fix: 一键同步模型接口同步修复并携带上游 base_url，防止 sub2api 账号 base_url 被清空或置为默认值

backend/app/routers/websites.py

@@ -1330,6 +1330,7 @@ def sync_website_accounts_upstream_models(
             if aid not in candidates:
                 candidates[aid] = {
                     "account_id": aid,
+                    "upstream_id": key.upstream_id,
                     "db_key": key
                 }
 
@@ -1340,6 +1341,9 @@ def sync_website_accounts_upstream_models(
                 items=[]
             )
 
+        upstream_ids = {cand["upstream_id"] for cand in candidates.values()}
+        upstreams_map = {up.id: up for up in db.query(Upstream).filter(Upstream.id.in_(upstream_ids)).all()}
+
         items = []
 
         for aid, cand in candidates.items():
@@ -1372,29 +1376,56 @@ def sync_website_accounts_upstream_models(
                 ))
                 continue
 
-            # 开始同步该账号的模型
+            upstream = upstreams_map.get(cand["upstream_id"])
+            upstream_base_url = upstream.base_url if upstream else None
+
+            if not upstream_base_url or not upstream_base_url.strip():
+                items.append(SyncUpstreamModelsItem(
+                    account_id=aid,
+                    account_name=acc_name,
+                    model_count=0,
+                    models=[],
+                    status="failed",
+                    message="上游 base_url 为空，跳过处理避免继续写坏账号"
+                ))
+                continue
+
+            # 1. 先安全修复 base_url (不依赖同步模型结果)
+            try:
+                c.update_account(aid, {"credentials": {"base_url": upstream_base_url}})
+            except Exception as e:
+                items.append(SyncUpstreamModelsItem(
+                    account_id=aid,
+                    account_name=acc_name,
+                    model_count=0,
+                    models=[],
+                    status="failed",
+                    message=f"修复 Base URL 失败: {e}"
+                ))
+                continue
+
+            # 2. 调用 sub2api 同步模型
             try:
                 raw_models = c.sync_account_upstream_models(aid)
                 # 过滤空、去重、排序
                 valid_models = sorted(list(set(m.strip() for m in raw_models if m and m.strip())))
 
                 if not valid_models:
-                    # 如果返回空模型，保守处理为失败，不清空已有的模型白名单配置
                     items.append(SyncUpstreamModelsItem(
                         account_id=aid,
                         account_name=acc_name,
                         model_count=0,
                         models=[],
                         status="failed",
-                        message="上游同步返回模型列表为空"
+                        message="已修复 Base URL，但上游同步返回模型列表为空"
                     ))
                     continue
 
-                # 构造 model_mapping
+                # 3. 模型同步成功后，写回 base_url 和 model_mapping
-                model_mapping = {m: m for m in valid_models}
+                c.update_account(aid, {"credentials": {
-
+                    "base_url": upstream_base_url,
-                # 直接发送增量字段进行更新，依赖远端的 JSONB merge 逻辑，避免覆盖/丢失敏感 credentials
+                    "model_mapping": {m: m for m in valid_models}
-                c.update_account(aid, {"credentials": {"model_mapping": model_mapping}})
+                }})
 
                 items.append(SyncUpstreamModelsItem(
                     account_id=aid,
@@ -1402,7 +1433,7 @@ def sync_website_accounts_upstream_models(
                     model_count=len(valid_models),
                     models=valid_models,
                     status="success",
-                    message=f"成功同步 {len(valid_models)} 个模型"
+                    message=f"已修复 Base URL 并同步 {len(valid_models)} 个模型"
                 ))
             except Exception as e:
                 items.append(SyncUpstreamModelsItem(
@@ -1411,7 +1442,7 @@ def sync_website_accounts_upstream_models(
                     model_count=0,
                     models=[],
                     status="failed",
-                    message=f"同步/保存上游模型失败: {e}"
+                    message=f"已修复 Base URL，但模型同步失败: {e}"
                 ))
 
         success_count = sum(1 for item in items if item.status == "success")

backend/test_sync_upstream_models.py

@@ -43,6 +43,9 @@ def test_sync_upstream_models_workflow(db_session, monkeypatch):
 
     up = Upstream(id=10, name="Upstream A", base_url="http://upstream-a")
     db_session.add(up)
+
+    up_empty = Upstream(id=11, name="Upstream Empty", base_url="")
+    db_session.add(up_empty)
     db_session.commit()
 
     # 2. 插入本地导入的 Key 记录
@@ -90,7 +93,7 @@ def test_sync_upstream_models_workflow(db_session, monkeypatch):
         imported_website_id=w.id,
         imported_account_id="1004",
     )
-    # k5: 属于本站，但同步接口返回空模型（应当标记为 failed，且不覆盖）
+    # k5: 属于本站，但同步接口返回空模型（应当标记为 failed，但应先安全修复 base_url）
     k5 = UpstreamGeneratedKey(
         id=105,
         upstream_id=up.id,
@@ -101,7 +104,7 @@ def test_sync_upstream_models_workflow(db_session, monkeypatch):
         imported_website_id=w.id,
         imported_account_id="1005",
     )
-    # k6: 属于本站，但同步接口抛错（应当标记为 failed，且不覆盖）
+    # k6: 属于本站，但同步接口抛错（应当标记为 failed，但应先安全修复 base_url）
     k6 = UpstreamGeneratedKey(
         id=106,
         upstream_id=up.id,
@@ -112,7 +115,18 @@ def test_sync_upstream_models_workflow(db_session, monkeypatch):
         imported_website_id=w.id,
         imported_account_id="1006",
     )
-    db_session.add_all([k1, k2, k3, k4, k5, k6])
+    # k7: 属于本站，但对应上游 base_url 为空（应当直接标记为 failed 且不调用同步和写回）
+    k7 = UpstreamGeneratedKey(
+        id=107,
+        upstream_id=up_empty.id,
+        group_id="g1",
+        key_name="key-107",
+        key_value="val-107",
+        status="active",
+        imported_website_id=w.id,
+        imported_account_id="1007",
+    )
+    db_session.add_all([k1, k2, k3, k4, k5, k6, k7])
     db_session.commit()
 
     # Mock Sub2Api 客户端
@@ -131,10 +145,11 @@ def test_sync_upstream_models_workflow(db_session, monkeypatch):
 
         def list_accounts(self):
             return [
-                {"id": 1001, "name": "acc-1001", "credentials": {"api_key": "k1", "model_mapping": {"old": "old"}}},
+                {"id": 1001, "name": "acc-1001", "credentials": {"api_key": "k1", "model_mapping": {"old": "old"}, "base_url": "http://default-base-url"}},
                 {"id": "abc", "name": "acc-abc", "credentials": {}},
                 {"id": 1005, "name": "acc-1005", "credentials": {"api_key": "k5"}},
                 {"id": 1006, "name": "acc-1006", "credentials": {"api_key": "k6"}},
+                {"id": 1007, "name": "acc-1007", "credentials": {"api_key": "k7"}},
             ]
 
         def extract_id(self, val):
@@ -160,25 +175,31 @@ def test_sync_upstream_models_workflow(db_session, monkeypatch):
     # 验证总体结果
     assert res.success is False  # 存在 failed 账号，success 应该为 False
     assert "成功 1 个" in res.message
-    assert "失败 2 个" in res.message
+    assert "失败 3 个" in res.message
     assert "跳过 2 个" in res.message
 
     items = res.items
-    assert len(items) == 5
+    assert len(items) == 6
 
-    # 1001 成功
+    # 1001 成功 (修复 Base URL 并同步)
     item_1001 = next(i for i in items if i.account_id == "1001")
     assert item_1001.status == "success"
     assert item_1001.model_count == 2
     assert item_1001.models == ["gpt-3.5-turbo", "gpt-4"]
+    assert "已修复 Base URL" in item_1001.message
 
-    # 1001 的 update 应当只传递增量的 model_mapping，防止敏感字段丢失
+    # 1001 的两次 update 调用检测
-    assert len(update_calls) == 1
+    calls_1001 = [call for call in update_calls if call[0] == "1001"]
-    assert update_calls[0][0] == "1001"
+    assert len(calls_1001) == 2
-    assert "api_key" not in update_calls[0][1]["credentials"]
+    # 第一次安全修复：只传递 base_url
-    assert update_calls[0][1]["credentials"]["model_mapping"] == {
+    assert calls_1001[0][1]["credentials"] == {"base_url": "http://upstream-a"}
-        "gpt-3.5-turbo": "gpt-3.5-turbo",
+    # 第二次同步成功写入：传递 base_url + model_mapping
-        "gpt-4": "gpt-4"
+    assert calls_1001[1][1]["credentials"] == {
+        "base_url": "http://upstream-a",
+        "model_mapping": {
+            "gpt-3.5-turbo": "gpt-3.5-turbo",
+            "gpt-4": "gpt-4"
+        }
     }
 
     # abc 跳过
@@ -191,15 +212,31 @@ def test_sync_upstream_models_workflow(db_session, monkeypatch):
     assert item_1004.status == "skipped"
     assert "不存在" in item_1004.message
 
-    # 1005 失败 (空模型)
+    # 1005 失败 (空模型，但也应当执行了第一次安全修复的 update_account)
     item_1005 = next(i for i in items if i.account_id == "1005")
     assert item_1005.status == "failed"
+    assert "已修复 Base URL" in item_1005.message
     assert "为空" in item_1005.message
+    calls_1005 = [call for call in update_calls if call[0] == "1005"]
+    assert len(calls_1005) == 1
+    assert calls_1005[0][1]["credentials"] == {"base_url": "http://upstream-a"}
 
-    # 1006 失败 (抛错)
+    # 1006 失败 (抛错，但应当执行了第一次安全修复的 update_account)
     item_1006 = next(i for i in items if i.account_id == "1006")
     assert item_1006.status == "failed"
+    assert "已修复 Base URL" in item_1006.message
     assert "Network Error" in item_1006.message
+    calls_1006 = [call for call in update_calls if call[0] == "1006"]
+    assert len(calls_1006) == 1
+    assert calls_1006[0][1]["credentials"] == {"base_url": "http://upstream-a"}
+
+    # 1007 失败 (上游 base_url 为空，不应当调用 update_account 且不调用同步模型)
+    item_1007 = next(i for i in items if i.account_id == "1007")
+    assert item_1007.status == "failed"
+    assert "base_url 为空" in item_1007.message
+    calls_1007 = [call for call in update_calls if call[0] == "1007"]
+    assert len(calls_1007) == 0
+    assert "1007" not in sync_calls
 
     assert closed_count == 1