修改模型

.env

@@ -6,6 +6,8 @@ PG_PORT=5432
 PG_USER=postgres
 PG_PASSWORD=feAeyR0u2fJGSS5ooFdHnSbyHQNY4WlV
 PG_DB=postgres
+PG_POOL_SIZE=10
+PG_MAX_OVERFLOW=10
 
 # MySQL（业务库）
 DB_HOST=8.163.14.142
@@ -13,6 +15,11 @@ DB_PORT=30006
 DB_USER=root
 DB_PASSWORD=^CgDatabase2020
 DB_NAME=offerpie
+MYSQL_POOL_SIZE=10
+MYSQL_MAX_OVERFLOW=10
+
+# 连接池获取连接等待超时（秒），两库共用
+DB_POOL_TIMEOUT=30
 
 
 # AI 供应商
@@ -23,19 +30,32 @@ VOLCENGINE_BASE_URL=https://ark.cn-beijing.volces.com/api/v3
 ANTHROPIC_API_KEY=sk-43ccdb29caa7e9ebe0db8ac0958c63f6d3a2d62e59064d3d26d94332055a9bc9
 ANTHROPIC_BASE_URL=https://code.warpdevloper.cloud
 
+# Claude 备用中转（Anthropic 风格）
+ANTHROPIC_ALT_API_KEY=sk-ac5544257dcf843de887fa153c4f4a11fa5007cbae4b8f4eec895a354c89ff7b
+ANTHROPIC_ALT_BASE_URL=http://98.142.250.143:8080
+
+# OpenAI（OpenAI 兼容风格）
+OPENAI_API_KEY=sk-12663c3a1ae5db0bc9aee78ffa64e7a0b95bcf7a15541040dd2bce573130f5c8
+OPENAI_BASE_URL=https://re.94xy.cn/v1
+
 # 岗位清洗参数
-CLEAN_BATCH_SIZE=20
-CLEAN_CONCURRENCY=20
-CLEAN_INTERVAL_SECONDS=100
+CLEAN_WORKER_COUNT=20
+CLEAN_IDLE_SLEEP=5.0
 CLEAN_TOTAL_LIMIT=0
 
 # 公司补充参数
-COMPANY_BATCH_SIZE=20
-COMPANY_CONCURRENCY=10
-COMPANY_INTERVAL_SECONDS=300
+COMPANY_WORKER_COUNT=10
+COMPANY_IDLE_SLEEP=10.0
 
 # 岗位下架
 JOB_EXPIRE_DAYS=7
 
 # 日志
 LOGGING_LEVEL=INFO
+
+# 阿里云 OSS
+OSS_ACCESS_KEY_ID=LTAI5tEdLKKQUKhTyUpfH5Mk
+OSS_ACCESS_KEY_SECRET=RjUdTrq0V5qA4b3BUElNhXqs3ZLp5k
+OSS_ENDPOINT=oss-cn-guangzhou.aliyuncs.com
+OSS_BUCKET=offerpie
+OSS_DOMAIN=https://offerpie.oss-cn-guangzhou.aliyuncs.com

.env.prod

@@ -6,6 +6,8 @@ PG_PORT=5432
 PG_USER=postgres
 PG_PASSWORD=feAeyR0u2fJGSS5ooFdHnSbyHQNY4WlV
 PG_DB=postgres
+PG_POOL_SIZE=10
+PG_MAX_OVERFLOW=10
 
 # MySQL（业务库）
 DB_HOST=8.163.14.142
@@ -13,23 +15,47 @@ DB_PORT=30006
 DB_USER=root
 DB_PASSWORD=^CgDatabase2020
 DB_NAME=offerpie
+MYSQL_POOL_SIZE=10
+MYSQL_MAX_OVERFLOW=10
+
+# 连接池获取连接等待超时（秒），两库共用
+DB_POOL_TIMEOUT=30
+
 
 # AI 供应商
 VOLCENGINE_API_KEY=fd065993-bee2-4f31-8bf2-56d5d3012c02
 VOLCENGINE_BASE_URL=https://ark.cn-beijing.volces.com/api/v3
 
+# Claude（Anthropic 风格）
+ANTHROPIC_API_KEY=sk-43ccdb29caa7e9ebe0db8ac0958c63f6d3a2d62e59064d3d26d94332055a9bc9
+ANTHROPIC_BASE_URL=https://code.warpdevloper.cloud
+
+# Claude 备用中转（Anthropic 风格）
+ANTHROPIC_ALT_API_KEY=sk-ac5544257dcf843de887fa153c4f4a11fa5007cbae4b8f4eec895a354c89ff7b
+ANTHROPIC_ALT_BASE_URL=http://98.142.250.143:8080
+
+# OpenAI（OpenAI 兼容风格）
+OPENAI_API_KEY=sk-12663c3a1ae5db0bc9aee78ffa64e7a0b95bcf7a15541040dd2bce573130f5c8
+OPENAI_BASE_URL=https://re.94xy.cn/v1
+
 # 岗位清洗参数
-CLEAN_BATCH_SIZE=100
-CLEAN_CONCURRENCY=80
-CLEAN_INTERVAL_SECONDS=200
+CLEAN_WORKER_COUNT=20
+CLEAN_IDLE_SLEEP=5.0
+CLEAN_TOTAL_LIMIT=0
 
 # 公司补充参数
-COMPANY_BATCH_SIZE=20
-COMPANY_CONCURRENCY=10
-COMPANY_INTERVAL_SECONDS=300
+COMPANY_WORKER_COUNT=10
+COMPANY_IDLE_SLEEP=10.0
 
 # 岗位下架
 JOB_EXPIRE_DAYS=7
 
 # 日志
 LOGGING_LEVEL=INFO
+
+# 阿里云 OSS
+OSS_ACCESS_KEY_ID=LTAI5tEdLKKQUKhTyUpfH5Mk
+OSS_ACCESS_KEY_SECRET=RjUdTrq0V5qA4b3BUElNhXqs3ZLp5k
+OSS_ENDPOINT=oss-cn-guangzhou.aliyuncs.com
+OSS_BUCKET=offerpie
+OSS_DOMAIN=https://offerpie.oss-cn-guangzhou.aliyuncs.com

.env.test

@@ -6,6 +6,8 @@ PG_PORT=5432
 PG_USER=postgres
 PG_PASSWORD=feAeyR0u2fJGSS5ooFdHnSbyHQNY4WlV
 PG_DB=postgres
+PG_POOL_SIZE=10
+PG_MAX_OVERFLOW=10
 
 # MySQL（业务库）
 DB_HOST=192.168.31.105
@@ -13,24 +15,42 @@ DB_PORT=3306
 DB_USER=root
 DB_PASSWORD=123456
 DB_NAME=offerpie
+MYSQL_POOL_SIZE=10
+MYSQL_MAX_OVERFLOW=10
+
+# 连接池获取连接等待超时（秒），两库共用
+DB_POOL_TIMEOUT=30
 
 
 # AI 供应商
 VOLCENGINE_API_KEY=fd065993-bee2-4f31-8bf2-56d5d3012c02
 VOLCENGINE_BASE_URL=https://ark.cn-beijing.volces.com/api/v3
 
+# Claude 备用中转（Anthropic 风格）
+ANTHROPIC_ALT_API_KEY=sk-ac5544257dcf843de887fa153c4f4a11fa5007cbae4b8f4eec895a354c89ff7b
+ANTHROPIC_ALT_BASE_URL=http://98.142.250.143:8080
+
+# OpenAI（OpenAI 兼容风格）
+OPENAI_API_KEY=sk-12663c3a1ae5db0bc9aee78ffa64e7a0b95bcf7a15541040dd2bce573130f5c8
+OPENAI_BASE_URL=https://re.94xy.cn/v1
+
 # 岗位清洗参数
-CLEAN_BATCH_SIZE=100
-CLEAN_CONCURRENCY=50
-CLEAN_INTERVAL_SECONDS=180
+CLEAN_WORKER_COUNT=20
+CLEAN_IDLE_SLEEP=5.0
 
 # 公司补充参数
-COMPANY_BATCH_SIZE=20
-COMPANY_CONCURRENCY=10
-COMPANY_INTERVAL_SECONDS=300
+COMPANY_WORKER_COUNT=10
+COMPANY_IDLE_SLEEP=10.0
 
 # 岗位下架
 JOB_EXPIRE_DAYS=7
 
 # 日志
 LOGGING_LEVEL=INFO
+
+# 阿里云 OSS
+OSS_ACCESS_KEY_ID=LTAI5tEdLKKQUKhTyUpfH5Mk
+OSS_ACCESS_KEY_SECRET=RjUdTrq0V5qA4b3BUElNhXqs3ZLp5k
+OSS_ENDPOINT=oss-cn-guangzhou.aliyuncs.com
+OSS_BUCKET=offerpie
+OSS_DOMAIN=https://offerpie.oss-cn-guangzhou.aliyuncs.com

app/ai/model_config.py

@@ -9,14 +9,14 @@ from app.ai.models import LLM
 class JobCleanModel:
     """岗位清洗模块"""
     # 第一次AI：结构化提取岗位信息
-    STRUCTURE = LLM.CLAUDE_OPUS.create(temperature=0)
+    STRUCTURE = LLM.CLAUDE_SONNET_4_5.create(temperature=0)
     # 第二次AI：专业匹配
-    MAJOR_MATCH = LLM.CLAUDE_OPUS.create(temperature=0)
+    MAJOR_MATCH = LLM.CLAUDE_SONNET_4_5.create(temperature=0)
     # 第三次AI：技能提取
-    SKILL_EXTRACT = LLM.CLAUDE_OPUS.create(temperature=0)
+    SKILL_EXTRACT = LLM.CLAUDE_SONNET_4_5.create(temperature=0)
 
 
 class CompanyCleanModel:
     """公司补充模块"""
     # 公司信息补充
-    ENRICH = LLM.CLAUDE_OPUS.create(temperature=0)
+    ENRICH = LLM.CLAUDE_SONNET_4_5.create(temperature=0)

app/ai/models.py

@@ -18,6 +18,8 @@ from app.config import settings
 # 供应商连接配置 = (api_key函数, base_url函数)
 _VOLCENGINE = (lambda: settings.volcengine_api_key, lambda: settings.volcengine_base_url)
 _ANTHROPIC = (lambda: settings.anthropic_api_key, lambda: settings.anthropic_base_url)
+_ANTHROPIC_ALT = (lambda: settings.anthropic_alt_api_key, lambda: settings.anthropic_alt_base_url)
+_OPENAI = (lambda: settings.openai_api_key, lambda: settings.openai_base_url)
 
 
 class LLM(Enum):
@@ -32,6 +34,10 @@ class LLM(Enum):
 
     # Claude（Anthropic 风格）
     CLAUDE_OPUS = ("claude-opus-4-6", ChatAnthropic, *_ANTHROPIC)
+    CLAUDE_SONNET_4_5 = ("claude-sonnet-4-5-20250929", ChatAnthropic, *_ANTHROPIC_ALT)
+
+    # OpenAI（OpenAI 兼容）
+    GPT_5_4_MINI = ("gpt-5.4-mini", ChatOpenAI, *_OPENAI)
 
     def __init__(self, model_name: str, cls, api_key_fn, base_url_fn):
         self.model_name = model_name

app/ai/prompts.py

@@ -50,18 +50,34 @@ MAJOR_MATCH_SYSTEM = """你是一个岗位专业匹配助手。根据岗位信
 
 
 # ──────────── 第三次AI：技能提取 ────────────
-SKILL_EXTRACT_SYSTEM = """你是一个技能提取助手。根据岗位信息，提取该岗位要求的核心专业能力和工具技能。
+# 说明：与 Java 简历侧（UserProfileAnalyzeService.extractSkillTags）配对使用。
+# 策略「岗位精、简历广」：岗位侧只提该岗位最核心、最不可或缺的技能（≤10），简历侧尽可能广地扩散提取。
+# 只要候选人的宽技能集覆盖了岗位的核心要求即视为匹配。评估看「岗位核心被简历覆盖率」而非交集。
+SKILL_EXTRACT_SYSTEM = """你是一个技能提取助手。根据岗位信息，只提取该岗位最核心、最不可或缺的硬技能（无论岗位是否明确写出，都抓最本质的那几个）。
 返回JSON数组格式，如：["java", "spring boot", "mysql", "redis"]
 规则：
-1. 统一使用小写字母
+1. 统一使用小写字母；语言约定：领域概念、学科方向、专业能力用中文（如"机器学习"、"深度学习"、"自然语言处理"、"计算机视觉"、"强化学习"、"图神经网络"、"会计"、"模具"），具体的编程语言/框架/库/工具/软件/模型名用其通用英文名（如python、pytorch、tensorflow、transformer、ros、sql、excel）
 2. 只保留核心词，去掉多余修饰（如"plc编程"→"plc"，"c语言"→"c"，"cad制图"→"cad"）
-3. 同一技能只保留最具体的表述，不要同时出现上位词和下位词（如有"机械设计"就不要再出"机械"）
-4. 提取范围包括：技术栈、专业领域知识、行业工具、专业资质能力等
-5. 不提取纯软技能（如沟通能力、团队协作、学习能力、积极主动）
-6. 不提取过于宽泛的标签（如"办公软件"、"windows"）
-7. 如果岗位完全没有专业能力要求（纯看态度和素质），返回空数组 []
-8. 最多15个，按重要性排序
-9. 只返回JSON数组，不要其他内容"""
+3. 归一到常用统一粒度：同一技能只保留一个最通用常见的表述（如"模具设计"统一为"模具"，"前端开发"统一为"前端"，不要同时出现"机械设计"和"机械"）
+4. 优先使用具体工具/技术名，不要用抽象概念词（如用"kafka"不用"消息队列"，用"redis"不用"缓存"，用"mysql"不用"数据库"）
+5. 只提取具体、有明确名称的硬技能：编程语言、框架、软件、工具、平台、专业资质证书、明确的技术/工艺名称，以及通用学科名（如"会计"、"模具"）
+6. 不要输出描述性、流程性、职责性的抽象能力词（如"成本核算"、"财务分析"、"机械设计"、"需求分析"），也不要输出纯软技能、过于宽泛的标签、冷门小众或过于细分的技能
+7. 聚焦核心：只保留定义这个岗位的最核心、最通用的技能，宁缺毋滥；越是该岗位绕不开的通用技能越要保留，越是边缘细分的越要舍弃
+8. 最多提取10个，按重要性排序
+9. 如果岗位完全没有专业能力要求（纯看态度和素质），返回空数组 []
+10. 只返回JSON数组，不要其他内容
+示例1（技术岗，只留核心）：
+输入：负责后端接口开发，要求熟悉Spring Boot、MyBatis、Redis
+输出：["java", "spring boot", "mysql", "redis", "sql"]
+示例2（财务岗，只留核心具体硬技能）：
+输入：负责费用管理与审核，会计学相关专业，熟练使用SAP
+输出：["会计", "sap", "excel"]
+示例3（制造岗，只留核心具体硬技能）：
+输入：负责模具开发，要求熟悉SolidWorks三维建模
+输出：["模具", "solidworks", "cad"]
+示例4（无专业能力）：
+输入：要求具备较强的沟通能力和创新意识，积极主动
+输出：[]"""
 
 
 # ──────────── 公司补充 ────────────

app/config/settings.py

@@ -14,7 +14,7 @@ class Settings(BaseSettings):
     pg_password: str = ""
     pg_db: str = "postgres"
     pg_pool_size: int = 10
-    pg_max_overflow: int = 20
+    pg_max_overflow: int = 10
 
     # ──────────── MySQL（业务库）────────────
     db_host: str = "192.168.31.105"
@@ -23,7 +23,10 @@ class Settings(BaseSettings):
     db_password: str = "123456"
     db_name: str = "offerpie"
     mysql_pool_size: int = 10
-    mysql_max_overflow: int = 20
+    mysql_max_overflow: int = 10
+
+    # 连接池获取连接的等待超时（秒），两个数据源共用
+    db_pool_timeout: int = 30
 
     # ──────────── AI 供应商 ────────────
     # 火山引擎（OpenAI 兼容风格）
@@ -34,20 +37,36 @@ class Settings(BaseSettings):
     anthropic_api_key: str = "sk-43ccdb29caa7e9ebe0db8ac0958c63f6d3a2d62e59064d3d26d94332055a9bc9"
     anthropic_base_url: str = "https://code.warpdevloper.cloud"
 
+    # Claude 备用中转（Anthropic 风格）
+    anthropic_alt_api_key: str = "sk-ac5544257dcf843de887fa153c4f4a11fa5007cbae4b8f4eec895a354c89ff7b"
+    anthropic_alt_base_url: str = "http://98.142.250.143:8080"
+
+    # OpenAI（OpenAI 兼容风格）
+    openai_api_key: str = "sk-12663c3a1ae5db0bc9aee78ffa64e7a0b95bcf7a15541040dd2bce573130f5c8"
+    openai_base_url: str = "https://re.94xy.cn/v1"
+
     # ──────────── 岗位清洗参数 ────────────
-    clean_batch_size: int = 100
-    clean_concurrency: int = 80
-    clean_interval_seconds: int = 200
+    clean_worker_count: int = 60  # 持续消费 worker 协程数量
+    clean_idle_sleep: float = 5.0  # 没数据时每个 worker 休眠秒数
     clean_total_limit: int = 0  # 累计清洗总数上限，达到后停止清洗任务；0 = 不限制
 
     # ──────────── 公司补充参数 ────────────
-    company_batch_size: int = 20
-    company_concurrency: int = 10
-    company_interval_seconds: int = 300
+    company_worker_count: int = 10  # 持续消费 worker 协程数量
+    company_idle_sleep: float = 10.0  # 没数据时每个 worker 休眠秒数
 
     # ──────────── 岗位下架参数 ────────────
     job_expire_days: int = 7
 
+    # ──────────── 阿里云 OSS ────────────
+    oss_access_key_id: str = ""
+    oss_access_key_secret: str = ""
+    oss_endpoint: str = "oss-cn-guangzhou.aliyuncs.com"
+    oss_bucket: str = "offerpie"
+    # 固定上传目录（对象 key 前缀）
+    oss_upload_dir: str = "company/logo"
+    # 下载访问域名
+    oss_domain: str = "https://offerpie.oss-cn-guangzhou.aliyuncs.com"
+
     # ──────────── 日志 ────────────
     logging_level: str = "INFO"
     log_file_name: str = "cleaner.log"

app/core/database.py

@@ -39,7 +39,9 @@ async def init_db() -> None:
         settings.pg_url,
         pool_size=settings.pg_pool_size,
         max_overflow=settings.pg_max_overflow,
+        pool_timeout=settings.db_pool_timeout,
         pool_recycle=3600,
+        pool_pre_ping=True,
         echo=False,
     )
     _pg_session_factory = async_sessionmaker(_pg_engine, expire_on_commit=False)
@@ -48,7 +50,9 @@ async def init_db() -> None:
         settings.mysql_url,
         pool_size=settings.mysql_pool_size,
         max_overflow=settings.mysql_max_overflow,
+        pool_timeout=settings.db_pool_timeout,
         pool_recycle=3600,
+        pool_pre_ping=True,
         echo=False,
     )
     _mysql_session_factory = async_sessionmaker(_mysql_engine, expire_on_commit=False)

app/core/oss.py

@@ -0,0 +1,42 @@
+"""阿里云 OSS 上传：往固定目录上传，返回下载地址"""
+
+import asyncio
+import base64
+import os
+from uuid import uuid4
+
+import oss2
+
+from app.config import settings
+from app.core.logger import log
+
+_auth = oss2.Auth(settings.oss_access_key_id, settings.oss_access_key_secret)
+_bucket = oss2.Bucket(_auth, f"https://{settings.oss_endpoint}", settings.oss_bucket)
+
+
+async def upload(data: bytes, file_name: str) -> str:
+    """上传字节内容到固定目录，返回下载地址"""
+    key = f"{settings.oss_upload_dir}/{uuid4().hex[:18]}{os.path.splitext(file_name)[1]}"
+    await asyncio.to_thread(_bucket.put_object, key, data)
+    url = f"{settings.oss_domain}/{key}"
+    log.info("OSS 上传成功 [{}] -> {}", file_name, url)
+    return url
+
+
+async def upload_base64(b64: str, file_name: str = "image.png") -> str | None:
+    """上传 base64 图片到固定目录，返回下载地址；空值或解码失败返回 None
+
+    :param b64: base64 字符串，兼容 data URI 前缀（data:image/png;base64,xxx）
+    :param file_name: 用于取后缀，默认 png
+    """
+    if not b64 or not b64.strip():
+        return None
+    # 去掉 data URI 前缀
+    if "," in b64 and b64.strip().startswith("data:"):
+        b64 = b64.split(",", 1)[1]
+    try:
+        data = base64.b64decode(b64)
+    except Exception as e:
+        log.warning("base64 解码失败: {}", e)
+        return None
+    return await upload(data, file_name)

app/main.py

@@ -1,16 +1,16 @@
-"""项目入口：初始化数据源、加载字典、启动调度器"""
+"""项目入口：初始化数据源、加载字典、启动持续消费 worker"""
 
 import asyncio
 import logging
 import signal
 import warnings
-from datetime import datetime
 
 from app.core.logger import log
 
 # 屏蔽 asyncmy INSERT IGNORE 产生的 Duplicate entry warnings
 warnings.filterwarnings("ignore", message=".*Duplicate entry.*")
 logging.getLogger("asyncmy").setLevel(logging.ERROR)
+
 from app.core.database import init_db, close_db
 from app.services.dict_cache_service import dict_cache
 from app.scheduler.tasks import create_scheduler
@@ -27,25 +27,29 @@ async def main():
     # 加载字典缓存
     await dict_cache.refresh()
 
-    # 创建并启动调度器
+    # 创建并启动调度器（僵尸恢复、岗位下架等辅助任务）
     scheduler = create_scheduler()
     scheduler.start()
+    log.info("调度器已启动，辅助定时任务已注册")
 
-    # 立即触发一次岗位清洗和公司补充
-    scheduler.modify_job("job_clean", next_run_time=datetime.now())
-    scheduler.modify_job("company_clean", next_run_time=datetime.now())
+    # 启动持续消费任务
+    from app.services.job_clean_service import run_job_clean, stop_job_clean
+    from app.services.company_clean_service import run_company_clean, stop_company_clean
 
-    log.info("调度器已启动，所有定时任务已注册")
+    job_clean_task = asyncio.create_task(run_job_clean())
+    company_clean_task = asyncio.create_task(run_company_clean())
+    log.info("岗位清洗 & 公司补充持续消费任务已启动")
 
     # 优雅关闭
     stop_event = asyncio.Event()
 
     def _shutdown(*args):
         log.info("收到关闭信号，正在关闭...")
+        stop_job_clean()
+        stop_company_clean()
         stop_event.set()
 
     loop = asyncio.get_running_loop()
-    # Unix: SIGINT + SIGTERM，Windows: 仅靠 KeyboardInterrupt
     for sig in (signal.SIGINT, signal.SIGTERM):
         try:
             loop.add_signal_handler(sig, _shutdown)
@@ -55,11 +59,14 @@ async def main():
     try:
         await stop_event.wait()
     except KeyboardInterrupt:
-        pass
-    finally:
-        scheduler.shutdown(wait=False)
-        await close_db()
-        log.info("OfferPie Job Cleaner 已关闭")
+        _shutdown()
+
+    # 等待 worker 协程退出
+    await asyncio.gather(job_clean_task, company_clean_task, return_exceptions=True)
+
+    scheduler.shutdown(wait=False)
+    await close_db()
+    log.info("OfferPie Job Cleaner 已关闭")
 
 
 if __name__ == "__main__":

app/models/pg/ai_call_log.py

@@ -0,0 +1,22 @@
+"""PostgreSQL: ai_call_log 表模型"""
+
+from datetime import datetime
+from typing import Optional
+
+from sqlalchemy import BigInteger, DateTime, String, Text
+from sqlalchemy.orm import Mapped, mapped_column
+
+from app.core.database import PgBase
+
+
+class AiCallLog(PgBase):
+    """AI 调用日志表"""
+
+    __tablename__ = "ai_call_log"
+
+    id: Mapped[int] = mapped_column(BigInteger, primary_key=True, comment="主键ID(雪花)")
+    scene: Mapped[str] = mapped_column(String(32), nullable=False, comment="场景标识: structure/major_match/skill_extract/company_enrich")
+    system_prompt: Mapped[str] = mapped_column(Text, nullable=False, comment="系统提示词")
+    user_message: Mapped[str] = mapped_column(Text, nullable=False, comment="发给AI的用户消息")
+    response: Mapped[Optional[str]] = mapped_column(Text, comment="AI返回原文")
+    created_at: Mapped[datetime] = mapped_column(DateTime, nullable=False, comment="创建时间")

app/models/pg/app_url_list.py

@@ -0,0 +1,35 @@
+"""PostgreSQL: app_url_list 表模型"""
+
+from datetime import datetime
+from typing import Optional
+
+from sqlalchemy import BigInteger, DateTime, Integer, String, Text
+from sqlalchemy.orm import Mapped, mapped_column
+
+from app.core.database import PgBase
+
+
+class AppUrlList(PgBase):
+    """爬虫 URL 任务列表"""
+
+    __tablename__ = "app_url_list"
+
+    id: Mapped[int] = mapped_column(BigInteger, primary_key=True, autoincrement=True, comment="自增主键")
+    status: Mapped[str] = mapped_column(String(32), default="pending", nullable=False, comment="任务状态")
+    input_company_name: Mapped[str] = mapped_column(String(255), nullable=False, comment="输入公司名")
+    input_url: Mapped[Optional[str]] = mapped_column(String(1024), comment="输入URL")
+    output_url: Mapped[Optional[str]] = mapped_column(String(1024), comment="输出URL")
+    post_login: Mapped[Optional[int]] = mapped_column(Integer, comment="发布登录")
+    crawl_login: Mapped[Optional[int]] = mapped_column(Integer, comment="爬取登录")
+    type_sz: Mapped[Optional[int]] = mapped_column(Integer, comment="类型sz")
+    type_sx: Mapped[Optional[int]] = mapped_column(Integer, comment="类型sx")
+    crawlok_at: Mapped[Optional[str]] = mapped_column(String(32), comment="-2没有数据或者数据全部过期或者数据缺失 -3 过滤后没数据， -4无法修复，")
+    pyname: Mapped[Optional[str]] = mapped_column(String(255), comment="脚本名字")
+    type_xz: Mapped[Optional[int]] = mapped_column(Integer, comment="类型xz")
+    error_message: Mapped[Optional[str]] = mapped_column(Text, comment="错误信息")
+    logo: Mapped[Optional[str]] = mapped_column(Text, comment="公司logo (base64编码的64x64 PNG图片)")
+    started_at: Mapped[Optional[datetime]] = mapped_column(DateTime, comment="开始时间")
+    finished_at: Mapped[Optional[datetime]] = mapped_column(DateTime, comment="完成时间")
+    created_at: Mapped[datetime] = mapped_column(DateTime, nullable=False, comment="创建时间")
+    updated_at: Mapped[datetime] = mapped_column(DateTime, nullable=False, comment="更新时间")
+    output_company_name: Mapped[Optional[str]] = mapped_column(String(255), comment="输出公司名")

app/scheduler/tasks.py

@@ -1,6 +1,11 @@
-"""定时任务注册"""
+"""定时任务注册
 
-from datetime import datetime, timedelta
+岗位清洗和公司补充改为持续消费模型（启动即运行），
+其余辅助任务保留定时调度。
+"""
+
+import asyncio
+from datetime import datetime
 
 from apscheduler.schedulers.asyncio import AsyncIOScheduler
 from apscheduler.triggers.interval import IntervalTrigger
@@ -11,30 +16,12 @@ from app.core.logger import log
 
 
 def create_scheduler() -> AsyncIOScheduler:
-    """创建并注册所有定时任务"""
+    """创建并注册辅助定时任务（岗位清洗/公司补充由 main 直接启动）"""
     scheduler = AsyncIOScheduler(
         timezone="Asia/Shanghai",
         job_defaults={"misfire_grace_time": 60},
     )
 
-    # 岗位清洗（每 N 秒）
-    scheduler.add_job(
-        _job_clean_task,
-        trigger=IntervalTrigger(seconds=settings.clean_interval_seconds),
-        id="job_clean",
-        name="岗位清洗",
-        max_instances=1,
-    )
-
-    # 公司补充（每 N 秒）
-    scheduler.add_job(
-        _company_clean_task,
-        trigger=IntervalTrigger(seconds=settings.company_interval_seconds),
-        id="company_clean",
-        name="公司补充",
-        max_instances=1,
-    )
-
     # 岗位僵尸恢复（每30分钟）
     scheduler.add_job(
         _job_zombie_task,
@@ -65,22 +52,6 @@ def create_scheduler() -> AsyncIOScheduler:
     return scheduler
 
 
-async def _job_clean_task():
-    from app.services.job_clean_service import run_job_clean
-    try:
-        await run_job_clean()
-    except Exception as e:
-        log.error("岗位清洗任务异常: {}", e)
-
-
-async def _company_clean_task():
-    from app.services.company_clean_service import run_company_clean
-    try:
-        await run_company_clean()
-    except Exception as e:
-        log.error("公司补充任务异常: {}", e)
-
-
 async def _job_zombie_task():
     from app.services.zombie_recover_service import recover_job_zombie
     try:

app/services/ai_tool.py

@@ -1,14 +1,24 @@
-"""AI 调用工具封装"""
+"""AI 调用工具封装
 
+每个协程独立重试，失败后 sleep 200ms 再试，直到成功。
+不再有全局门闸，互不影响。
+"""
+
+import asyncio
 import re
-from typing import Any
+from datetime import datetime
+from typing import Any, Optional
 
 from json_repair import repair_json
 from langchain_core.language_models import BaseChatModel
 from langchain_core.messages import SystemMessage, HumanMessage
+from snowflake import SnowflakeGenerator
 
 from app.core.logger import log
 
+# AI 日志专用雪花ID
+_log_id_gen = SnowflakeGenerator(instance=2)
+
 # 匹配 <think>任意内容</think>，用于剥离推理模型的思考过程
 _THINK_RE = re.compile(r"<think>.*?</think>", re.DOTALL | re.IGNORECASE)
 
@@ -28,19 +38,31 @@ def parse_llm_json(text: str) -> Any:
     return repair_json(cleaned, return_objects=True)
 
 
-async def ai_chat(llm: BaseChatModel, system_prompt: str, user_message: str) -> str:
-    """异步调用 LLM，返回原始文本"""
-    messages = [
-        SystemMessage(content=system_prompt),
-        HumanMessage(content=user_message),
-    ]
-    response = await llm.ainvoke(messages)
-    return response.content
+async def ai_chat(llm: BaseChatModel, system_prompt: str, user_message: str, scene: Optional[str] = None) -> str:
+    """异步调用 LLM，返回原始文本。失败后独立重试直到成功。"""
+    attempt = 0
+
+    while True:
+        attempt += 1
+        try:
+            messages = [
+                SystemMessage(content=system_prompt),
+                HumanMessage(content=user_message),
+            ]
+            response = await llm.ainvoke(messages)
+            content = response.content
+            # 记录 AI 调用日志（失败不影响主流程）
+            if scene:
+                await _save_call_log(scene, system_prompt, user_message, content)
+            return content
+        except Exception as e:
+            log.warning("AI 调用失败(第{}次), 200ms 后重试: {}", attempt, e)
+            await asyncio.sleep(0.2)
 
 
-async def ai_chat_json(llm: BaseChatModel, system_prompt: str, user_message: str) -> Any:
+async def ai_chat_json(llm: BaseChatModel, system_prompt: str, user_message: str, scene: Optional[str] = None) -> Any:
     """异步调用 LLM，返回解析后的 JSON 对象"""
-    raw = await ai_chat(llm, system_prompt, user_message)
+    raw = await ai_chat(llm, system_prompt, user_message, scene=scene)
     if not raw or not raw.strip():
         log.warning("AI 返回为空")
         return None
@@ -49,3 +71,26 @@ async def ai_chat_json(llm: BaseChatModel, system_prompt: str, user_message: str
     except Exception as e:
         log.warning("AI JSON 解析失败: {}, raw={}", e, raw[:200])
         return None
+
+
+async def _save_call_log(scene: str, system_prompt: str, user_message: str, response: Optional[str]) -> None:
+    """异步写入 AI 调用日志到 PG，失败只 warning 不影响主流程"""
+    try:
+        from app.core.database import PgSession
+        from sqlalchemy import insert
+        from app.models.pg.ai_call_log import AiCallLog
+
+        async with PgSession() as pg:
+            await pg.execute(
+                insert(AiCallLog).values(
+                    id=next(_log_id_gen),
+                    scene=scene,
+                    system_prompt=system_prompt,
+                    user_message=user_message,
+                    response=response,
+                    created_at=datetime.now(),
+                )
+            )
+            await pg.commit()
+    except Exception as e:
+        log.warning("AI 调用日志写入失败: {}", e)

app/services/company_clean_service.py

@@ -1,4 +1,8 @@
-"""公司数据补充服务（协程版）"""
+"""公司数据补充服务（持续消费模型）
+
+启动 N 个 worker 协程，每个 worker 循环：取一条 → 处理 → 取下一条。
+没数据时短暂休眠后重试。
+"""
 
 import asyncio
 from datetime import datetime
@@ -13,47 +17,65 @@ from app.ai.prompts import COMPANY_ENRICH_SYSTEM
 from app.services.ai_tool import ai_chat_json
 from app.services.dict_cache_service import dict_cache
 
+# 停止信号
+_stop_event = asyncio.Event()
+
+
+def stop_company_clean():
+    """外部调用，通知所有 worker 停止"""
+    _stop_event.set()
+
 
 async def run_company_clean() -> None:
-    """一次批量公司补充任务"""
-    # 锁定一批待完善公司
+    """启动 N 个 worker 协程持续消费"""
+    _stop_event.clear()
+    worker_count = settings.company_worker_count
+    log.info("公司补充：启动 {} 个 worker 协程", worker_count)
+
+    workers = [asyncio.create_task(_worker(i)) for i in range(worker_count)]
+    await asyncio.gather(*workers)
+
+    log.info("公司补充：所有 worker 已退出")
+
+
+async def _worker(worker_id: int) -> None:
+    """单个 worker：循环取一条、处理一条"""
+    while not _stop_event.is_set():
+        data = await _fetch_one()
+        if data is None:
+            await asyncio.sleep(settings.company_idle_sleep)
+            continue
+
+        try:
+            await _do_clean(data)
+        except Exception as e:
+            log.error("[worker-{}] 公司补充异常, id={}, shortName={}: {}",
+                      worker_id, data["id"], data.get("short_name"), e)
+
+
+async def _fetch_one() -> dict | None:
+    """从 MySQL 锁定一条待完善公司并标记为 status=3"""
     async with MysqlSession() as mysql:
+        # MySQL 不支持 UPDATE ... RETURNING，分两步
         result = await mysql.execute(
             text("""
                 SELECT * FROM bg_company
                 WHERE status = 0
-                LIMIT :limit
+                LIMIT 1
                 FOR UPDATE SKIP LOCKED
             """),
-            {"limit": settings.company_batch_size},
         )
-        rows = result.mappings().all()
-        if not rows:
-            return
+        row = result.mappings().first()
+        if not row:
+            return None
 
-        ids = [r["id"] for r in rows]
-        # MySQL 批量 IN 用 format 拼接（id 是 bigint，安全）
-        ids_str = ",".join(str(i) for i in ids)
+        company_id = row["id"]
         await mysql.execute(
-            text(f"UPDATE bg_company SET status = 3, update_time = NOW() WHERE id IN ({ids_str})"),
+            text("UPDATE bg_company SET status = 3, update_time = NOW() WHERE id = :id"),
+            {"id": company_id},
         )
         await mysql.commit()
-
-    log.info("公司补充：锁定{}条数据", len(rows))
-
-    # 协程并发，信号量限流
-    sem = asyncio.Semaphore(settings.company_concurrency)
-    tasks = [_clean_one(sem, dict(r)) for r in rows]
-    await asyncio.gather(*tasks, return_exceptions=True)
-
-
-async def _clean_one(sem: asyncio.Semaphore, company: dict) -> None:
-    """单条公司补充"""
-    async with sem:
-        try:
-            await _do_clean(company)
-        except Exception as e:
-            log.error("公司补充异常, id={}, shortName={}: {}", company["id"], company.get("short_name"), e)
+        return dict(row)
 
 
 async def _do_clean(company: dict) -> None:
@@ -62,7 +84,7 @@ async def _do_clean(company: dict) -> None:
     short_name = company.get("short_name", "")
 
     user_msg = f"【公司简称】\n{short_name}\n\n【行业列表】\n{dict_cache.industry_text}"
-    result = await ai_chat_json(CompanyCleanModel.ENRICH, COMPANY_ENRICH_SYSTEM, user_msg)
+    result = await ai_chat_json(CompanyCleanModel.ENRICH, COMPANY_ENRICH_SYSTEM, user_msg, scene="company_enrich")
 
     if result is None or not result.get("valid", False):
         await _update_status(company_id, 4)

app/services/dict_cache_service.py

@@ -64,9 +64,10 @@ class DictCacheService:
                 for m in majors
             )
 
-            # 地区（省市级）
+            # 地区（仅市级：province_code 不为空且 city_code 为空；直辖市取市级避免与省级同名重复）
             result = await session.execute(text("""
-                SELECT code, name FROM bg_china_regions_code WHERE city_code IS NULL
+                SELECT code, name FROM bg_china_regions_code
+                WHERE city_code IS NULL AND province_code IS NOT NULL
             """))
             self._region_list = [dict(r) for r in result.mappings().all()]
 

app/services/job_clean_service.py

@@ -1,4 +1,8 @@
-"""岗位清洗服务（协程版）"""
+"""岗位清洗服务（持续消费模型）
+
+启动 N 个 worker 协程，每个 worker 循环：取一条 → 处理 → 取下一条。
+没数据时短暂休眠后重试。
+"""
 
 import asyncio
 import json
@@ -10,6 +14,7 @@ from sqlalchemy import text, insert
 from app.config import settings
 from app.core.database import PgSession, MysqlSession
 from app.core.logger import log
+from app.core.oss import upload_base64
 from app.ai.model_config import JobCleanModel
 from app.ai.prompts import JOB_STRUCTURE_SYSTEM, MAJOR_MATCH_SYSTEM, SKILL_EXTRACT_SYSTEM
 from app.models.mysql.job import Job
@@ -27,70 +32,83 @@ _company_lock = asyncio.Lock()
 # 累计已清洗总数（用于总数上限控制，进程内统计）
 _cleaned_total = 0
 
+# 停止信号
+_stop_event = asyncio.Event()
+
 
 def is_clean_limit_reached() -> bool:
     """是否已达到累计清洗总数上限（clean_total_limit=0 表示不限制）"""
     return 0 < settings.clean_total_limit <= _cleaned_total
 
 
+def stop_job_clean():
+    """外部调用，通知所有 worker 停止"""
+    _stop_event.set()
+
+
 async def run_job_clean() -> None:
-    """一次批量清洗任务"""
+    """启动 N 个 worker 协程持续消费，直到收到停止信号或达到上限"""
+    _stop_event.clear()
+    worker_count = settings.clean_worker_count
+    log.info("岗位清洗：启动 {} 个 worker 协程", worker_count)
+
+    workers = [asyncio.create_task(_worker(i)) for i in range(worker_count)]
+    await asyncio.gather(*workers)
+
+    log.info("岗位清洗：所有 worker 已退出，累计清洗 {} 条", _cleaned_total)
+
+
+async def _worker(worker_id: int) -> None:
+    """单个 worker：循环取一条、处理一条"""
     global _cleaned_total
 
-    # 总数上限：已达上限直接跳过
-    if is_clean_limit_reached():
-        return
+    while not _stop_event.is_set():
+        # 总数上限检查
+        if is_clean_limit_reached():
+            return
 
-    # 1. 从 PG 锁定一批待清洗数据
+        # 从 PG 取一条待处理数据
+        data = await _fetch_one()
+        if data is None:
+            # 没数据，休眠后重试
+            await asyncio.sleep(settings.clean_idle_sleep)
+            continue
+
+        # 处理
+        try:
+            await _do_clean(data)
+            _cleaned_total += 1
+        except Exception as e:
+            log.error("[worker-{}] 岗位清洗异常, id={}: {}", worker_id, data["id"], e)
+
+        # 达到上限后通知所有 worker 停止
+        if is_clean_limit_reached():
+            log.info("岗位清洗：已达累计上限 {} 条，停止", settings.clean_total_limit)
+            _stop_event.set()
+            return
+
+
+async def _fetch_one() -> dict | None:
+    """从 PG 锁定一条待清洗数据并标记为 cleaning"""
     async with PgSession() as pg:
         result = await pg.execute(
-            text("""
-                SELECT * FROM app_job_data
-                WHERE clean_status = 'pending' AND  recruit_category in (0,1,2)
-                LIMIT :limit
-                FOR UPDATE SKIP LOCKED
-            """),
-            {"limit": settings.clean_batch_size},
-        )
-        rows = result.mappings().all()
-        if not rows:
-            return
-
-        ids = [r["id"] for r in rows]
-        await pg.execute(
             text("""
                 UPDATE app_job_data
                 SET clean_status = 'cleaning', clean_started_at = NOW()
-                WHERE id = ANY(:ids)
+                WHERE id = (
+                    SELECT id FROM app_job_data
+                    WHERE clean_status = 'pending' AND verify_status = 'success' AND recruit_category IN (0,1,2)
+                    LIMIT 1
+                    FOR UPDATE SKIP LOCKED
+                )
+                RETURNING *
             """),
-            {"ids": ids},
         )
-        await pg.commit()
-
-    log.info("岗位清洗：锁定{}条数据", len(rows))
-
-    # 2. 协程并发清洗，信号量限流
-    sem = asyncio.Semaphore(settings.clean_concurrency)
-    tasks = [_clean_one(sem, dict(r)) for r in rows]
-    results = await asyncio.gather(*tasks, return_exceptions=True)
-
-    # 汇总
-    errors = sum(1 for r in results if isinstance(r, Exception))
-    _cleaned_total += len(rows)
-    log.info("岗位清洗：本批完成，共{}条，异常{}条，累计{}条", len(rows), errors, _cleaned_total)
-
-    if is_clean_limit_reached():
-        log.info("岗位清洗：已达累计上限 {} 条，停止清洗任务", settings.clean_total_limit)
-
-
-async def _clean_one(sem: asyncio.Semaphore, data: dict) -> None:
-    """单条岗位清洗"""
-    async with sem:
-        try:
-            await _do_clean(data)
-        except Exception as e:
-            log.error("岗位清洗异常, id={}: {}", data["id"], e)
-            # 保持 cleaning 状态，由僵尸恢复任务重置
+        row = result.mappings().first()
+        if row:
+            await pg.commit()
+            return dict(row)
+        return None
 
 
 async def _do_clean(data: dict) -> None:
@@ -125,8 +143,8 @@ async def _do_clean(data: dict) -> None:
                 return
 
     # 第一次AI：结构化提取
-    user_message = _build_user_message(data)
-    result = await ai_chat_json(JobCleanModel.STRUCTURE, JOB_STRUCTURE_SYSTEM, user_message)
+    user_message = await _build_user_message(data)
+    result = await ai_chat_json(JobCleanModel.STRUCTURE, JOB_STRUCTURE_SYSTEM, user_message, scene="structure")
     if result is None or not result.get("valid", False):
         log.info("[id={}] 丢弃：AI判定无效", data_id)
         await _update_pg_status(data_id, "discarded")
@@ -134,7 +152,7 @@ async def _do_clean(data: dict) -> None:
 
     # 公司处理
     company_short_name = result.get("companyShortName") or data.get("company") or ""
-    company_id = await _find_or_create_company(company_short_name)
+    company_id = await _find_or_create_company(company_short_name, data.get("urllistid"))
 
     # 地区处理
     region_codes = []
@@ -216,7 +234,7 @@ async def _match_major(job_id: int, result: dict) -> None:
     req = result.get("requirement", "")
     user_msg = f"【岗位信息】\n标题: {title}\n职责: {desc}\n要求: {req}\n\n【专业分类列表】\n{dict_cache.major_category_text}"
 
-    data = await ai_chat_json(JobCleanModel.MAJOR_MATCH, MAJOR_MATCH_SYSTEM, user_msg)
+    data = await ai_chat_json(JobCleanModel.MAJOR_MATCH, MAJOR_MATCH_SYSTEM, user_msg, scene="major_match")
     if data is None:
         return
 
@@ -241,7 +259,7 @@ async def _extract_skill_tags(job_id: int, result: dict) -> None:
     req = result.get("requirement", "")
     user_msg = f"【岗位信息】\n标题: {title}\n职责: {desc}\n要求: {req}"
 
-    skills = await ai_chat_json(JobCleanModel.SKILL_EXTRACT, SKILL_EXTRACT_SYSTEM, user_msg)
+    skills = await ai_chat_json(JobCleanModel.SKILL_EXTRACT, SKILL_EXTRACT_SYSTEM, user_msg, scene="skill_extract")
     if not skills or not isinstance(skills, list):
         return
 
@@ -252,7 +270,6 @@ async def _extract_skill_tags(job_id: int, result: dict) -> None:
         if not name or len(name) > 50:
             continue
 
-        # 每个 tag 单独 session，避免死锁
         real_id = await _find_or_create_skill_tag(name)
         if real_id and real_id not in tag_ids:
             tag_ids.append(real_id)
@@ -284,8 +301,8 @@ async def _find_or_create_skill_tag(name: str) -> int | None:
         return row.scalar()
 
 
-async def _find_or_create_company(short_name: str) -> int:
-    """查找或创建公司（加锁防并发重复）"""
+async def _find_or_create_company(short_name: str, urllistid: int | None = None) -> int:
+    """查找或创建公司（加锁防并发重复）；新建公司时若有 logo 则上传 OSS 并回填地址"""
     async with _company_lock:
         async with MysqlSession() as mysql:
             row = await mysql.execute(
@@ -309,7 +326,35 @@ async def _find_or_create_company(short_name: str) -> int:
                 )
             )
             await mysql.commit()
-            return company_id
+
+    # 锁外处理 logo
+    try:
+        logo_b64 = await _get_logo_base64(urllistid)
+        if logo_b64:
+            logo_url = await upload_base64(logo_b64, "logo.png")
+            if logo_url:
+                async with MysqlSession() as mysql:
+                    await mysql.execute(
+                        text("UPDATE bg_company SET logo_url = :url, update_time = :t WHERE id = :id"),
+                        {"url": logo_url, "t": datetime.now(), "id": company_id},
+                    )
+                    await mysql.commit()
+    except Exception as e:
+        log.warning("[company={}] logo 上传失败: {}", company_id, e)
+
+    return company_id
+
+
+async def _get_logo_base64(urllistid: int | None) -> str | None:
+    """从 PG app_url_list 按 urllistid 读取 logo（base64）"""
+    if not urllistid:
+        return None
+    async with PgSession() as pg:
+        row = await pg.execute(
+            text("SELECT logo FROM app_url_list WHERE id = :id"),
+            {"id": urllistid},
+        )
+        return row.scalar()
 
 
 async def _update_pg_status(data_id: int, status: str) -> None:
@@ -328,14 +373,18 @@ async def _update_pg_status(data_id: int, status: str) -> None:
         await pg.commit()
 
 
-def _build_user_message(data: dict) -> str:
+async def _build_user_message(data: dict) -> str:
     """构建第一次AI的用户消息"""
+    # 从 app_url_list 取来源公司名，作为 companyShortName 的参考
+    input_company_name = await _get_input_company_name(data.get("urllistid"))
+
     parts = [
         "【原始数据】",
         f"岗位名称: {data.get('job_title') or ''}",
         f"薪资: {data.get('salary') or ''}",
         f"工作地点: {data.get('location') or ''}",
         f"公司: {data.get('company') or ''}",
+        f"来源公司名(参考): {input_company_name or ''}",
         f"经验要求: {data.get('experience') or ''}",
         f"学历要求: {data.get('education') or ''}",
         f"岗位详情: {data.get('description') or ''}",
@@ -345,3 +394,15 @@ def _build_user_message(data: dict) -> str:
         f"【行业列表】\n{dict_cache.industry_text}",
     ]
     return "\n".join(parts)
+
+
+async def _get_input_company_name(urllistid: int | None) -> str | None:
+    """从 PG app_url_list 按 urllistid 读取 input_company_name"""
+    if not urllistid:
+        return None
+    async with PgSession() as pg:
+        row = await pg.execute(
+            text("SELECT input_company_name FROM app_url_list WHERE id = :id"),
+            {"id": urllistid},
+        )
+        return row.scalar()

requirements.txt

@@ -13,4 +13,5 @@ langchain-core>=0.3
 # 工具
 loguru>=0.7
 snowflake-id>=1.0
-json-repair>=0.30
+json-repair>=0.30
+oss2>=2.18