_int(t, r"(?:iran|iranian)[\s\w]*(\d+)[\s\w]*(?:wounded|injured)")
+    if v is not None:
+        loss_ir["personnel_wounded"] = v
+
+    # 平民伤亡（多不区分阵营，计入双方或仅 us 因多为美国基地周边）
+    v = _first_int(t, r"(\d+)[\s\w]*(?:civilian|civil)[\s\w]*(?:killed|dead)")
+    if v is not None:
+        loss_us["civilian_killed"] = v
+    v = _first_int(t, r"(\d+)[\s\w]*(?:civilian|civil)[\s\w]*(?:wounded|injured)")
+    if v is not None:
+        loss_us["civilian_wounded"] = v
+
+    # 基地损毁（美方基地居多）
+    v = _first_int(t, r"(\d+)[\s\w]*(?:base)[\s\w]*(?:destroyed|leveled)")
+    if v is not None:
+        loss_us["bases_destroyed"] = v
+    v = _first_int(t, r"(\d+)[\s\w]*(?:base)[\s\w]*(?:damaged|hit|struck)")
+    if v is not None:
+        loss_us["bases_damaged"] = v
+    if "base" in t and ("destroy" in t or "level" in t) and not loss_us.get("bases_destroyed"):
+        loss_us["bases_destroyed"] = 1
+    if "base" in t and ("damage" in t or "hit" in t or "struck" in t or "strike" in t) and not loss_us.get("bases_damaged"):
+        loss_us["bases_damaged"] = 1
+
+    # 战机 / 舰船（根据上下文判断阵营）
+    v = _first_int(t, r"(\d+)[\s\w]*(?:aircraft|plane|jet|fighter|f-?16|f-?35|f-?18)[\s\w]*(?:down|destroyed|lost|shot)")
+    if v is not None:
+        if "us" in t or "american" in t or "u.s" in t:
+            loss_us["aircraft"] = v
+        elif "iran" in t:
+            loss_ir["aircraft"] = v
+        else:
+            loss_us["aircraft"] = v
+    v = _first_int(t, r"(\d+)[\s\w]*(?:ship|destroyer|warship|vessel)[\s\w]*(?:hit|damaged|sunk)")
+    if v is not None:
+        if "iran" in t:
+            loss_ir["warships"] = v
+        else:
+            loss_us["warships"] = v
 
     if loss_us:
         out.setdefault("combat_losses_delta", {})["us"] = loss_us

crawler/realtime_conflict_service.py

@@ -14,13 +14,13 @@ from datetime import datetime
 from pathlib import Path
 from typing import List, Optional
 
+import asyncio
 import logging
 import requests
 from fastapi import FastAPI
 from fastapi.middleware.cors import CORSMiddleware
-from apscheduler.schedulers.background import BackgroundScheduler
 
-logging.getLogger("apscheduler.scheduler").setLevel(logging.ERROR)
+logging.getLogger("uvicorn").setLevel(logging.INFO)
 app = FastAPI(title="GDELT Conflict Service")
 app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"])
 
@@ -263,8 +263,9 @@ def _extract_and_merge_panel_data(items: list) -> None:
             from extractor_ai import extract_from_news
         from datetime import timezone
         merged_any = False
-        # 只对前几条有足够文本的新闻做提取，避免 Ollama 调用过多
-        for it in items[:5]:
+        # 规则模式可多处理几条（无 Ollama）；AI 模式限制 5 条避免调用过多
+        limit = 10 if os.environ.get("CLEANER_AI_DISABLED", "0") == "1" else 5
+        for it in items[:limit]:
             text = (it.get("title", "") or "") + " " + (it.get("summary", "") or "")
             if len(text.strip()) < 20:
                 continue
@@ -292,12 +293,22 @@ def _extract_and_merge_panel_data(items: list) -> None:
 
 
 # ==========================
-# 定时任务（RSS 更频繁，优先保证事件脉络实时）
+# 定时任务（asyncio 后台任务，避免 APScheduler executor 关闭竞态）
 # ==========================
-scheduler = BackgroundScheduler()
-scheduler.add_job(fetch_news, "interval", seconds=RSS_INTERVAL_SEC, max_instances=2, coalesce=True)
-scheduler.add_job(fetch_gdelt_events, "interval", seconds=FETCH_INTERVAL_SEC, max_instances=2, coalesce=True)
-scheduler.start()
+_bg_task: Optional[asyncio.Task] = None
+
+
+async def _periodic_fetch() -> None:
+    loop = asyncio.get_event_loop()
+    while True:
+        try:
+            await loop.run_in_executor(None, fetch_news)
+            await loop.run_in_executor(None, fetch_gdelt_events)
+        except asyncio.CancelledError:
+            break
+        except Exception as e:
+            print(f"  [warn] 定时抓取: {e}")
+        await asyncio.sleep(min(RSS_INTERVAL_SEC, FETCH_INTERVAL_SEC))
 
 
 # ==========================
@@ -356,18 +367,23 @@ def _get_conflict_stats() -> dict:
 
 
 @app.on_event("startup")
-def startup():
-    # 新闻优先启动，确保事件脉络有数据
-    fetch_news()
-    fetch_gdelt_events()
+async def startup():
+    global _bg_task
+    loop = asyncio.get_event_loop()
+    await loop.run_in_executor(None, fetch_news)
+    await loop.run_in_executor(None, fetch_gdelt_events)
+    _bg_task = asyncio.create_task(_periodic_fetch())
 
 
 @app.on_event("shutdown")
-def shutdown():
-    try:
-        scheduler.shutdown(wait=False)
-    except Exception:
-        pass
+async def shutdown():
+    global _bg_task
+    if _bg_task and not _bg_task.done():
+        _bg_task.cancel()
+        try:
+            await _bg_task
+        except asyncio.CancelledError:
+            pass
 
 
 if __name__ == "__main__":

crawler/requirements.txt

@@ -2,5 +2,4 @@ requests>=2.31.0
 feedparser>=6.0.0
 fastapi>=0.109.0
 uvicorn>=0.27.0
-apscheduler>=3.10.0
 deep-translator>=1.11.0