feat(avatar): honor square interaction authorization

Reviewed-on: #19

backend/app/services/avatar_service.py

@@ -1,16 +1,24 @@
 """数字分身管理服务层 — 同步连接数字分身应用的 SQLite 数据库"""
+import json
 import os
+from datetime import datetime, timedelta
 from typing import Optional, Tuple
 
-from sqlalchemy import create_engine, text
+from sqlalchemy import create_engine, select, text
 from sqlalchemy.orm import sessionmaker, Session
 
 from app.core.config import settings
+from app.core.logger import logger
+from app.models import UserPersonality, VirtualUser
 
 
 _engine = None
 _SessionLocal: Optional[sessionmaker] = None
 
+AVATAR_ACCOUNT_PREFIX = "__avatar__:"
+SQUARE_INTERACTION_PERMISSION = "interact"
+SQUARE_INTERACTION_ACTIONS = frozenset({"like", "collect", "comment", "reply"})
+
 
 def _get_engine_and_session():
     global _engine, _SessionLocal
@@ -66,6 +74,185 @@ def _get_global_token_balance(db: Session) -> int:
         return 0
 
 
+def _decode_config(value) -> dict:
+    if isinstance(value, dict):
+        return value
+    if isinstance(value, str):
+        try:
+            decoded = json.loads(value)
+            return decoded if isinstance(decoded, dict) else {}
+        except (json.JSONDecodeError, ValueError):
+            return {}
+    return {}
+
+
+def is_delegated_avatar_user(user: VirtualUser | None) -> bool:
+    return bool(user and (user.account or "").startswith(AVATAR_ACCOUNT_PREFIX))
+
+
+def delegated_avatar_id(user: VirtualUser | None) -> str:
+    if not is_delegated_avatar_user(user):
+        return ""
+    return (user.account or "")[len(AVATAR_ACCOUNT_PREFIX):]
+
+
+def _list_square_interaction_authorizations(db: Session) -> list[dict]:
+    """读取已明确授权分身参与广场互动的身份与会会令牌。"""
+    rows = db.execute(text("""
+        SELECT
+            a.id AS avatar_id,
+            a.name AS avatar_name,
+            a.display_name AS avatar_display_name,
+            a.description AS avatar_description,
+            a.photo_url AS avatar_photo_url,
+            a.config AS avatar_config,
+            u.huihui_user_id,
+            u.nickname AS owner_nickname,
+            u.avatar_url AS owner_avatar_url,
+            u.huihui_token
+        FROM avatars a
+        JOIN users u ON u.huihui_user_id = a.owner_id
+        WHERE a.status = 'active'
+    """)).fetchall()
+
+    authorized = []
+    for row in rows:
+        config = _decode_config(row.avatar_config)
+        permissions = config.get("authorizationPermissions", [])
+        if not isinstance(permissions, list) or SQUARE_INTERACTION_PERMISSION not in permissions:
+            continue
+        platform_uid = str(row.huihui_user_id or "").strip()
+        token = str(row.huihui_token or "").strip()
+        if not platform_uid or not token:
+            continue
+        authorized.append({
+            "avatar_id": str(row.avatar_id),
+            "avatar_name": row.avatar_display_name or row.avatar_name or row.owner_nickname or "数字分身",
+            "avatar_description": row.avatar_description or "",
+            "avatar_url": _resolve_photo_url(row.avatar_photo_url or row.owner_avatar_url or ""),
+            "config": config,
+            "platform_uid": platform_uid,
+            "token": token,
+        })
+    return authorized
+
+
+def get_square_interaction_permissions(avatar_id: str) -> frozenset[str]:
+    """实时复核授权；数据库不可用、令牌失效或撤权时一律拒绝执行。"""
+    avatar_db = get_session()
+    if avatar_db is None:
+        return frozenset()
+    try:
+        authorized_ids = {
+            item["avatar_id"] for item in _list_square_interaction_authorizations(avatar_db)
+        }
+        return SQUARE_INTERACTION_ACTIONS if avatar_id in authorized_ids else frozenset()
+    except Exception as exc:
+        logger.error(f"读取数字分身广场互动授权失败: {exc}")
+        return frozenset()
+    finally:
+        avatar_db.close()
+
+
+def _word_count_range(config: dict) -> tuple[int, int]:
+    ranges = {
+        "short": (10, 35),
+        "medium": (20, 60),
+        "long": (30, 80),
+    }
+    return ranges.get(str(config.get("responseLength") or "medium"), (20, 60))
+
+
+async def sync_square_interaction_users(db) -> set[str]:
+    """把已授权分身同步为调度器身份，并刷新其会会会话。"""
+    avatar_db = get_session()
+    if avatar_db is None:
+        logger.warning("数字分身数据库不可用，跳过广场互动授权同步")
+        return set()
+    try:
+        authorized = _list_square_interaction_authorizations(avatar_db)
+    except Exception as exc:
+        logger.error(f"同步数字分身广场互动授权失败: {exc}")
+        return set()
+    finally:
+        avatar_db.close()
+
+    from app.core.redis_client import delete_session, set_session
+
+    result = await db.execute(
+        select(VirtualUser).where(VirtualUser.account.like(f"{AVATAR_ACCOUNT_PREFIX}%"))
+    )
+    existing_users = {delegated_avatar_id(user): user for user in result.scalars().all()}
+    authorized_ids = {item["avatar_id"] for item in authorized}
+
+    for avatar_id, user in existing_users.items():
+        if avatar_id not in authorized_ids:
+            user.is_enabled = 0
+            user.status = 0
+            user.session_token = None
+            user.session_expires_at = None
+            await delete_session(user.id)
+
+    for item in authorized:
+        avatar_id = item["avatar_id"]
+        user = existing_users.get(avatar_id)
+        if user is None:
+            user = VirtualUser(
+                nickname=item["avatar_name"],
+                account=f"{AVATAR_ACCOUNT_PREFIX}{avatar_id}",
+                password_enc="",
+                status=2,
+                is_enabled=1,
+                platform_uid=item["platform_uid"],
+                remark="用户授权的数字分身广场互动身份",
+            )
+            db.add(user)
+            await db.flush()
+
+        expires_at = datetime.now() + timedelta(days=1)
+        user.nickname = item["avatar_name"]
+        user.real_name = item["avatar_name"]
+        user.avatar_url = item["avatar_url"]
+        user.platform_uid = item["platform_uid"]
+        user.session_token = item["token"]
+        user.session_expires_at = expires_at
+        user.last_login_at = datetime.now()
+        user.status = 2
+        user.is_enabled = 1
+
+        config = item["config"]
+        personality_result = await db.execute(
+            select(UserPersonality).where(UserPersonality.user_id == user.id)
+        )
+        personality = personality_result.scalar_one_or_none()
+        word_min, word_max = _word_count_range(config)
+        prompt_parts = [item["avatar_description"], str(config.get("systemPrompt") or "")]
+        style_prompt = "\n".join(part.strip() for part in prompt_parts if part and part.strip())
+        if personality is None:
+            personality = UserPersonality(user_id=user.id)
+            db.add(personality)
+        personality.language_style = str(config.get("replyStyle") or "professional")
+        personality.personality_desc = item["avatar_description"]
+        personality.comment_style_prompt = style_prompt
+        personality.word_count_min = word_min
+        personality.word_count_max = word_max
+
+        await set_session(user.id, {
+            "token": item["token"],
+            "session_id": f"avatar:{avatar_id}",
+            "platform_uid": item["platform_uid"],
+            "org_id": "",
+            "login_time": datetime.now().isoformat(),
+            "nickname": item["avatar_name"],
+            "real_name": item["avatar_name"],
+            "avatar": item["avatar_url"],
+            "delegated_avatar_id": avatar_id,
+        }, expire=86400)
+
+    await db.commit()
+    return authorized_ids
+
+
 class AvatarService:
 
     @staticmethod

backend/app/services/scheduler.py

@@ -23,6 +23,7 @@ class SchedulerService:
         from app.core.database import AsyncSessionLocal
         logger.info("⚡ 立即触发互动任务")
         async with AsyncSessionLocal() as session:
+            await self._sync_delegated_avatar_users(session)
             try:
                 max_concurrent = int(await self._get_config(session, "max_concurrent_users", "5"))
             except (TypeError, ValueError):
@@ -146,7 +147,9 @@ class SchedulerService:
     async def _check_sessions(self):
         """定时校验登录状态"""
         from app.services.news_service import news_service
+        from app.services.avatar_service import is_delegated_avatar_user
         async with AsyncSessionLocal() as db:
+            await self._sync_delegated_avatar_users(db)
             result = await db.execute(
                 select(VirtualUser).where(VirtualUser.status == 2, VirtualUser.is_enabled == 1)
             )
@@ -154,7 +157,7 @@ class SchedulerService:
             for user in users:
                 try:
                     valid = await news_service.check_session(db, user)
-                    if not valid:
+                    if not valid and not is_delegated_avatar_user(user):
                         logger.warning(f"用户 {user.account} 会话失效，尝试重登")
                         await news_service.login(db, user)
                 except Exception as e:
@@ -163,6 +166,7 @@ class SchedulerService:
     async def _run_interactions(self):
         """执行互动任务"""
         async with AsyncSessionLocal() as db:
+            await self._sync_delegated_avatar_users(db)
             # 检查调度器开关
             enabled = await self._get_config(db, "scheduler_enabled", "true")
             if enabled != "true":
@@ -184,8 +188,11 @@ class SchedulerService:
                 logger.debug(f"[调度] 当前北京时间 {now_time} 不在互动时段 {start_str}-{end_str}")
                 return
 
-            # 获取最小互动间隔（秒）
-            min_interval = int(await self._get_config(db, "interact_min_interval", "300"))
+            # 获取互动间隔范围（秒），与调度设置页面字段保持一致
+            min_interval = await self._get_int_config(db, "interact_interval_min", 300)
+            max_interval = await self._get_int_config(db, "interact_interval_max", min_interval)
+            min_interval = max(0, min_interval)
+            max_interval = max(min_interval, max_interval)
 
             # 获取最大并发
             max_concurrent = int(await self._get_config(db, "max_concurrent_users", "5"))
@@ -204,7 +211,7 @@ class SchedulerService:
                 await self._try_login_users(db)
                 return
 
-            # 检查互动间隔：过滤掉最近 min_interval 秒内已互动的用户
+            # 每个用户在其最小/最大间隔内取得稳定随机值，直到下次互动后再变化
             now_dt = datetime.now()
             eligible = []
             for u in all_users:
@@ -212,11 +219,17 @@ class SchedulerService:
                     eligible.append(u)
                 else:
                     elapsed = (now_dt - u.last_interact_at).total_seconds()
-                    if elapsed >= min_interval:
+                    interval = random.Random(
+                        f"{u.id}:{u.last_interact_at.isoformat()}"
+                    ).randint(min_interval, max_interval)
+                    if elapsed >= interval:
                         eligible.append(u)
 
             if not eligible:
-                logger.debug(f"[调度] 所有 {len(all_users)} 个用户在 {min_interval}s 内已互动，跳过本次")
+                logger.debug(
+                    f"[调度] 所有 {len(all_users)} 个用户尚未达到 "
+                    f"{min_interval}-{max_interval}s 随机互动间隔，跳过本次"
+                )
                 return
 
             # 按最后互动时间升序排序：最久没互动的用户优先
@@ -257,10 +270,12 @@ class SchedulerService:
     async def _try_login_users(self, db):
         """尝试登录未登录的用户"""
         from app.services.news_service import news_service
+        from app.services.avatar_service import AVATAR_ACCOUNT_PREFIX
         result = await db.execute(
             select(VirtualUser).where(
                 VirtualUser.status.in_([0, 3]),
-                VirtualUser.is_enabled == 1
+                VirtualUser.is_enabled == 1,
+                ~VirtualUser.account.like(f"{AVATAR_ACCOUNT_PREFIX}%"),
             ).limit(3)
         )
         users = result.scalars().all()
@@ -275,6 +290,11 @@ class SchedulerService:
         """执行单用户互动 - 基于真实接口"""
         from app.services.news_service import news_service
         from app.services.ai_service import ai_service
+        from app.services.avatar_service import (
+            delegated_avatar_id,
+            get_square_interaction_permissions,
+            is_delegated_avatar_user,
+        )
 
         async with AsyncSessionLocal() as db:
             try:
@@ -289,6 +309,23 @@ class SchedulerService:
                         "interactions": [],
                     }
 
+                allowed_actions = {"like", "collect", "comment", "reply", "forward"}
+                if is_delegated_avatar_user(user):
+                    allowed_actions = set(
+                        get_square_interaction_permissions(delegated_avatar_id(user))
+                    )
+                    if not allowed_actions:
+                        user.status = 0
+                        user.is_enabled = 0
+                        await db.commit()
+                        return {
+                            "user_id": user.id,
+                            "account": user.account,
+                            "status": "skipped",
+                            "reason": "avatar_interaction_not_authorized",
+                            "interactions": [],
+                        }
+
                 # 检查今日评论限额
                 can_comment = True
                 if user.today_comment_count >= user.daily_comment_limit:
@@ -398,14 +435,53 @@ class SchedulerService:
                 interactions_done = []
                 action_failures = []
 
-                # ① 先记录阅读（每次必做，模拟真实用户打开文章）
+                done_on_this = today_done.get(news_id, set())
+                wants = {
+                    "like": (
+                        "like" in allowed_actions
+                        and "like" not in done_on_this
+                        and random.random() < like_prob
+                    ),
+                    "collect": (
+                        "collect" in allowed_actions
+                        and "collect" not in done_on_this
+                        and random.random() < collect_prob
+                    ),
+                    "forward": (
+                        "forward" in allowed_actions
+                        and "forward" not in done_on_this
+                        and random.random() < forward_prob
+                    ),
+                    "reply": (
+                        "reply" in allowed_actions
+                        and can_comment
+                        and personality is not None
+                        and random.random() < reply_prob
+                    ),
+                    "comment": (
+                        "comment" in allowed_actions
+                        and can_comment
+                        and personality is not None
+                        and not already_commented_this
+                        and random.random() < comment_prob
+                    ),
+                }
+                if not any(wants.values()):
+                    return {
+                        "user_id": user.id,
+                        "account": user.account,
+                        "status": "skipped",
+                        "reason": "no_actions_triggered",
+                        "interactions": [],
+                        "article_id": news_id,
+                        "article_title": news_title,
+                    }
+
+                # 只有动作命中调度概率后才打开文章
                 await news_service.read_news(db, user, news_id)
 
-                # 今日已对此文章做过的互动类型
-                done_on_this = today_done.get(news_id, set())
-
                 # ② 点赞（每篇文章每用户每天只点赞一次）
-                if "like" not in done_on_this and random.random() < like_prob:
+                if wants["like"]:
                     success, err = await news_service.like_news(db, user, news_id, org_id=article_org_id, to_user_id=news_author, title=news_title)
                     await self._save_record(db, user, news_id, news_title, "like", None, 0, success, err)
                     if success:
@@ -415,16 +491,17 @@ class SchedulerService:
                         action_failures.append({"type": "like", "error": err})
 
                 # ③ 收藏（每篇文章每用户每天只收藏一次）
-                if "collect" not in done_on_this and random.random() < collect_prob:
+                if wants["collect"]:
                     success, err = await news_service.collect_news(db, user, news_id, org_id=article_org_id, to_user_id=news_author, title=news_title)
                     await self._save_record(db, user, news_id, news_title, "collect", None, 0, success, err)
                     if success:
                         interactions_done.append("collect")
+                        await self._incr_total(db, user_id)
                     else:
                         action_failures.append({"type": "collect", "error": err})
 
                 # ④ 转发（每篇文章每用户每天只转发一次）
-                if "forward" not in done_on_this and random.random() < forward_prob:
+                if wants["forward"]:
                     success, err = await news_service.forward_news(db, user, news_id)
                     await self._save_record(db, user, news_id, news_title, "forward", None, 0, success, err)
                     if success:
@@ -438,7 +515,7 @@ class SchedulerService:
                     style_prompt = personality.comment_style_prompt or ""
                     safe_word_max = min(personality.word_count_max, 80)
 
-                    if random.random() < reply_prob:
+                    if wants["reply"]:
                         reply_actions, reply_failures = await self._run_reply_interaction_chain(
                             db=db,
                             starter=user,
@@ -455,7 +532,7 @@ class SchedulerService:
                         action_failures.extend(reply_failures)
 
                     # 每篇文章每个用户每天只发一条顶层评论；回复不再要求先评论
-                    if not already_commented_this and random.random() < comment_prob:
+                    if wants["comment"]:
                         comment_text, tokens = await ai_service.generate_comment(
                             db, news_title, news_content,
                             style_prompt, personality.word_count_min, safe_word_max
@@ -679,6 +756,7 @@ class SchedulerService:
 
         async with AsyncSessionLocal() as db:
             try:
+                await self._sync_delegated_avatar_users(db)
                 now = datetime.now()
                 await db.execute(
                     update(PendingReplyTask)
@@ -706,6 +784,12 @@ class SchedulerService:
                 logger.error(f"待发送回复队列处理异常: {e}")
 
     async def _process_pending_reply_task(self, db, task: PendingReplyTask, news_service, ai_service):
+        from app.services.avatar_service import (
+            delegated_avatar_id,
+            get_square_interaction_permissions,
+            is_delegated_avatar_user,
+        )
+
         task.status = 1
         task.locked_at = datetime.now()
         task.attempts = (task.attempts or 0) + 1
@@ -716,6 +800,13 @@ class SchedulerService:
             task.status = 3
             task.last_error = "用户未登录或已禁用"
             return
+        if (
+            is_delegated_avatar_user(actor)
+            and "reply" not in get_square_interaction_permissions(delegated_avatar_id(actor))
+        ):
+            task.status = 3
+            task.last_error = "数字分身广场互动授权已撤销"
+            return
 
         reply_result = await self._post_contextual_reply(
             db=db,
@@ -858,6 +949,16 @@ class SchedulerService:
         except (TypeError, ValueError):
             return default
 
