Remove CLS collection and support a full news AI reset

backend/src/alphapulse/crawler/events.py

@@ -11,8 +11,8 @@ LOG_CHANNEL = f"{KEY_PREFIX}pub:logs"
 NEWS_CHANNEL = f"{KEY_PREFIX}pub:news"
 CONTROL_CHANNEL = f"{KEY_PREFIX}ctl:crawler"
 STATUS_KEY = f"{KEY_PREFIX}crawler:status"
-# 列表接口支持按时间点往前翻的来源；cls_depth 列表只能翻约 3 页，无法回补历史
-HISTORY_SOURCES = ("cls_telegraph", "wscn_live", "wscn_news")
+# Sources whose list endpoints support historical pagination.
+HISTORY_SOURCES = ("wscn_live", "wscn_news")
 
 
 def start_urls_key(source_code: str) -> str:

backend/src/alphapulse/crawler/remove_cls.py

@@ -0,0 +1,117 @@
+"""Fixed-scope operator cleanup; callers must stop all application writers."""
+
+import argparse
+import asyncio
+import json
+
+from sqlalchemy import text
+
+from alphapulse.core.db import engine
+
+SOURCES = ('cls_depth', 'cls_telegraph')
+AI_TABLES = (
+    'news_research_references', 'news_research_versions', 'news_research_memberships',
+    'news_research_attempts', 'news_research_packages', 'news_research_events',
+    'news_ai_reprocess_targets', 'news_ai_reprocess_items', 'news_ai_reprocess_jobs',
+    'news_ai_event_links',
+    'news_ai_event_versions', 'news_ai_event_members', 'news_ai_events',
+    'news_ai_package_corrections', 'news_package_evidence', 'news_ai_summary_versions',
+    'news_ai_grouping_versions', 'news_ai_summary_articles', 'news_ai_summaries',
+    'news_ai_article_analyses', 'risk_snapshots', 'risk_analyses', 'risk_events', 'news_signals',
+)
+# These retired models were never included in the migration chain.
+OPTIONAL_AI_TABLES = ('news_ai_event_evidence', 'news_ai_event_corrections')
+SOURCE_TABLES = ('news_items', 'raw_responses', 'crawl_logs', 'crawl_runs', 'sources')
+
+
+async def retained_news(connection):
+    return (await connection.execute(text('''
+        SELECT count(*), md5(coalesce(string_agg(id::text || ':' || content_hash, ',' ORDER BY id), ''))
+        FROM news_items WHERE NOT (source_code = ANY(:sources))
+    '''), {'sources': list(SOURCES)})).one()._asdict()
+
+
+async def protected_records(connection):
+    tables = list(await connection.scalars(text("""
+        SELECT tablename FROM pg_tables WHERE schemaname='public' AND
+        (tablename IN ('ai_providers','skill_nodes','skill_versions',
+         'news_research_vocabulary_versions','chat_sessions','chat_messages','context_versions')
+         OR starts_with(tablename, 'trading_')) ORDER BY tablename
+    """)))
+    quote = connection.dialect.identifier_preparer.quote
+    result = {}
+    for table in tables:
+        result[table] = (await connection.execute(text(
+            "SELECT count(*), md5(coalesce(string_agg(digest,'' ORDER BY digest),'')) FROM "
+            '(SELECT md5(row_to_json(record)::text) AS digest FROM ' + quote(table) + ' record) records'
+        ))).one()._asdict()
+    return result
+
+
+async def cleanup(connection, execute=False):
+    params = {'sources': list(SOURCES)}
+    optional = tuple([table for table in OPTIONAL_AI_TABLES
+                     if await connection.scalar(text('SELECT to_regclass(:name)'), {'name': table})])
+    ai_tables = (*optional, *AI_TABLES)
+    if execute:
+        await connection.execute(text("SET LOCAL lock_timeout = '10s'"))
+        tables = (*ai_tables, 'news_revisions', *SOURCE_TABLES, 'raw_bodies',
+                  'news_research_settings', 'risk_settings')
+        await connection.execute(text('LOCK TABLE ' + ', '.join(tables) + ' IN SHARE ROW EXCLUSIVE MODE'))
+    counts = {table: await connection.scalar(text('SELECT count(*) FROM ' + table)) for table in ai_tables}
+    counts.update({table: await connection.scalar(text('SELECT count(*) FROM ' + table +
+                   ' WHERE ' + ('code' if table == 'sources' else 'source_code') + ' = ANY(:sources)'), params)
+                   for table in SOURCE_TABLES})
+    counts['news_revisions'] = await connection.scalar(text('''SELECT count(*) FROM news_revisions
+        WHERE news_id IN (SELECT id FROM news_items WHERE source_code = ANY(:sources))'''), params)
+    before = await retained_news(connection)
+    protected = await protected_records(connection)
+    if execute:
+        await connection.execute(text('''CREATE TEMP TABLE IF NOT EXISTS removed_cls_bodies ON COMMIT DROP AS
+            SELECT DISTINCT body_sha256 FROM raw_responses WHERE source_code = ANY(:sources)'''), params)
+        for table in ('news_research_settings', 'risk_settings'):
+            await connection.execute(text('UPDATE ' + table + ' SET enabled = false'))
+        await connection.execute(text('UPDATE news_research_settings SET control_revision = control_revision + 1'))
+        for table in ai_tables:
+            await connection.execute(text('DELETE FROM ' + table))
+        await connection.execute(text('''DELETE FROM news_revisions WHERE news_id IN
+            (SELECT id FROM news_items WHERE source_code = ANY(:sources))'''), params)
+        for table in SOURCE_TABLES:
+            await connection.execute(text('DELETE FROM ' + table + ' WHERE ' +
+                ('code' if table == 'sources' else 'source_code') + ' = ANY(:sources)'), params)
+        result = await connection.execute(text('''DELETE FROM raw_bodies body USING removed_cls_bodies removed
+            WHERE body.sha256 = removed.body_sha256 AND NOT EXISTS
+            (SELECT 1 FROM raw_responses response WHERE response.body_sha256 = body.sha256)'''))
+        counts['unreferenced_raw_bodies'] = result.rowcount
+        if protected != await protected_records(connection):
+            raise RuntimeError('保留业务数据校验失败，清理已回滚')
+        if before != await retained_news(connection):
+            raise RuntimeError('其他来源原文校验失败，清理已回滚')
+    return {'executed': execute, 'protected_records': protected, 'counts': counts, 'retained_news': before, 'absent_legacy_tables': sorted(set(OPTIONAL_AI_TABLES) - set(optional))}
+
+
+async def main(execute=False):
+    try:
+        async with engine.begin() as connection:
+            if not execute:
+                await connection.execute(text('SET TRANSACTION READ ONLY'))
+            result = await cleanup(connection, execute)
+        if execute:
+            from alphapulse.core.redis import redis
+            async for key in redis.scan_iter(match='ap:risk:*', count=100):
+                await redis.delete(key)
+            for source in SOURCES:
+                async for key in redis.scan_iter(match=f'ap:{source}:*', count=100):
+                    await redis.delete(key)
+            await redis.hdel('ap:crawler:status', *SOURCES)
+            await redis.delete('ap:rl:www.cls.cn')
+            await redis.aclose()
+        print(json.dumps(result, ensure_ascii=False))
+    finally:
+        await engine.dispose()
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument('--execute', action='store_true')
+    asyncio.run(main(parser.parse_args().execute))

backend/src/alphapulse/crawler/spiders/cls.py

@@ -1,113 +0,0 @@
-import hashlib
-import html
-import json
-import re
-from datetime import datetime
-from urllib.parse import urlencode
-
-from scrapy import Request
-from scrapy.http import Response
-
-from alphapulse.crawler.spiders.base import NewsSpider, ts
-
-
-def signed_url(path: str, params: dict[str, str]) -> str:
-    # 财联社网页前端的公开签名算法：md5(sha1(按键排序的查询串))
-    query = {"app": "CailianpressWeb", "os": "web", "sv": "8.4.6", **params}
-    query["sign"] = hashlib.md5(hashlib.sha1(urlencode(sorted(query.items())).encode()).hexdigest().encode()).hexdigest()
-    return f"https://www.cls.cn{path}?{urlencode(query)}"
-
-
-class ClsTelegraphSpider(NewsSpider):
-    """财联社·电报：列表即全文；翻页需 refresh_type=1"""
-
-    name = "cls_telegraph"
-    request_headers = {"Referer": "https://www.cls.cn/telegraph"}
-
-    def first_requests(self) -> list[Request]:
-        return [self.request(signed_url("/v1/roll/get_roll_list", {"category": "", "last_time": "", "rn": "20"}), self.parse_list, page=1)]
-
-    def list_request(self, last_time: int, page: int, **meta) -> Request:
-        params = {"category": "", "last_time": str(last_time), "refresh_type": "1", "rn": "20"}
-        return self.request(signed_url("/v1/roll/get_roll_list", params), self.parse_list, page=page, **meta)
-
-    def history_request(self, before: datetime) -> Request:
-        return self.list_request(int(before.timestamp()), 1, history=True)
-
-    async def parse_list(self, response: Response):
-        body = response.json()
-        if body["errno"] != 0:
-            raise ValueError(f"接口返回错误：{body}")
-        rows = [x for x in body["data"]["roll_data"] if not x["is_ad"]]
-        # 先查已入库集合，再产出条目：否则管道可能先落库，导致误判为没有新条目而停止翻页
-        ids = [str(x["id"]) for x in rows]
-        known = await self.known_items(ids)
-        for x in rows:
-            yield self.item(
-                response,
-                external_id=str(x["id"]),
-                url=f"https://www.cls.cn/detail/{x['id']}",
-                title=x["title"],
-                summary=x["brief"],
-                content_html=f"<p>{html.escape(x['content'])}</p>",
-                published_at=ts(x["ctime"]),
-                source_updated_at=ts(x["modified_time"]) if x.get("modified_time") else None,
-                author=x.get("author"),
-                tags=[s["subject_name"] for s in x.get("subjects") or []],
-                extra={"level": x.get("level"), "stock_list": x.get("stock_list")},
-            )
-        page = response.meta["page"]
-        last_time = min(x["ctime"] for x in rows) if rows else None
-        oldest = ts(last_time) if rows else None
-        if response.meta.get("history"):
-            if self.history_step(oldest):
-                yield self.list_request(last_time, page + 1, history=True)
-        elif rows and self.should_continue(ids, known, oldest, page):
-            yield self.list_request(last_time, page + 1)
-
-
-class ClsDepthSpider(NewsSpider):
-    """财联社·深度（头条）：列表只有摘要，正文从详情页 __NEXT_DATA__ 解析"""
-
-    name = "cls_depth"
-    request_headers = {"Referer": "https://www.cls.cn/depth?id=1000"}
-
-    def first_requests(self) -> list[Request]:
-        # 首页不带 last_time 时 rn 无效，返回约 30 条且不按时间排序
-        return [self.request(signed_url("/v3/depth/list/1000", {}), self.parse_list, page=1)]
-
-    async def parse_list(self, response: Response):
-        body = response.json()
-        if body["errno"] != 0:
-            raise ValueError(f"接口返回错误：{body}")
-        rows = [x for x in body["data"] if not x["is_ad"]]
-        ids = [str(x["id"]) for x in rows]
-        known = await self.known_items(ids)
-        for x in rows:
-            if known.get(str(x["id"])) != (x["title"], x["brief"]):
-                # 详情页不一定有 brief，摘要以列表为准（也用于下次比较是否变化）
-                yield self.request(f"https://www.cls.cn/detail/{x['id']}", self.parse_detail, summary=x["brief"])
-        page = response.meta["page"]
-        if rows and self.should_continue(ids, known, ts(min(x["ctime"] for x in rows)), page):
-            params = {"last_time": str(min(x["ctime"] for x in rows)), "rn": "20"}
-            yield self.request(signed_url("/v3/depth/list/1000", params), self.parse_list, page=page + 1)
-
-    async def parse_detail(self, response: Response):
-        match = re.search(r'<script id="__NEXT_DATA__" type="application/json">(.*?)</script>', response.text, re.S)
-        if match is None:
-            raise ValueError("详情页缺少 __NEXT_DATA__，页面结构可能已变化")
-        a = json.loads(match.group(1))["props"]["pageProps"]["articleDetail"]
-        yield self.item(
-            response,
-            external_id=str(a["id"]),
-            url=f"https://www.cls.cn/detail/{a['id']}",
-            title=a["title"],
-            summary=response.meta["summary"],
-            content_html=a["content"],
-            published_at=ts(a["ctime"]),
-            author=(a.get("author") or {}).get("name"),
-            tags=[s["name"] for s in a.get("subject") or []],
-            # 付费内容只保存公开部分
-            is_paywalled=not a["isFree"],
-            extra={"level": a.get("level")},
-        )