i/Aether
1
0
Fork 0
mirror of https://github.com/fawney19/Aether.git synced 2026-09-03 01:40:21 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat: 添加格式转换追踪、模型过滤规则和 Provider 超时配置

Browse Source 
1. Usage 格式转换追踪
   - 新增 endpoint_api_format 和 has_format_conversion 字段
   - 在用量记录中显示格式转换信息（请求格式 → 端点格式）
   - 兼容历史数据的回填逻辑

2. Provider API Key 模型过滤规则
   - 新增 model_include_patterns 和 model_exclude_patterns 字段
   - 支持 * 和 ? 通配符，不区分大小写
   - 自动获取模型时应用过滤规则

3. Provider 超时配置
   - 新增 stream_first_byte_timeout 和 request_timeout 字段
   - 支持每个 Provider 单独配置超时时间
   - 优先使用 Provider 配置，否则回退到全局配置

Close #122, Close #123

Co-Authored-By: AAEE86 <33052466+AAEE86@users.noreply.github.com>
This commit is contained in:
fawney19
2026-01-28 00:04:47 +08:00
parent 3c0bf5fdae
commit c732a03263
23 changed files with 652 additions and 97 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
src/api/handlers/base/base_handler.py
View File

@@ -109,6 +109,9 @@ class MessageTelemetry:
provider_endpoint_id: Optional[str] = None,
provider_api_key_id: Optional[str] = None,
api_format: Optional[str] = None,
# 格式转换追踪
endpoint_api_format: Optional[str] = None, # 端点原生 API 格式
has_format_conversion: bool = False, # 是否发生了格式转换
# 模型映射信息
target_model: Optional[str] = None,
# Provider 响应元数据（如 Gemini 的 modelVersion）
@@ -135,6 +138,8 @@ class MessageTelemetry:
cache_read_input_tokens=cache_read_tokens,
request_type="chat",
api_format=api_format,
endpoint_api_format=endpoint_api_format,
has_format_conversion=has_format_conversion,
is_stream=is_stream,
response_time_ms=response_time_ms,
first_byte_time_ms=first_byte_time_ms, # 传递首字时间
@@ -195,6 +200,9 @@ class MessageTelemetry:
response_body: Optional[Dict[str, Any]] = None,
response_headers: Optional[Dict[str, Any]] = None,
client_response_headers: Optional[Dict[str, Any]] = None,
# 格式转换追踪
endpoint_api_format: Optional[str] = None,
has_format_conversion: bool = False,
# 模型映射信息
target_model: Optional[str] = None,
) -> None:
@@ -230,6 +238,8 @@ class MessageTelemetry:
cache_read_input_tokens=cache_read_tokens,
request_type="chat",
api_format=api_format,
endpoint_api_format=endpoint_api_format,
has_format_conversion=has_format_conversion,
is_stream=is_stream,
response_time_ms=response_time_ms,
status_code=status_code,
@@ -265,6 +275,9 @@ class MessageTelemetry:
response_body: Optional[Dict[str, Any]] = None,
response_headers: Optional[Dict[str, Any]] = None,
client_response_headers: Optional[Dict[str, Any]] = None,
# 格式转换追踪
endpoint_api_format: Optional[str] = None,
has_format_conversion: bool = False,
target_model: Optional[str] = None,
) -> None:
"""
@@ -286,6 +299,8 @@ class MessageTelemetry:
cache_read_input_tokens=cache_read_tokens,
request_type="chat",
api_format=api_format,
endpoint_api_format=endpoint_api_format,
has_format_conversion=has_format_conversion,
is_stream=is_stream,
response_time_ms=response_time_ms,
first_byte_time_ms=first_byte_time_ms,
@@ -488,6 +503,9 @@ class BaseMessageHandler:
key_id = ctx.key_id
first_byte_time_ms = ctx.first_byte_time_ms
api_format = ctx.api_format
# 格式转换追踪
endpoint_api_format = ctx.provider_api_format or None
has_format_conversion = ctx.needs_conversion
# 如果 provider 为空，记录警告（不应该发生，但用于调试）
if not provider:
@@ -512,6 +530,8 @@ class BaseMessageHandler:
provider_api_key_id=key_id,
first_byte_time_ms=first_byte_time_ms,
api_format=api_format,
endpoint_api_format=endpoint_api_format,
has_format_conversion=has_format_conversion,
)
finally:
db.close()

18
src/api/handlers/base/chat_handler_base.py
View File

@@ -713,7 +713,8 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
)
# 流式请求使用 stream_first_byte_timeout 作为首字节超时
request_timeout = config.stream_first_byte_timeout
# 优先使用 Provider 配置，否则使用全局配置
request_timeout = provider.stream_first_byte_timeout or config.stream_first_byte_timeout
# 创建 HTTP 客户端（支持代理配置，从 Provider 读取）
from src.clients.http_client import HTTPClientPool
@@ -851,6 +852,9 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
provider_request_headers=ctx.provider_request_headers,
response_headers=ctx.response_headers,
client_response_headers=client_response_headers,
# 格式转换追踪
endpoint_api_format=ctx.provider_api_format or None,
has_format_conversion=ctx.needs_conversion,
target_model=ctx.mapped_model,
)
@@ -987,7 +991,8 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
from src.clients.http_client import HTTPClientPool
# 非流式请求使用 http_request_timeout 作为整体超时
request_timeout = config.http_request_timeout
# 优先使用 Provider 配置，否则使用全局配置
request_timeout = provider.request_timeout or config.http_request_timeout
http_client = await HTTPClientPool.get_proxy_client(
proxy_config=provider.proxy,
)
@@ -1140,6 +1145,9 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
is_stream=False,
provider_request_headers=provider_request_headers,
api_format=api_format,
# 格式转换追踪
endpoint_api_format=provider_api_format_for_error or None,
has_format_conversion=needs_conversion_for_error,
provider_id=provider_id,
provider_endpoint_id=endpoint_id,
provider_api_key_id=key_id,
@@ -1203,6 +1211,9 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
provider_request_headers=provider_request_headers,
response_headers=response_headers,
client_response_headers={"content-type": "application/json"},
# 格式转换追踪
endpoint_api_format=provider_api_format_for_error or None,
has_format_conversion=needs_conversion_for_error,
target_model=mapped_model_result,
)
client_format = (client_api_format_for_error or "").upper()
@@ -1249,6 +1260,9 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
response_headers=error_response_headers,
# 非流式失败返回给客户端的是 JSON 错误响应
client_response_headers={"content-type": "application/json"},
# 格式转换追踪
endpoint_api_format=provider_api_format_for_error or None,
has_format_conversion=needs_conversion_for_error,
# 模型映射信息
target_model=mapped_model_result,
)

