feat: 实现 GlobalModel 别名匹配系统

主要更改: - GlobalModel 支持 model_aliases 配置，允许使用正则表达式定义别名规则 - Provider Key 的 allowed_models 现在可以通过别名规则匹配 GlobalModel - 新增 ModelAliasesTab 组件用于管理模型别名配置 - Provider 详情页新增别名映射预览功能，展示 Key 白名单与 GlobalModel 别名的匹配关系 - 路由预览 API 返回 Key 的 allowed_models 信息安全特性: - 使用 regex 库的原生超时保护（100ms）防止 ReDoS 攻击 - 别名规则数量限制（50 条/模型）和长度限制（200 字符） - 别名映射预览 API 添加超时保护和结果截断其他改进: - GlobalModel 更新/删除时使用行级锁防止并发竞态 - 缓存失效逻辑优化，支持异步清理和正则缓存清空 - 路由 Tab 布局重构，使用 flexbox 替代绝对定位
2026-09-02 17:30:23 +08:00 · 2026-01-13 16:04:15 +08:00
parent 9fea71a70c
commit 85decd7487
21 changed files with 3845 additions and 2308 deletions
--- a/src/api/handlers/base/chat_handler_base.py
+++ b/src/api/handlers/base/chat_handler_base.py
@@ -53,6 +53,7 @@ from src.models.database import (
    ProviderEndpoint,
    User,
 )
+from src.services.cache.aware_scheduler import ProviderCandidate
 from src.services.provider.transport import build_provider_url


@@ -312,6 +313,7 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
            provider: Provider,
            endpoint: ProviderEndpoint,
            key: ProviderAPIKey,
+            candidate: ProviderCandidate,
        ) -> AsyncGenerator[bytes, None]:
            return await self._execute_stream_request(
                ctx,
@@ -322,6 +324,7 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
                original_request_body,
                original_headers,
                query_params,
+                candidate,
            )

        try:
@@ -411,6 +414,7 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
        original_request_body: Dict[str, Any],
        original_headers: Dict[str, str],
        query_params: Optional[Dict[str, str]] = None,
+        candidate: Optional[ProviderCandidate] = None,
    ) -> AsyncGenerator[bytes, None]:
        """执行流式请求并返回流生成器"""
        # 重置上下文状态（重试时清除之前的数据）
@@ -425,11 +429,13 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
            provider_api_format=str(endpoint.api_format) if endpoint.api_format else None,
        )

-        # 获取模型映射
-        mapped_model = await self._get_mapped_model(
-            source_model=ctx.model,
-            provider_id=str(provider.id),
-        )
+        # 获取模型映射（优先使用别名匹配到的模型，其次是 Provider 级别的映射）
+        mapped_model = candidate.alias_matched_model if candidate else None
+        if not mapped_model:
+            mapped_model = await self._get_mapped_model(
+                source_model=ctx.model,
+                provider_id=str(provider.id),
+            )

        # 应用模型映射到请求体
        if mapped_model:
@@ -650,17 +656,20 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
            provider: Provider,
            endpoint: ProviderEndpoint,
            key: ProviderAPIKey,
+            candidate: ProviderCandidate,
        ) -> Dict[str, Any]:
            nonlocal provider_name, response_json, status_code, response_headers
            nonlocal provider_request_headers, provider_request_body, mapped_model_result

            provider_name = str(provider.name)

-            # 获取模型映射
-            mapped_model = await self._get_mapped_model(
-                source_model=model,
-                provider_id=str(provider.id),
-            )
+            # 获取模型映射（优先使用别名匹配到的模型，其次是 Provider 级别的映射）
+            mapped_model = candidate.alias_matched_model if candidate else None
+            if not mapped_model:
+                mapped_model = await self._get_mapped_model(
+                    source_model=model,
+                    provider_id=str(provider.id),
+                )

            # 应用模型映射
            if mapped_model:
--- a/src/api/handlers/base/cli_handler_base.py
+++ b/src/api/handlers/base/cli_handler_base.py
@@ -64,6 +64,7 @@ from src.models.database import (
    ProviderEndpoint,
    User,
 )
+from src.services.cache.aware_scheduler import ProviderCandidate
 from src.services.provider.transport import build_provider_url
 from src.utils.sse_parser import SSEEventParser
 from src.utils.timeout import read_first_chunk_with_ttfb_timeout
@@ -317,6 +318,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
            provider: Provider,
            endpoint: ProviderEndpoint,
            key: ProviderAPIKey,
+            candidate: ProviderCandidate,
        ) -> AsyncGenerator[bytes, None]:
            return await self._execute_stream_request(
                ctx,
@@ -326,6 +328,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
                original_request_body,
                original_headers,
                query_params,
+                candidate,
            )

        try:
@@ -405,6 +408,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
        original_request_body: Dict[str, Any],
        original_headers: Dict[str, str],
        query_params: Optional[Dict[str, str]] = None,
+        candidate: Optional[ProviderCandidate] = None,
    ) -> AsyncGenerator[bytes, None]:
        """执行流式请求并返回流生成器"""
        # 重置上下文状态（重试时清除之前的数据，避免累积）
@@ -432,11 +436,13 @@ class CliMessageHandlerBase(BaseMessageHandler):
        ctx.provider_api_format = str(endpoint.api_format) if endpoint.api_format else ""
        ctx.client_api_format = ctx.api_format  # 已在 process_stream 中设置

-        # 获取模型映射（映射名称 → 实际模型名）
-        mapped_model = await self._get_mapped_model(
-            source_model=ctx.model,
-            provider_id=str(provider.id),
-        )
+        # 获取模型映射（优先使用别名匹配到的模型，其次是 Provider 级别的映射）
+        mapped_model = candidate.alias_matched_model if candidate else None
+        if not mapped_model:
+            mapped_model = await self._get_mapped_model(
+                source_model=ctx.model,
+                provider_id=str(provider.id),
+            )

        # 应用模型映射到请求体（子类可覆盖此方法处理不同格式）
        if mapped_model:
@@ -1247,14 +1253,29 @@ class CliMessageHandlerBase(BaseMessageHandler):
        stream_generator: AsyncGenerator[bytes, None],
    ) -> AsyncGenerator[bytes, None]:
        """创建带监控的流生成器"""
+        import time as time_module
+
+        last_chunk_time = time_module.time()
+        chunk_count = 0
        try:
            async for chunk in stream_generator:
+                last_chunk_time = time_module.time()
+                chunk_count += 1
                yield chunk
        except asyncio.CancelledError:
+            # 计算距离上次收到 chunk 的时间
+            time_since_last_chunk = time_module.time() - last_chunk_time
            # 如果响应已完成，不标记为失败
            if not ctx.has_completion:
                ctx.status_code = 499
                ctx.error_message = "Client disconnected"
+                logger.warning(
+                    f"ID:{ctx.request_id} | Stream cancelled: "
+                    f"chunks={chunk_count}, "
+                    f"has_completion={ctx.has_completion}, "
+                    f"time_since_last_chunk={time_since_last_chunk:.2f}s, "
+                    f"output_tokens={ctx.output_tokens}"
+                )
            raise
        except httpx.TimeoutException as e:
            ctx.status_code = 504
@@ -1536,16 +1557,19 @@ class CliMessageHandlerBase(BaseMessageHandler):
            provider: Provider,
            endpoint: ProviderEndpoint,
            key: ProviderAPIKey,
+            candidate: ProviderCandidate,
        ) -> Dict[str, Any]:
            nonlocal provider_name, response_json, status_code, response_headers, provider_api_format, provider_request_headers, provider_request_body, mapped_model_result, response_metadata_result
            provider_name = str(provider.name)
            provider_api_format = str(endpoint.api_format) if endpoint.api_format else ""

-            # 获取模型映射（映射名称 → 实际模型名）
-            mapped_model = await self._get_mapped_model(
-                source_model=model,
-                provider_id=str(provider.id),
-            )
+            # 获取模型映射（优先使用别名匹配到的模型，其次是 Provider 级别的映射）
+            mapped_model = candidate.alias_matched_model if candidate else None
+            if not mapped_model:
+                mapped_model = await self._get_mapped_model(
+                    source_model=model,
+                    provider_id=str(provider.id),
+                )

            # 应用模型映射到请求体（子类可覆盖此方法处理不同格式）
            if mapped_model: