Files
huihuiSquare/digital-avatar-app/backend/services/takeover_service.py
T

614 lines
22 KiB
Python

"""Restart-safe automatic replies over Huihui's self-hosted BOXIM."""
import asyncio
import hashlib
import logging
import re
import secrets
import time
from datetime import datetime, timedelta
from typing import Callable
from sqlalchemy.orm import Session
from models import (
Avatar,
TakeoverCursor,
TakeoverMessage,
TakeoverReplyTask,
User,
)
from services.boxim_client import BoxIMClient, BoxIMError
logger = logging.getLogger(__name__)
ACTIVE_TASK_STATUSES = ("pending", "generating", "ready", "sending")
GENERATABLE_TASK_STATUSES = ("pending",)
MAX_PROMPT_LENGTH = 4000
MAX_STALE_SECONDS = 120
STUCK_LOCK_SECONDS = 90
TAKEOVER_PERMISSION = "takeover"
def _utcnow() -> datetime:
return datetime.utcnow()
def _takeover_enabled(avatar: Avatar | None) -> bool:
if not avatar or avatar.status != "active":
return False
permissions = (avatar.config or {}).get("authorizationPermissions", [])
return isinstance(permissions, list) and TAKEOVER_PERMISSION in permissions
def _boxim_time(value, fallback: datetime) -> datetime:
try:
timestamp = float(value)
if timestamp > 10_000_000_000:
timestamp /= 1000
return datetime.utcfromtimestamp(timestamp)
except (TypeError, ValueError, OSError, OverflowError):
return fallback
def _numeric_id(value) -> int:
try:
return int(value)
except (TypeError, ValueError):
return 0
def _plain_text_reply(value: str) -> str:
"""BOXIM is plain text, so remove Markdown markers without damaging paragraphs."""
text = (value or "").replace("\r\n", "\n").replace("\r", "\n")
text = re.sub(r"```(?:\w+)?\n?(.*?)```", r"\1", text, flags=re.S)
text = re.sub(r"\*\*(.*?)\*\*|__(.*?)__", lambda m: m.group(1) or m.group(2), text)
text = re.sub(r"(?<!\*)\*([^*\n]+)\*(?!\*)", r"\1", text)
text = re.sub(r"`([^`]+)`", r"\1", text)
text = re.sub(r"^\s{0,3}#{1,6}\s*", "", text, flags=re.M)
lines = [line.strip() for line in text.split("\n")]
return "\n".join(line for line in lines if line).strip()
class TakeoverService:
"""Poll BOXIM, prepare replies during the grace period, then send at +3s."""
def __init__(
self,
session_factory: Callable[[], Session],
boxim_client: BoxIMClient,
*,
reply_delay_seconds: int = 3,
now: Callable[[], datetime] = _utcnow,
):
self.session_factory = session_factory
self.boxim = boxim_client
self.reply_delay_seconds = reply_delay_seconds
self.now = now
self._sessions: dict[str, dict] = {}
self._run_lock = asyncio.Lock()
async def poll_and_process_messages(self):
"""Run one complete cycle; polling always happens before reply dispatch."""
if self._run_lock.locked():
return
async with self._run_lock:
self._recover_stuck_tasks()
avatar_ids = self._enabled_avatar_ids()
self._cancel_disabled_tasks(set(avatar_ids))
for avatar_id in avatar_ids:
await self._sync_avatar(avatar_id)
generated = await self._prepare_replies()
if generated:
# Catch a human reply sent while the model was preparing its answer.
for avatar_id in avatar_ids:
await self._sync_avatar(avatar_id)
await self._dispatch_ready_replies()
def _enabled_avatar_ids(self) -> list[str]:
db = self.session_factory()
try:
return [
avatar.id
for avatar in db.query(Avatar).filter(Avatar.status == "active").all()
if _takeover_enabled(avatar)
]
finally:
db.close()
def _cancel_disabled_tasks(self, enabled_avatar_ids: set[str]):
db = self.session_factory()
try:
tasks = (
db.query(TakeoverReplyTask)
.filter(TakeoverReplyTask.status.in_(ACTIVE_TASK_STATUSES))
.all()
)
changed = False
for task in tasks:
if task.avatar_id not in enabled_avatar_ids:
task.status = "cancelled"
task.cancel_reason = "takeover_disabled"
task.locked_at = None
changed = True
if changed:
db.commit()
finally:
db.close()
def _recover_stuck_tasks(self):
db = self.session_factory()
try:
threshold = self.now() - timedelta(seconds=STUCK_LOCK_SECONDS)
tasks = (
db.query(TakeoverReplyTask)
.filter(
TakeoverReplyTask.status.in_(("generating", "sending")),
TakeoverReplyTask.locked_at.isnot(None),
TakeoverReplyTask.locked_at < threshold,
)
.all()
)
for task in tasks:
task.status = "pending" if task.status == "generating" else "ready"
task.locked_at = None
task.last_error = "上次处理意外中断,已自动恢复"
if tasks:
db.commit()
finally:
db.close()
async def _boxim_session(self, user: User) -> dict:
token_fingerprint = hashlib.sha256((user.huihui_token or "").encode()).hexdigest()
cached = self._sessions.get(user.id)
if (
cached
and cached["expires_at"] > time.monotonic()
and cached["token_fingerprint"] == token_fingerprint
):
return cached
token_data = await self.boxim.exchange_access_token(user.huihui_token)
access_token = token_data["accessToken"]
profile = await self.boxim.get_self(access_token)
try:
expires_in = int(token_data.get("accessTokenExpiresIn") or 3600)
except (TypeError, ValueError):
expires_in = 3600
if expires_in > 86_400:
expires_in //= 1000
cache_for = max(60, min(expires_in - 60, 3600))
cached = {
"access_token": access_token,
"boxim_owner_id": str(profile["id"]),
"expires_at": time.monotonic() + cache_for,
"token_fingerprint": token_fingerprint,
}
self._sessions[user.id] = cached
return cached
def _forget_boxim_session(self, user_id: str):
self._sessions.pop(user_id, None)
def _disable_after_connection_failure(
self,
db: Session,
avatar: Avatar,
cursor: TakeoverCursor,
message: str,
):
permissions = (avatar.config or {}).get("authorizationPermissions", [])
avatar.config = {
**(avatar.config or {}),
"authorizationPermissions": [
permission
for permission in permissions
if permission != TAKEOVER_PERMISSION
],
}
cursor.last_error = message
cursor.last_polled_at = self.now()
tasks = (
db.query(TakeoverReplyTask)
.filter(
TakeoverReplyTask.avatar_id == avatar.id,
TakeoverReplyTask.status.in_(ACTIVE_TASK_STATUSES),
)
.all()
)
for task in tasks:
task.status = "cancelled"
task.cancel_reason = "connection_failed"
task.locked_at = None
async def _sync_avatar(self, avatar_id: str) -> bool:
db = self.session_factory()
try:
avatar = db.query(Avatar).filter(Avatar.id == avatar_id).first()
if not _takeover_enabled(avatar):
return False
user = db.query(User).filter(User.huihui_user_id == avatar.owner_id).first()
cursor = db.query(TakeoverCursor).filter(TakeoverCursor.avatar_id == avatar.id).first()
if not cursor:
cursor = TakeoverCursor(avatar_id=avatar.id, owner_id=avatar.owner_id)
db.add(cursor)
db.flush()
if not user or not user.huihui_token:
self._disable_after_connection_failure(
db,
avatar,
cursor,
"请重新登录会会生产账号后再开启主动接管",
)
db.commit()
return False
try:
session = await self._boxim_session(user)
owner_boxim_id = session["boxim_owner_id"]
if cursor.boxim_owner_id and cursor.boxim_owner_id != owner_boxim_id:
cursor.initialized = False
cursor.last_message_id = "0"
cursor.boxim_owner_id = owner_boxim_id
messages = await self.boxim.fetch_private_messages(
session["access_token"], cursor.last_message_id or "0"
)
except Exception as exc:
if isinstance(exc, BoxIMError) and exc.auth_error:
self._forget_boxim_session(user.id)
message = "BOXIM 授权已失效,请重新登录会会生产账号"
else:
message = f"BOXIM 暂时连接失败:{str(exc)[:160]}"
self._disable_after_connection_failure(db, avatar, cursor, message)
db.commit()
logger.warning("BOXIM sync failed for avatar %s: %s", avatar.id, exc)
return False
messages.sort(key=lambda item: (_numeric_id(item.get("id")), item.get("sendTime") or 0))
priming = not bool(cursor.initialized)
max_message_id = _numeric_id(cursor.last_message_id)
read_receipts: dict[str, int] = {}
for message in messages:
self._record_message(
db,
avatar,
cursor.boxim_owner_id,
message,
schedule_reply=not priming,
)
message_id = _numeric_id(message.get("id"))
max_message_id = max(max_message_id, message_id)
send_id = str(message.get("sendId") or "")
recv_id = str(message.get("recvId") or "")
if recv_id == cursor.boxim_owner_id and send_id and message_id:
read_receipts[send_id] = max(read_receipts.get(send_id, 0), message_id)
# BOXIM publishes this HTTP state change to connected socket clients.
# Do it before advancing the cursor so a failed receipt is retried.
for peer_id, message_id in read_receipts.items():
await self.boxim.mark_private_messages_read(
session["access_token"], peer_id, message_id
)
cursor.last_message_id = str(max_message_id)
cursor.initialized = True
cursor.last_polled_at = self.now()
cursor.last_error = ""
db.commit()
return True
except Exception:
db.rollback()
logger.exception("Failed to persist BOXIM messages for avatar %s", avatar_id)
return False
finally:
db.close()
def _record_message(
self,
db: Session,
avatar: Avatar,
boxim_owner_id: str,
message: dict,
*,
schedule_reply: bool,
):
message_id = str(message.get("id") or "").strip()
if not message_id:
return
local_id = str(message.get("localId") or "").strip() or None
if (
db.query(TakeoverMessage)
.filter(
TakeoverMessage.owner_id == avatar.owner_id,
TakeoverMessage.boxim_message_id == message_id,
)
.first()
):
return
send_id = str(message.get("sendId") or "")
recv_id = str(message.get("recvId") or "")
if send_id == boxim_owner_id:
direction, peer_id = "outgoing", recv_id
elif recv_id == boxim_owner_id:
direction, peer_id = "incoming", send_id
else:
return
if not peer_id:
return
now = self.now()
send_time = _boxim_time(message.get("sendTime"), now)
is_avatar = False
if direction == "outgoing" and local_id:
is_avatar = bool(
db.query(TakeoverReplyTask)
.filter(
TakeoverReplyTask.owner_id == avatar.owner_id,
TakeoverReplyTask.boxim_local_id == local_id,
TakeoverReplyTask.status == "sent",
)
.first()
)
event = TakeoverMessage(
avatar_id=avatar.id,
owner_id=avatar.owner_id,
boxim_message_id=message_id,
boxim_local_id=local_id,
peer_id=peer_id,
direction=direction,
message_type=int(message.get("type") or 0),
content=str(message.get("content") or ""),
is_avatar=is_avatar,
send_time=send_time,
)
db.add(event)
db.flush()
if direction == "outgoing":
if not is_avatar:
self._cancel_conversation(db, avatar.owner_id, peer_id, "owner_replied")
return
if not schedule_reply or event.message_type != 0 or not event.content.strip():
return
if (now - send_time).total_seconds() > MAX_STALE_SECONDS:
return
self._schedule_reply(db, avatar, event)
@staticmethod
def _cancel_conversation(db: Session, owner_id: str, peer_id: str, reason: str):
tasks = (
db.query(TakeoverReplyTask)
.filter(
TakeoverReplyTask.owner_id == owner_id,
TakeoverReplyTask.peer_id == peer_id,
TakeoverReplyTask.status.in_(ACTIVE_TASK_STATUSES),
)
.all()
)
for task in tasks:
task.status = "cancelled"
task.cancel_reason = reason
task.locked_at = None
def _schedule_reply(self, db: Session, avatar: Avatar, event: TakeoverMessage):
active_tasks = (
db.query(TakeoverReplyTask)
.filter(
TakeoverReplyTask.owner_id == avatar.owner_id,
TakeoverReplyTask.peer_id == event.peer_id,
TakeoverReplyTask.status.in_(("pending", "generating", "ready")),
)
.order_by(TakeoverReplyTask.created_at.desc())
.all()
)
prompt_parts = []
source_ids = []
if active_tasks:
latest = active_tasks[0]
prompt_parts.append(latest.prompt)
source_ids.extend(latest.source_message_ids or [])
for task in active_tasks:
task.status = "cancelled"
task.cancel_reason = "newer_incoming_message"
task.locked_at = None
prompt_parts.append(event.content.strip())
source_ids.append(event.boxim_message_id)
prompt = "\n".join(part for part in prompt_parts if part).strip()[-MAX_PROMPT_LENGTH:]
due_at = event.send_time + timedelta(seconds=self.reply_delay_seconds)
task_id = secrets.token_hex(16)
local_id = int(time.time() * 1000) * 1000 + secrets.randbelow(1000)
db.add(
TakeoverReplyTask(
id=task_id,
avatar_id=avatar.id,
owner_id=avatar.owner_id,
peer_id=event.peer_id,
trigger_message_id=event.boxim_message_id,
source_message_ids=source_ids,
prompt=prompt,
status="pending",
scheduled_at=due_at,
boxim_local_id=str(local_id),
)
)
async def _prepare_replies(self) -> int:
db = self.session_factory()
try:
task_ids = [
row[0]
for row in (
db.query(TakeoverReplyTask.id)
.filter(
TakeoverReplyTask.status.in_(GENERATABLE_TASK_STATUSES),
TakeoverReplyTask.response_text == "",
)
.order_by(TakeoverReplyTask.created_at.asc())
.limit(10)
.all()
)
]
finally:
db.close()
generated = 0
for task_id in task_ids:
if await asyncio.to_thread(self._generate_reply, task_id):
generated += 1
return generated
def _generate_reply(self, task_id: str) -> bool:
db = self.session_factory()
try:
task = db.query(TakeoverReplyTask).filter(TakeoverReplyTask.id == task_id).first()
if not task or task.status != "pending":
return False
avatar = db.query(Avatar).filter(Avatar.id == task.avatar_id).first()
if not _takeover_enabled(avatar):
task.status = "cancelled"
task.cancel_reason = "takeover_disabled"
db.commit()
return False
task.status = "generating"
task.locked_at = self.now()
db.commit()
excluded_ids = set(task.source_message_ids or [])
events = (
db.query(TakeoverMessage)
.filter(
TakeoverMessage.owner_id == task.owner_id,
TakeoverMessage.peer_id == task.peer_id,
)
.order_by(TakeoverMessage.send_time.desc())
.limit(30)
.all()
)
history = []
for event in reversed(events):
if event.boxim_message_id in excluded_ids or not event.content.strip():
continue
history.append(
{
"role": "user" if event.direction == "incoming" else "assistant",
"content": event.content.strip(),
}
)
history = history[-10:]
from routers.chat import _resolve_reply
result = _resolve_reply(db, avatar, task.prompt, history)
answer = _plain_text_reply(result.get("answer", ""))
db.refresh(task)
if task.status != "generating":
return False
if not answer:
raise RuntimeError("分身没有生成有效回复")
task.response_text = answer
task.status = "ready"
task.locked_at = None
task.last_error = ""
db.commit()
return True
except Exception as exc:
db.rollback()
task = db.query(TakeoverReplyTask).filter(TakeoverReplyTask.id == task_id).first()
if task and task.status in ("pending", "generating"):
task.attempts = (task.attempts or 0) + 1
task.status = "pending" if task.attempts < 3 else "failed"
task.locked_at = None
task.last_error = str(exc)[:300]
db.commit()
logger.warning("Failed to prepare takeover reply %s: %s", task_id, exc)
return False
finally:
db.close()
async def _dispatch_ready_replies(self):
db = self.session_factory()
try:
task_ids = [
row[0]
for row in (
db.query(TakeoverReplyTask.id)
.filter(
TakeoverReplyTask.status == "ready",
TakeoverReplyTask.scheduled_at <= self.now(),
)
.order_by(TakeoverReplyTask.scheduled_at.asc())
.limit(10)
.all()
)
]
finally:
db.close()
for task_id in task_ids:
await self._send_task(task_id)
async def _send_task(self, task_id: str) -> bool:
db = self.session_factory()
user = None
try:
task = db.query(TakeoverReplyTask).filter(TakeoverReplyTask.id == task_id).first()
if not task or task.status != "ready":
return False
avatar = db.query(Avatar).filter(Avatar.id == task.avatar_id).first()
if not _takeover_enabled(avatar):
task.status = "cancelled"
task.cancel_reason = "takeover_disabled"
db.commit()
return False
if (self.now() - task.scheduled_at).total_seconds() > MAX_STALE_SECONDS:
task.status = "cancelled"
task.cancel_reason = "stale_reply"
db.commit()
return False
user = db.query(User).filter(User.huihui_user_id == task.owner_id).first()
if not user or not user.huihui_token:
raise BoxIMError("缺少会会登录凭证", auth_error=True)
task.status = "sending"
task.locked_at = self.now()
db.commit()
session = await self._boxim_session(user)
result = await self.boxim.send_private_message(
session["access_token"],
task.peer_id,
task.response_text,
local_id=task.boxim_local_id,
)
db.refresh(task)
if task.status != "sending":
return False
task.status = "sent"
task.sent_at = self.now()
task.locked_at = None
task.last_error = ""
task.boxim_sent_message_id = str(result.get("id") or "")
db.commit()
logger.info("BOXIM takeover reply sent for task %s", task.id)
return True
except Exception as exc:
db.rollback()
if user and isinstance(exc, BoxIMError) and exc.auth_error:
self._forget_boxim_session(user.id)
task = db.query(TakeoverReplyTask).filter(TakeoverReplyTask.id == task_id).first()
if task and task.status in ("ready", "sending"):
task.attempts = (task.attempts or 0) + 1
task.status = "ready" if task.attempts < 3 else "failed"
task.locked_at = None
task.last_error = str(exc)[:300]
if task.status == "ready":
task.scheduled_at = self.now() + timedelta(seconds=2 ** task.attempts)
db.commit()
logger.warning("Failed to send takeover reply %s: %s", task_id, exc)
return False
finally:
db.close()