24
src/api/handlers/base/cli_handler_base.py
View File

@@ -739,7 +739,8 @@ class CliMessageHandlerBase(BaseMessageHandler):
)
# 流式请求使用 stream_first_byte_timeout 作为首字节超时
request_timeout = config.stream_first_byte_timeout
# 优先使用 Provider 配置，否则使用全局配置
request_timeout = provider.stream_first_byte_timeout or config.stream_first_byte_timeout
logger.debug(
f" └─ [{self.request_id}] 发送流式请求: "
@@ -1068,7 +1069,8 @@ class CliMessageHandlerBase(BaseMessageHandler):
provider_parser = self.parser
# 使用共享的 TTFB 超时函数读取首字节
ttfb_timeout = config.stream_first_byte_timeout
# 优先使用 Provider 配置，否则使用全局配置
ttfb_timeout = provider.stream_first_byte_timeout or config.stream_first_byte_timeout
first_chunk, aiter = await read_first_chunk_with_ttfb_timeout(
byte_iterator,
timeout=ttfb_timeout,
@@ -1804,6 +1806,9 @@ class CliMessageHandlerBase(BaseMessageHandler):
response_body=response_body,
response_headers=ctx.response_headers,
client_response_headers=client_response_headers,
# 格式转换追踪
endpoint_api_format=ctx.provider_api_format or None,
has_format_conversion=ctx.needs_conversion,
# 模型映射信息
target_model=ctx.mapped_model,
)
@@ -1845,6 +1850,9 @@ class CliMessageHandlerBase(BaseMessageHandler):
is_stream=True,
provider_request_headers=ctx.provider_request_headers,
api_format=ctx.api_format,
# 格式转换追踪
endpoint_api_format=ctx.provider_api_format or None,
has_format_conversion=ctx.needs_conversion,
# Provider 侧追踪信息（用于记录真实成本）
provider_id=ctx.provider_id,
provider_endpoint_id=ctx.endpoint_id,
@@ -1977,6 +1985,9 @@ class CliMessageHandlerBase(BaseMessageHandler):
provider_request_headers=ctx.provider_request_headers,
response_headers=ctx.response_headers,
client_response_headers=client_response_headers,
# 格式转换追踪
endpoint_api_format=ctx.provider_api_format or None,
has_format_conversion=ctx.needs_conversion,
# 模型映射信息
target_model=ctx.mapped_model,
)
@@ -2101,7 +2112,8 @@ class CliMessageHandlerBase(BaseMessageHandler):
from src.clients.http_client import HTTPClientPool
# 非流式请求使用 http_request_timeout 作为整体超时
request_timeout = config.http_request_timeout
# 优先使用 Provider 配置，否则使用全局配置
request_timeout = provider.request_timeout or config.http_request_timeout
http_client = await HTTPClientPool.get_proxy_client(
proxy_config=provider.proxy,
)
@@ -2274,6 +2286,9 @@ class CliMessageHandlerBase(BaseMessageHandler):
is_stream=False,
provider_request_headers=provider_request_headers,
api_format=api_format,
# 格式转换追踪
endpoint_api_format=provider_api_format or None,
has_format_conversion=needs_conversion,
# Provider 侧追踪信息（用于记录真实成本）
provider_id=provider_id,
provider_endpoint_id=endpoint_id,
@@ -2346,6 +2361,9 @@ class CliMessageHandlerBase(BaseMessageHandler):
response_headers=error_response_headers,
# 非流式失败返回给客户端的是 JSON 错误响应
client_response_headers={"content-type": "application/json"},
# 格式转换追踪
endpoint_api_format=provider_api_format or None,
has_format_conversion=needs_conversion,
# 模型映射信息
target_model=mapped_model_result,
)

5
src/api/handlers/base/stream_processor.py
View File

@@ -213,7 +213,8 @@ class StreamProcessor:
try:
# 使用共享的 TTFB 超时函数读取首字节
ttfb_timeout = config.stream_first_byte_timeout
# 优先使用 Provider 配置，否则使用全局配置
ttfb_timeout = provider.stream_first_byte_timeout or config.stream_first_byte_timeout
first_chunk, aiter = await read_first_chunk_with_ttfb_timeout(
byte_iterator,
timeout=ttfb_timeout,
@@ -422,6 +423,8 @@ class StreamProcessor:
f"[{self.request_id}] needs_conversion=True 但 provider_format 为空，回退到透传模式"
)
needs_conversion = False
# 保持 ctx 与实际行为一致，避免 Usage 记录误标记为转换
ctx.needs_conversion = False
def _mark_stream_started() -> None:
nonlocal start_time, streaming_started