fix: 防止分页漏读和上游异常快照误标 orphaned

backend/app/services/website_sync.py

@@ -848,6 +848,7 @@ def reconcile_upstream_keys(
     active_group_ids: set[str] | None,
     remote_key_ids: set[str] | None,
     captured_at: datetime,
+    prev_active_group_ids: set[str] | None = None,
 ) -> None:
     """对账上游 Key 的本地缓存与远端状态。
 
@@ -857,6 +858,7 @@ def reconcile_upstream_keys(
     """
     if active_group_ids is None and remote_key_ids is None:
         return
+
     key_rows = (
         db.query(UpstreamGeneratedKey)
         .filter(
@@ -865,16 +867,40 @@ def reconcile_upstream_keys(
         .all()
     )
     for row in key_rows:
+        # 1. 增加自恢复逻辑：已导入的 orphaned key 如果后续快照中来源分组恢复，且远端 key id 仍存在，则自动恢复为 imported 并清空错误。
+        if (
+            row.status == "orphaned"
+            and row.error == "来源分组已不存在"
+            and row.imported_website_id is not None
+            and row.imported_account_id is not None
+            and active_group_ids is not None
+            and row.group_id in active_group_ids
+            and remote_key_ids is not None
+            and row.key_id in remote_key_ids
+        ):
+            row.status = "imported"
+            row.error = None
+            row.updated_at = captured_at
+            logger.info("recovered orphaned key %s to imported (group %s recovered and key_id %s exists remotely)", row.id, row.group_id, row.key_id)
+
+        # 2. 修改上游 key 对账逻辑：来源分组缺失时，不因单次快照缺失立即把已导入 key 标为 orphaned。
+        # 分组级 orphan 判断改为“两次连续快照都缺失该分组”才生效，避免上游接口临时返回不完整分组导致误标记。
         if active_group_ids is not None and row.group_id not in active_group_ids:
-            if row.imported_website_id and row.imported_account_id:
-                row.status = "orphaned"
-                row.error = "来源分组已不存在"
-                row.updated_at = captured_at
-                logger.info("marked key %s orphaned (group %s no longer in snapshot)", row.id, row.group_id)
+            if prev_active_group_ids is not None and row.group_id not in prev_active_group_ids:
+                if row.imported_website_id and row.imported_account_id:
+                    row.status = "orphaned"
+                    row.error = "来源分组已不存在"
+                    row.updated_at = captured_at
+                    logger.info("marked key %s orphaned (group %s no longer in snapshot consecutively)", row.id, row.group_id)
+                else:
+                    db.delete(row)
+                    logger.info("removed key %s (group %s no longer in snapshot consecutively)", row.id, row.group_id)
+                continue
             else:
-                db.delete(row)
-                logger.info("removed key %s (group %s no longer in snapshot)", row.id, row.group_id)
-            continue
+                # 只有单次快照缺失，跳过标记/清理，避免误标
+                logger.info("skipped marking key %s orphaned (group %s missing in single snapshot, waiting for next consecutive check)", row.id, row.group_id)
+
+        # 3. 远端 key id 级清理的现有保守策略：只有远端 key 列表成功拉取后，才允许判断 key 是否不存在。
         if row.key_id and remote_key_ids is not None and row.key_id not in remote_key_ids:
             if row.imported_website_id and row.imported_account_id:
                 row.status = "orphaned"
@@ -885,6 +911,7 @@ def reconcile_upstream_keys(
                 db.delete(row)
                 logger.info("removed key %s (key_id %s gone from remote)", row.id, row.key_id)
             continue
+
         if remote_key_ids is not None and row.key_id in remote_key_ids:
             row.updated_at = captured_at
 
@@ -914,11 +941,31 @@ def reconcile_upstream_keys_full(db: Session, upstream_id: int) -> bool:
     now = datetime.now(timezone.utc)
 
     # 从最新快照获取活跃分组 ID（与调度器 _sync_upstream_keys 一致）
+    prev_active_group_ids: set[str] | None = None
     groups = latest_rate_map(db, upstream_id)
     if groups:
         active_group_ids = set(groups.keys())
         groups_fetched = True
 
+        # 获取倒数第二个快照作为 previous snapshot
+        from app.models.snapshot import UpstreamRateSnapshot
+        snapshots = (
+            db.query(UpstreamRateSnapshot)
+            .filter(UpstreamRateSnapshot.upstream_id == upstream_id)
+            .order_by(UpstreamRateSnapshot.captured_at.desc())
+            .limit(2)
+            .all()
+        )
+        if len(snapshots) >= 2:
+            prev_snapshot = snapshots[1]
+            try:
+                prev_data = json.loads(prev_snapshot.snapshot_json or "{}")
+                prev_groups = prev_data.get("groups") or {}
+                if isinstance(prev_groups, dict):
+                    prev_active_group_ids = set(prev_groups.keys())
+            except Exception:
+                pass
+
     try:
         with UpstreamClient(
             base_url=upstream.base_url,
@@ -944,6 +991,7 @@ def reconcile_upstream_keys_full(db: Session, upstream_id: int) -> bool:
         active_group_ids if groups_fetched else None,
         remote_key_ids if keys_fetched else None,
         now,
+        prev_active_group_ids=prev_active_group_ids,
     )
     db.commit()
     return keys_fetched