Commit Graph

3 Commits

Author SHA1 Message Date
fawney19
6ec8df97e8 refactor: 限制流式文本收集内存增长,降低默认连接池和缓存上限
- StreamContext.append_text 增加 16KB 上限,超出后仅计数不存储,
  避免长流式响应导致内存持续增长;token 估算改用 collected_text_length
- 降低 DB 连接池上限 (30->15) 和 HTTP 连接池上限 (200->100)
- tiktoken 编码器缓存从 32 缩减到 4(实际编码种类只有几种)
- dev.sh 添加开发环境低配连接池默认值,uvicorn 热重载仅监视 src 目录
2026-03-10 15:33:46 +08:00
fawney19
7b66505634 refactor: 全局适配 ApiFamily/EndpointKind 结构化标识体系
将新的 (ApiFamily, EndpointKind) / `family:kind` 签名体系应用到整个代码库:
- API Handlers: 所有 adapter/handler 使用新的签名格式
- Services: provider, model, usage, cache, auth 等服务层适配
- Database: ProviderEndpoint 新增 api_family/endpoint_kind 字段
- Frontend: Provider 管理、Usage 表格等组件适配
- Tests: 更新所有相关测试用例
2026-02-01 17:28:00 +08:00
fawney19
ad1c8c394c refactor(handler): optimize stream processing and telemetry pipeline
- Enhance stream context for better token and latency tracking
- Refactor stream processor for improved performance metrics
- Improve telemetry integration with first_byte_time_ms support
- Add comprehensive stream context unit tests
2025-12-16 02:39:03 +08:00