+    async def _sync_delegated_avatar_users(self, db):
+        from app.services.avatar_service import sync_square_interaction_users
+
+        try:
+            return await sync_square_interaction_users(db)
+        except Exception as exc:
+            await db.rollback()
+            logger.error(f"数字分身广场互动身份同步异常: {exc}")
+            return set()
+
     async def _incr_total(self, db, user_id: int):
         await db.execute(
             update(VirtualUser).where(VirtualUser.id == user_id).values(

backend/tests/test_avatar_square_authorization.py

@@ -0,0 +1,126 @@
+import json
+import os
+import sqlite3
+import tempfile
+import unittest
+from types import SimpleNamespace
+from unittest.mock import patch
+
+from app.services import avatar_service
+
+
+class AvatarSquareAuthorizationTests(unittest.TestCase):
+    def setUp(self):
+        fd, self.db_path = tempfile.mkstemp(suffix=".db")
+        os.close(fd)
+        connection = sqlite3.connect(self.db_path)
+        connection.executescript("""
+            CREATE TABLE users (
+                huihui_user_id TEXT,
+                nickname TEXT,
+                avatar_url TEXT,
+                huihui_token TEXT
+            );
+            CREATE TABLE avatars (
+                id TEXT,
+                owner_id TEXT,
+                name TEXT,
+                display_name TEXT,
+                description TEXT,
+                photo_url TEXT,
+                config TEXT,
+                status TEXT
+            );
+        """)
+        connection.execute(
+            "INSERT INTO users VALUES (?, ?, ?, ?)",
+            ("huihui-7", "主人", "/owner.jpg", "huihui-token"),
+        )
+        connection.commit()
+        connection.close()
+        avatar_service._engine = None
+        avatar_service._SessionLocal = None
+        self.path_patch = patch.object(avatar_service.settings, "AVATAR_DB_PATH", self.db_path)
+        self.path_patch.start()
+
+    def tearDown(self):
+        self.path_patch.stop()
+        if avatar_service._engine is not None:
+            avatar_service._engine.dispose()
+        avatar_service._engine = None
+        avatar_service._SessionLocal = None
+        os.unlink(self.db_path)
+
+    def _insert_avatar(self, permissions, *, status="active", token=None):
+        connection = sqlite3.connect(self.db_path)
+        connection.execute(
+            "INSERT INTO avatars VALUES (?, ?, ?, ?, ?, ?, ?, ?)",
+            (
+                "avatar-7",
+                "huihui-7",
+                "avatar",
+                "小会",
+                "语气友好，表达简洁",
+                "/avatar.jpg",
+                json.dumps({
+                    "authorizationPermissions": permissions,
+                    "replyStyle": "warm",
+                    "responseLength": "short",
+                }),
+                status,
+            ),
+        )
+        if token is not None:
+            connection.execute(
+                "UPDATE users SET huihui_token = ? WHERE huihui_user_id = ?",
+                (token, "huihui-7"),
+            )
+        connection.commit()
+        connection.close()
+
+    def test_interact_permission_exposes_only_requested_square_actions(self):
+        self._insert_avatar(["chat", "interact"])
+
+        permissions = avatar_service.get_square_interaction_permissions("avatar-7")
+
+        self.assertEqual(
+            permissions,
+            frozenset({"like", "collect", "comment", "reply"}),
+        )
+        self.assertNotIn("forward", permissions)
+
+    def test_missing_permission_inactive_avatar_or_missing_token_denies_execution(self):
+        scenarios = [
+            (["chat"], "active", "huihui-token"),
+            (["interact"], "inactive", "huihui-token"),
+            (["interact"], "active", ""),
+        ]
+        for permissions, status, token in scenarios:
+            with self.subTest(permissions=permissions, status=status, token=token):
+                connection = sqlite3.connect(self.db_path)
+                connection.execute("DELETE FROM avatars")
+                connection.commit()
+                connection.close()
+                self._insert_avatar(permissions, status=status, token=token)
+                self.assertEqual(
+                    avatar_service.get_square_interaction_permissions("avatar-7"),
+                    frozenset(),
+                )
+
+    def test_delegated_avatar_identity_is_recognized_without_matching_normal_users(self):
+        delegated = SimpleNamespace(account="__avatar__:avatar-7")
+        normal = SimpleNamespace(account="13800000000")
+
+        self.assertTrue(avatar_service.is_delegated_avatar_user(delegated))
+        self.assertEqual(avatar_service.delegated_avatar_id(delegated), "avatar-7")
+        self.assertFalse(avatar_service.is_delegated_avatar_user(normal))
+        self.assertEqual(avatar_service.delegated_avatar_id(normal), "")
+
+    def test_response_length_maps_to_scheduler_comment_limits(self):
+        self.assertEqual(avatar_service._word_count_range({"responseLength": "short"}), (10, 35))
+        self.assertEqual(avatar_service._word_count_range({"responseLength": "long"}), (30, 80))
+        self.assertEqual(avatar_service._word_count_range({"responseLength": "unknown"}), (20, 60))
+
+
+if __name__ == "__main__":
+    unittest.main()

digital-avatar-app/backend/routers/knowledge.py

@@ -1,4 +1,7 @@
 import os
+import json
+import shutil
+import time
 import uuid
 
 from fastapi import APIRouter, UploadFile, File, Depends, Header, HTTPException
@@ -19,6 +22,9 @@ os.makedirs(UPLOAD_DIR, exist_ok=True)
 ALLOWED_EXT = {".md", ".txt", ".pdf", ".doc", ".docx", ".xlsx"}
 MAX_UPLOAD_BYTES = 50 * 1024 * 1024
 UPLOAD_CHUNK_BYTES = 1024 * 1024
+MULTIPART_CHUNK_BYTES = 5 * 1024 * 1024
+MULTIPART_ROOT = ".multipart"
+MULTIPART_TTL_SECONDS = 24 * 60 * 60
 
 
 class QAIn(BaseModel):
@@ -31,6 +37,74 @@ class EnabledIn(BaseModel):
     enabled: bool = True
 
 
+class MultipartUploadIn(BaseModel):
+    filename: str
+    fileSize: int
+    totalChunks: int
+
+
+def _validate_document(filename: str, file_size: int):
+    ext = os.path.splitext(filename or "")[1].lower()
+    if ext not in ALLOWED_EXT:
+        return None, f"不支持的文件类型：{ext or '空'}，仅支持 md/txt/pdf/doc/docx/xlsx"
+    if file_size <= 0:
+        return None, "文件内容不能为空"
+    if file_size > MAX_UPLOAD_BYTES:
+        return None, "文件不能超过 50MB"
+    return ext, ""
+
+
+def _multipart_dir(avatar_id: str, upload_id: str) -> str:
+    safe_avatar_id = os.path.basename(avatar_id)
+    safe_upload_id = os.path.basename(upload_id)
+    if (
+        safe_avatar_id != avatar_id
+        or safe_upload_id != upload_id
+        or len(upload_id) != 32
+        or any(character not in "0123456789abcdef" for character in upload_id)
+    ):
+        raise HTTPException(status_code=400, detail="上传标识无效")
+    return os.path.join(UPLOAD_DIR, MULTIPART_ROOT, safe_avatar_id, safe_upload_id)
+
+
+def _purge_stale_multipart_uploads(avatar_id: str):
+    avatar_upload_root = os.path.join(UPLOAD_DIR, MULTIPART_ROOT, os.path.basename(avatar_id))
+    if not os.path.isdir(avatar_upload_root):
+        return
+    cutoff = time.time() - MULTIPART_TTL_SECONDS
+    for entry in os.scandir(avatar_upload_root):
+        if entry.is_dir(follow_symlinks=False) and entry.stat(follow_symlinks=False).st_mtime < cutoff:
+            shutil.rmtree(entry.path, ignore_errors=True)
+
+
+def _read_multipart_metadata(avatar_id: str, upload_id: str) -> tuple[str, dict]:
+    upload_dir = _multipart_dir(avatar_id, upload_id)
+    metadata_path = os.path.join(upload_dir, "metadata.json")
+    if not os.path.isfile(metadata_path):
+        raise HTTPException(status_code=404, detail="上传任务不存在或已过期")
+    with open(metadata_path, "r", encoding="utf-8") as stream:
+        return upload_dir, json.load(stream)
+
+
+def _create_knowledge_doc(db: Session, avatar_id: str, filename: str, ext: str, file_size: int, stored: str):
+    doc = KnowledgeDoc(
+        id=uuid.uuid4().hex,
+        avatar_id=avatar_id,
+        filename=filename,
+        file_type=ext.lstrip("."),
+        file_size=file_size,
+        file_url=f"/api/files/{avatar_id}/{stored}",
+        status="parsing",
+        index_stage="queued",
+        index_progress=0,
+    )
+    db.add(doc)
+    db.commit()
+    db.refresh(doc)
+    knowledge_vectorizer.enqueue(doc.id)
+    return doc
+
+
 def _doc_payload(doc: KnowledgeDoc) -> dict:
     payload = doc.to_dict()
     stored_name = os.path.basename(doc.file_url or "")
@@ -74,9 +148,9 @@ def list_docs(avatar_id: str, authorization: str = Header(None), db: Session = D
 @router.post("/avatar/{avatar_id}/knowledge/docs")
 async def upload_doc(avatar_id: str, file: UploadFile = File(...), authorization: str = Header(None), db: Session = Depends(get_db)):
     _require_owned_avatar(db, avatar_id, authorization)
-    ext = os.path.splitext(file.filename or "")[1].lower()
-    if ext not in ALLOWED_EXT:
-        return fail(f"不支持的文件类型：{ext or '空'}，仅支持 md/txt/pdf/doc/docx/xlsx", code=400)
+    ext, validation_error = _validate_document(file.filename or "", 1)
+    if validation_error:
+        return fail(validation_error, code=400)
     avatar_dir = os.path.join(UPLOAD_DIR, avatar_id)
     os.makedirs(avatar_dir, exist_ok=True)
     stored = f"{uuid.uuid4().hex}{ext}"
@@ -94,25 +168,124 @@ async def upload_doc(avatar_id: str, file: UploadFile = File(...), authorization
         if os.path.exists(path):
             os.remove(path)
         return fail(str(exc), code=400)
-    doc = KnowledgeDoc(
-        id=uuid.uuid4().hex,
-        avatar_id=avatar_id,
-        filename=file.filename,
-        file_type=ext.lstrip("."),
-        file_size=file_size,
-        file_url=f"/api/files/{avatar_id}/{stored}",
-        status="parsing",
-        index_stage="queued",
-        index_progress=0,
+    if file_size == 0:
+        if os.path.exists(path):
+            os.remove(path)
+        return fail("文件内容不能为空", code=400)
+
+    doc = _create_knowledge_doc(db, avatar_id, file.filename or stored, ext, file_size, stored)
+    return ok(_doc_payload(doc))
+
+
+@router.post("/avatar/{avatar_id}/knowledge/uploads")
+def create_multipart_upload(
+    avatar_id: str,
+    body: MultipartUploadIn,
+    authorization: str = Header(None),
+    db: Session = Depends(get_db),
+):
+    _require_owned_avatar(db, avatar_id, authorization)
+    ext, validation_error = _validate_document(body.filename, body.fileSize)
+    if validation_error:
+        return fail(validation_error, code=400)
+    expected_chunks = (body.fileSize + MULTIPART_CHUNK_BYTES - 1) // MULTIPART_CHUNK_BYTES
+    if body.totalChunks != expected_chunks:
+        return fail("文件分片数量不正确", code=400)
+
+    _purge_stale_multipart_uploads(avatar_id)
+    upload_id = uuid.uuid4().hex
+    upload_dir = _multipart_dir(avatar_id, upload_id)
+    os.makedirs(upload_dir, exist_ok=False)
+    metadata = {
+        "filename": body.filename,
+        "fileSize": body.fileSize,
+        "totalChunks": body.totalChunks,
+        "extension": ext,
+    }
+    with open(os.path.join(upload_dir, "metadata.json"), "w", encoding="utf-8") as stream:
+        json.dump(metadata, stream, ensure_ascii=False)
+    return ok({"uploadId": upload_id, "chunkSize": MULTIPART_CHUNK_BYTES})
+
+
+@router.post("/avatar/{avatar_id}/knowledge/uploads/{upload_id}/chunks/{chunk_index}")
+async def upload_multipart_chunk(
+    avatar_id: str,
+    upload_id: str,
+    chunk_index: int,
+    file: UploadFile = File(...),
+    authorization: str = Header(None),
+    db: Session = Depends(get_db),
+):
+    _require_owned_avatar(db, avatar_id, authorization)
+    upload_dir, metadata = _read_multipart_metadata(avatar_id, upload_id)
+    total_chunks = int(metadata["totalChunks"])
+    if chunk_index < 0 or chunk_index >= total_chunks:
+        return fail("文件分片序号不正确", code=400)
+
+    expected_size = min(
+        MULTIPART_CHUNK_BYTES,
+        int(metadata["fileSize"]) - chunk_index * MULTIPART_CHUNK_BYTES,
     )
+    part_path = os.path.join(upload_dir, f"{chunk_index}.part")
+    temporary_path = f"{part_path}.uploading"
+    received = 0
+    try:
+        with open(temporary_path, "wb") as stream:
+            while chunk := await file.read(UPLOAD_CHUNK_BYTES):
+                received += len(chunk)
+                if received > expected_size:
+                    raise ValueError("文件分片大小不正确")
+                stream.write(chunk)
+        if received != expected_size:
+            raise ValueError("文件分片大小不正确")
+        os.replace(temporary_path, part_path)
+    except ValueError as exc:
+        if os.path.exists(temporary_path):
+            os.remove(temporary_path)
+        return fail(str(exc), code=400)
+    return ok({"chunkIndex": chunk_index, "uploadedBytes": received})
 
-    # Persist and acknowledge the upload first. Extraction and embeddings may take
-    # minutes for a PDF and must never consume the browser request timeout.
-    db.add(doc)
-    db.commit()
-    db.refresh(doc)
-    knowledge_vectorizer.enqueue(doc.id)
 
+@router.post("/avatar/{avatar_id}/knowledge/uploads/{upload_id}/complete")
+def complete_multipart_upload(
+    avatar_id: str,
+    upload_id: str,
+    authorization: str = Header(None),
+    db: Session = Depends(get_db),
+):
+    _require_owned_avatar(db, avatar_id, authorization)
+    upload_dir, metadata = _read_multipart_metadata(avatar_id, upload_id)
+    total_chunks = int(metadata["totalChunks"])
+    part_paths = [os.path.join(upload_dir, f"{index}.part") for index in range(total_chunks)]
+    if not all(os.path.isfile(path) for path in part_paths):
+        return fail("文件分片尚未上传完整", code=400)
+    if sum(os.path.getsize(path) for path in part_paths) != int(metadata["fileSize"]):
+        return fail("文件分片总大小不正确", code=400)
+
+    avatar_dir = os.path.join(UPLOAD_DIR, avatar_id)
+    os.makedirs(avatar_dir, exist_ok=True)
+    stored = f"{uuid.uuid4().hex}{metadata['extension']}"
+    final_path = os.path.join(avatar_dir, stored)
+    temporary_path = f"{final_path}.assembling"
+    try:
+        with open(temporary_path, "wb") as output:
+            for part_path in part_paths:
+                with open(part_path, "rb") as source:
+                    shutil.copyfileobj(source, output, UPLOAD_CHUNK_BYTES)
+        os.replace(temporary_path, final_path)
+        doc = _create_knowledge_doc(
+            db,
+            avatar_id,
+            metadata["filename"],
+            metadata["extension"],
+            int(metadata["fileSize"]),
+            stored,
+        )
+    except Exception:
+        if os.path.exists(temporary_path):
+            os.remove(temporary_path)
+        raise
+    shutil.rmtree(upload_dir, ignore_errors=True)
     return ok(_doc_payload(doc))
 
 

digital-avatar-app/backend/tests/test_knowledge_storage.py

@@ -87,6 +87,84 @@ def test_upload_rejects_oversize_file_before_queuing_indexing(
     assert not list((tmp_path / context["avatar"].id).glob("*"))
 
 
+def test_multipart_upload_reassembles_file_before_queuing_indexing(
+    tmp_path: Path,
+    authorization_context,
+):
+    context = authorization_context
+    avatar_id = context["avatar"].id
+    content = b"0123456789"
+    with (
+        patch("routers.knowledge.UPLOAD_DIR", str(tmp_path)),
+        patch("routers.knowledge.MULTIPART_CHUNK_BYTES", 4),
+        patch("routers.knowledge.knowledge_vectorizer.enqueue") as enqueue,
+    ):
+        created = client.post(
+            f"/api/avatar/{avatar_id}/knowledge/uploads",
+            headers=context["owner_headers"],
+            json={"filename": "large.pdf", "fileSize": len(content), "totalChunks": 3},
+        ).json()["data"]
+
+        for index, chunk in enumerate((content[:4], content[4:8], content[8:])):
+            response = client.post(
+                f"/api/avatar/{avatar_id}/knowledge/uploads/{created['uploadId']}/chunks/{index}",
+                headers=context["owner_headers"],
+                files={"file": (f"chunk-{index}", chunk, "application/octet-stream")},
+            )
+            assert response.json()["code"] == 200
+
+        completed = client.post(
+            f"/api/avatar/{avatar_id}/knowledge/uploads/{created['uploadId']}/complete",
+            headers=context["owner_headers"],
+        ).json()["data"]
+
+    assert completed["status"] == "parsing"
+    assert completed["fileSize"] == len(content)
+    enqueue.assert_called_once_with(completed["id"])
+    stored_path = tmp_path / avatar_id / Path(completed["fileUrl"]).name
+    assert stored_path.read_bytes() == content
+    assert not (tmp_path / ".multipart" / avatar_id / created["uploadId"]).exists()
+
+    db = SessionLocal()
+    try:
+        stored = db.query(KnowledgeDoc).filter(KnowledgeDoc.id == completed["id"]).one()
+        db.delete(stored)
+        db.commit()
+    finally:
+        db.close()
+
+
+def test_multipart_upload_rejects_incomplete_parts(
+    tmp_path: Path,
+    authorization_context,
+):
+    context = authorization_context
+    avatar_id = context["avatar"].id
+    with (
+        patch("routers.knowledge.UPLOAD_DIR", str(tmp_path)),
+        patch("routers.knowledge.MULTIPART_CHUNK_BYTES", 4),
+        patch("routers.knowledge.knowledge_vectorizer.enqueue") as enqueue,
+    ):
+        created = client.post(
+            f"/api/avatar/{avatar_id}/knowledge/uploads",
+            headers=context["owner_headers"],
+            json={"filename": "large.pdf", "fileSize": 6, "totalChunks": 2},
+        ).json()["data"]
+        client.post(
+            f"/api/avatar/{avatar_id}/knowledge/uploads/{created['uploadId']}/chunks/0",
+            headers=context["owner_headers"],
+            files={"file": ("chunk-0", b"0123", "application/octet-stream")},
+        )
+        response = client.post(
+            f"/api/avatar/{avatar_id}/knowledge/uploads/{created['uploadId']}/complete",
+            headers=context["owner_headers"],
+        )
+
+    assert response.json()["code"] == 400
+    assert response.json()["message"] == "文件分片尚未上传完整"
+    enqueue.assert_not_called()
+
+
 def test_background_vectorizer_commits_ready_document_and_chunks_together(
     tmp_path: Path,
     authorization_context,

digital-avatar-app/src/api/index.ts

@@ -333,12 +333,76 @@ export interface SearchResult {
 export const getKnowledgeDocs = (avatarId: string) =>
   request.get<KnowledgeDoc[]>(`/avatar/${avatarId}/knowledge/docs`)
 
-// 上传文档（支持 md/txt/pdf/doc/docx/xlsx）
-export const uploadKnowledgeDoc = (
+const KNOWLEDGE_UPLOAD_CHUNK_SIZE = 5 * 1024 * 1024
+
+const uploadKnowledgeChunk = async (
+  avatarId: string,
+  uploadId: string,
+  chunkIndex: number,
+  chunk: Blob,
+  onProgress?: (loaded: number) => void
+) => {
+  const form = new FormData()
+  form.append('file', chunk, `chunk-${chunkIndex}`)
+  let reportedLoaded = 0
+  for (let attempt = 1; attempt <= 3; attempt += 1) {
+    try {
+      await request.post(
+        `/avatar/${avatarId}/knowledge/uploads/${uploadId}/chunks/${chunkIndex}`,
+        form,
+        {
+          headers: { 'Content-Type': 'multipart/form-data' },
+          timeout: 2 * 60 * 1000,
+          onUploadProgress: (event) => {
+            reportedLoaded = Math.max(reportedLoaded, Math.min(event.loaded, chunk.size))
+            onProgress?.(reportedLoaded)
+          }
+        }
+      )
+      return
+    } catch (error: any) {
+      const status = Number(error?.response?.status || 0)
+      const retryable = !status || status === 408 || status === 429 || status >= 500
+      if (!retryable || attempt === 3) throw error
+      await new Promise((resolve) => window.setTimeout(resolve, attempt * 800))
+    }
+  }
+}
+
+// 大文件拆成 5MB 分片，避免生产代理的请求体限制拦截整个文件。
+export const uploadKnowledgeDoc = async (
   avatarId: string,
   file: File,
   onUploadProgress?: (loaded: number, total: number) => void
 ) => {
+  if (file.size > KNOWLEDGE_UPLOAD_CHUNK_SIZE) {
+    const totalChunks = Math.ceil(file.size / KNOWLEDGE_UPLOAD_CHUNK_SIZE)
+    const upload: any = await request.post(`/avatar/${avatarId}/knowledge/uploads`, {
+      filename: file.name,
+      fileSize: file.size,
+      totalChunks
+    })
+    let uploadedBytes = 0
+    for (let index = 0; index < totalChunks; index += 1) {
+      const start = index * KNOWLEDGE_UPLOAD_CHUNK_SIZE
+      const chunk = file.slice(start, Math.min(start + KNOWLEDGE_UPLOAD_CHUNK_SIZE, file.size))
+      await uploadKnowledgeChunk(
+        avatarId,
+        upload.uploadId,
+        index,
+        chunk,
+        (chunkLoaded) => onUploadProgress?.(uploadedBytes + chunkLoaded, file.size)
+      )
+      uploadedBytes += chunk.size
+      onUploadProgress?.(uploadedBytes, file.size)
+    }
+    return request.post<KnowledgeDoc>(
+      `/avatar/${avatarId}/knowledge/uploads/${upload.uploadId}/complete`,
+      undefined,
+      { timeout: 2 * 60 * 1000 }
+    )
+  }
+
   const form = new FormData()
   form.append('file', file)
   return request.post<KnowledgeDoc>(`/avatar/${avatarId}/knowledge/docs`, form, {

digital-avatar-app/src/views/AuthorizationManage.vue

@@ -192,7 +192,7 @@ const permissionItems: Array<{
   {
     key: 'interact',
     title: '广场互动操作',
-    description: '点赞、收藏、评论、回复等操作',
+    description: '允许分身代表你点赞、收藏、评论和回复；时段、间隔与触发概率由广场调度设置统一控制',
     tone: 'pink',
   },
   {

digital-avatar-app/src/views/KnowledgeManage.vue

@@ -32,7 +32,7 @@
         </div>
 
         <div v-if="displayDocs.length" class="mobile-card-list">
-          <article v-for="doc in displayDocs" :key="doc.id" class="knowledge-card">
+          <article v-for="doc in displayDocs" :key="doc.id" class="knowledge-card document-card">
             <div class="card-icon">{{ fileEmoji(doc.fileType) }}</div>
             <div class="card-content">
               <div class="card-title-row">
@@ -46,9 +46,14 @@
               </div>
             </div>
             <div class="card-actions">
-              <button v-if="documentState(doc).tone === 'failed'" class="card-retry" @click="retryDoc(doc.id)">重新索引</button>
               <button v-if="!doc.localUploading" class="card-delete" @click="removeDoc(doc.id)">{{ doc.localOnly ? '移除' : '删除' }}</button>
             </div>
+            <div v-if="canRetryDoc(doc)" class="card-retry-area">
+              <span v-if="retryErrors[doc.id]" class="card-retry-error">{{ retryErrors[doc.id] }}</span>
+              <button class="card-retry" :disabled="retryingDocs[doc.id]" @click="retryDoc(doc)">
+                {{ retryingDocs[doc.id] ? '重新索引中…' : '重新索引' }}
+              </button>
+            </div>
           </article>
         </div>
         <div v-else class="card-empty">📂 暂无文档，先上传一个知识文件</div>
@@ -115,6 +120,8 @@ const uploading = computed(() => pendingUploads.value.some((doc) => doc.localUpl
 const uploadError = ref('')
 const dragOver = ref(false)
 const fileInput = ref<HTMLInputElement | null>(null)
+const retryingDocs = ref<Record<string, boolean>>({})
+const retryErrors = ref<Record<string, string>>({})
 let documentPollingTimer: ReturnType<typeof setInterval> | undefined
 
 const query = ref('')
@@ -250,14 +257,24 @@ const uploadOne = async (file: File, ext: string) => {
   }
 }
 
-const retryDoc = async (id: string) => {
-  if (!avatarId.value) return
-  uploadError.value = ''
+const canRetryDoc = (doc: any) =>
+  !doc.localOnly && doc.filePresent !== false && documentState(doc).tone === 'failed'
+
+const retryDoc = async (doc: any) => {
+  if (!avatarId.value || !canRetryDoc(doc) || retryingDocs.value[doc.id]) return
+  retryingDocs.value = { ...retryingDocs.value, [doc.id]: true }
+  retryErrors.value = { ...retryErrors.value, [doc.id]: '' }
   try {
-    await retryKnowledgeDoc(avatarId.value, id)
-    await loadDocs()
+    const updated: any = await retryKnowledgeDoc(avatarId.value, doc.id)
+    Object.assign(doc, updated)
+    startDocumentPolling()
   } catch (e: any) {
-    uploadError.value = e?.message || '重新索引失败'
+    retryErrors.value = {
+      ...retryErrors.value,
+      [doc.id]: e?.response?.data?.message || e?.response?.data?.detail || e?.message || '重新索引失败'
+    }
+  } finally {
+    retryingDocs.value = { ...retryingDocs.value, [doc.id]: false }
   }
 }
 
@@ -378,6 +395,7 @@ onUnmounted(stopDocumentPolling)
 .panel-heading p { margin: -5px 0 0; color: #9398AE; font-size: 12px; }
 .mobile-card-list { display: grid; grid-template-columns: minmax(0, 1fr); width: 100%; min-width: 0; gap: 10px; }
 .knowledge-card { display: flex; align-items: center; width: 100%; min-width: 0; box-sizing: border-box; gap: 11px; padding: 14px; background: #fff; border: 1px solid #F1E1D3; border-radius: 16px; box-shadow: 0 5px 16px rgba(112, 62, 22, .04); }
+.document-card { display: grid; grid-template-columns: 42px minmax(0, 1fr) auto; align-items: center; }
 .card-icon { flex: 0 0 auto; width: 42px; height: 42px; display: grid; place-items: center; border-radius: 13px; background: #FFF3E6; font-size: 22px; }
 .card-content { min-width: 0; flex: 1; overflow: hidden; }
 .card-title-row { display: flex; align-items: center; gap: 8px; min-width: 0; }
@@ -388,10 +406,13 @@ onUnmounted(stopDocumentPolling)
 .card-meta, .card-detail { margin: 5px 0 0; color: #9398AE; font-size: 11px; line-height: 1.4; }.card-detail { color: #8B6B58; }
 .progress-track { width: 100%; height: 4px; margin-top: 8px; overflow: hidden; border-radius: 999px; background: #FDE7D1; }
 .progress-fill { display: block; height: 100%; border-radius: inherit; background: linear-gradient(90deg, #FB923C, #F97316); transition: width .25s ease; }
-.card-actions { flex: 0 0 auto; display: flex; flex-direction: column; align-items: stretch; gap: 6px; }
+.card-actions { flex: 0 0 auto; display: flex; align-items: center; }
 .card-delete, .card-retry { align-self: center; border: 0; border-radius: 8px; padding: 7px 9px; font-size: 12px; cursor: pointer; white-space: nowrap; }
 .card-delete { color: #EF4444; background: #FEF2F2; }
 .card-retry { color: #C15F18; background: #FFF3E6; }
+.card-retry:disabled { cursor: wait; opacity: .65; }
+.card-retry-area { grid-column: 1 / -1; display: flex; align-items: center; justify-content: flex-end; gap: 10px; min-width: 0; }
+.card-retry-error { min-width: 0; overflow: hidden; color: #DC2626; font-size: 11px; line-height: 1.35; text-overflow: ellipsis; white-space: nowrap; }
 .card-empty { padding: 42px 16px; border: 1px dashed #F1D9C3; border-radius: 16px; color: #9398AE; background: #fff; font-size: 14px; text-align: center; }
 .qa-card { align-items: stretch; text-align: left; }.qa-card.qa-disabled { opacity: .58; }
 .qa-card .card-content,
@@ -617,8 +638,11 @@ onUnmounted(stopDocumentPolling)
 @media (max-width: 520px) {
   .knowledge-panel { padding: 0 12px; }
   .knowledge-card { display: grid; grid-template-columns: 42px minmax(0, 1fr); align-items: start; gap: 10px; padding: 13px; }
+  .document-card { grid-template-columns: 42px minmax(0, 1fr) auto; }
   .card-content { grid-column: 2; }
-  .card-delete { grid-column: 2; justify-self: end; margin-top: -2px; }
+  .card-actions { grid-column: 3; grid-row: 1; }
+  .card-delete { justify-self: end; margin-top: -2px; }
+  .card-retry-area { grid-column: 1 / -1; }
   .qa-card { display: block; }
   .qa-card .card-content { width: 100%; grid-column: 1; }
   .card-title-row { align-items: flex-start; flex-wrap: wrap; gap: 5px 7px; }