fix(onlyoffice): 修复云端台账保存与备份重开

backend/app/services/onlyoffice_collaboration_service.py

@@ -59,33 +59,55 @@ async def _active_session(
         ).order_by(CollaborationSession.created_at.desc())
     )
     if session:
-        if session.status == "ERROR":
-            recovered = await _recover_forgotten_content(session.document_key, item.file_type)
-            if recovered is not None:
-                revision, _ = await collaboration_service.append_revision(
-                    db,
-                    item,
-                    recovered,
-                    source="SERVER_RECOVERY",
-                    created_by=user_id,
-                    change_summary="自动恢复在线文档服务器备份",
-                )
-                session.base_revision_id = revision.id
-            # The failed key points to Document Server's recovery cache. A new
-            # generation must use a new key or every subsequent open falls back
-            # to the same unsaved backup again.
-            item.generation += 1
-            session.status = "RECOVERED" if recovered is not None else "CLOSED"
-            session.closed_at = datetime.now(timezone.utc)
-            await db.commit()
-            session = None
-        else:
-            if session.status != "ACTIVE":
-                session.status = "ACTIVE"
-                session.closed_at = None
-                await db.commit()
-                await db.refresh(session)
-            return session
+        should_recover = session.status == "ERROR"
+        if session.status == "ACTIVE":
+            created_at = session.created_at
+            if created_at.tzinfo is None:
+                created_at = created_at.replace(tzinfo=timezone.utc)
+            if (
+                session.last_callback_at is None
+                and datetime.now(timezone.utc) - created_at < timedelta(seconds=15)
+            ):
+                # The editor config can be requested twice before the first
+                # browser has connected and emitted status 1. Keep a short
+                # connection grace period so the second request does not retire
+                # the freshly issued key as a false stale session.
+                return session
+            live_users = await _document_server_users(session.document_key)
+            if live_users:
+                return session
+            # A service restart or rejected final callback can leave the row
+            # ACTIVE after Document Server has already retired the editing
+            # process. Reusing that key opens its forgotten copy as an
+            # unbound server backup, so retire it exactly like a final callback.
+            should_recover = True
+
+        # A final callback ends the editing lifecycle for this key. Never
+        # reactivate it: Document Server can retain a cached or forgotten copy
+        # for the old key, especially across a container restart.
+        recovered = (
+            await _recover_forgotten_content(session.document_key, item.file_type)
+            if should_recover
+            else None
+        )
+        if recovered is not None:
+            revision, _ = await collaboration_service.append_revision(
+                db,
+                item,
+                recovered,
+                source="SERVER_RECOVERY",
+                created_by=user_id,
+                change_summary="自动恢复在线文档服务器备份",
+            )
+            session.base_revision_id = revision.id
+        # The retired key can still point to Document Server's cache. A new
+        # generation must use a new key or a later open can fall back to the
+        # same server-side copy again.
+        item.generation += 1
+        session.status = "RECOVERED" if recovered is not None else "CLOSED"
+        session.closed_at = datetime.now(timezone.utc)
+        await db.commit()
+        session = None
     if not item.current_revision_id:
         raise HTTPException(status_code=status.HTTP_409_CONFLICT, detail="协作文件尚无可编辑内容")
     session = CollaborationSession(
@@ -413,6 +435,67 @@ async def _recover_forgotten_content(document_key: str, file_type: str) -> bytes
     return content
 
 
+async def _document_server_users(document_key: str) -> list[str]:
+    """Return live editor ids for a key without trusting stale database state."""
+    command = {"c": "info", "key": document_key}
+    token = jwt.encode(command, settings.ONLYOFFICE_JWT_SECRET or "", algorithm="HS256")
+    command_url = f"{settings.ONLYOFFICE_INTERNAL_URL.rstrip('/')}/command"
+    try:
+        async with httpx.AsyncClient(timeout=10.0, follow_redirects=False) as client:
+            response = await client.post(
+                command_url,
+                params={"shardkey": document_key},
+                json={**command, "token": token},
+            )
+        if response.status_code != status.HTTP_200_OK:
+            raise HTTPException(
+                status_code=status.HTTP_503_SERVICE_UNAVAILABLE,
+                detail="在线文档会话检查服务暂不可用",
+            )
+        payload = response.json()
+    except (httpx.HTTPError, ValueError) as exc:
+        raise HTTPException(
+            status_code=status.HTTP_503_SERVICE_UNAVAILABLE,
+            detail="在线文档会话检查服务暂不可用",
+        ) from exc
+
+    error = payload.get("error") if isinstance(payload, dict) else None
+    if error == 1:
+        # The database can retain an ACTIVE row after an interrupted callback,
+        # while Document Server no longer has a live editing process for it.
+        return []
+    users = payload.get("users") if isinstance(payload, dict) else None
+    if (
+        error != 0
+        or payload.get("key") != document_key
+        or not isinstance(users, list)
+        or any(not isinstance(user_id, str) or not user_id for user_id in users)
+    ):
+        raise HTTPException(
+            status_code=status.HTTP_502_BAD_GATEWAY,
+            detail="在线文档服务器返回的会话信息无效",
+        )
+    return list(dict.fromkeys(users))
+
+
+async def list_live_editing_sessions(db: AsyncSession) -> list[tuple[str, int]]:
+    """List file titles and live editor counts for deployment safety checks."""
+    rows = (
+        await db.execute(
+            select(CollaborationSession.document_key, CollaborationFile.title)
+            .join(CollaborationFile, CollaborationFile.id == CollaborationSession.file_id)
+            .where(CollaborationSession.status == "ACTIVE")
+            .order_by(CollaborationFile.title)
+        )
+    ).all()
+    active: list[tuple[str, int]] = []
+    for document_key, title in rows:
+        users = await _document_server_users(document_key)
+        if users:
+            active.append((title, len(users)))
+    return active
+
+
 async def _callback_user(
     db: AsyncSession, payload: CollaborationCallbackPayload, session: CollaborationSession
 ) -> User | None: