mirror of
https://github.com/fawney19/Aether.git
synced 2026-09-02 01:10:23 +08:00
refactor: 移除 Python 后端源码，全面迁移至 Rust gateway 架构
- 删除全部 Python 源码 (src/) 及 Alembic 迁移脚本，归档至 _deprecated_py_src/ - 重构 Rust gateway ai_pipeline: 拆分 planner/finalize 模块，新增 contracts/adaptation 层 - 重组 handlers 模块为 admin/public/proxy/internal/shared 子模块结构 - 新增 executor 模块，引入 Rust 原生数据库迁移 (aether-data/migrations) - 简化 CI/Docker 构建流程，移除 base image 二级构建，统一为单一 app image - 移除 Python 相关基础设施文件 (entrypoint.sh, gunicorn_conf.py, Dockerfile.base)
This commit is contained in:
fawney19
25
_deprecated_py_src/api/handlers/gemini_cli/__init__.py
Normal file
25
_deprecated_py_src/api/handlers/gemini_cli/__init__.py
Normal file
@@ -0,0 +1,25 @@
|
||||
"""Gemini CLI handler package (lazy exports)."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
from importlib import import_module
|
||||
from typing import Any
|
||||
|
||||
_LAZY_EXPORTS: dict[str, tuple[str, str]] = {
|
||||
"GeminiCliAdapter": (".adapter", "GeminiCliAdapter"),
|
||||
"build_gemini_cli_adapter": (".adapter", "build_gemini_cli_adapter"),
|
||||
"GeminiCliMessageHandler": (".handler", "GeminiCliMessageHandler"),
|
||||
}
|
||||
|
||||
__all__ = list(_LAZY_EXPORTS.keys())
|
||||
|
||||
|
||||
def __getattr__(name: str) -> Any:
|
||||
if name not in _LAZY_EXPORTS:
|
||||
raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
|
||||
|
||||
module_name, attr_name = _LAZY_EXPORTS[name]
|
||||
module = import_module(module_name, __name__)
|
||||
value = getattr(module, attr_name)
|
||||
globals()[name] = value
|
||||
return value
|
||||
171
_deprecated_py_src/api/handlers/gemini_cli/adapter.py
Normal file
171
_deprecated_py_src/api/handlers/gemini_cli/adapter.py
Normal file
@@ -0,0 +1,171 @@
|
||||
"""
|
||||
Gemini CLI Adapter - 基于通用 CLI Adapter 基类的实现
|
||||
|
||||
继承 CliAdapterBase，处理 Gemini CLI 格式的请求。
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
from typing import Any
|
||||
|
||||
from fastapi import Request
|
||||
|
||||
from src.api.handlers.base.cli_adapter_base import CliAdapterBase, register_cli_adapter
|
||||
from src.api.handlers.base.cli_handler_base import CliMessageHandlerBase
|
||||
from src.api.handlers.gemini.adapter import GeminiCapabilityDetector
|
||||
from src.config.settings import config
|
||||
from src.core.api_format import ApiFamily, get_auth_handler
|
||||
from src.core.api_format.enums import AuthMethod
|
||||
|
||||
|
||||
@register_cli_adapter
|
||||
class GeminiCliAdapter(CliAdapterBase):
|
||||
"""
|
||||
Gemini CLI API 适配器
|
||||
|
||||
处理 Gemini CLI 格式的请求（透传模式，最小验证）。
|
||||
"""
|
||||
|
||||
FORMAT_ID = "gemini:cli"
|
||||
API_FAMILY = ApiFamily.GEMINI
|
||||
name = "gemini.cli"
|
||||
|
||||
@property
|
||||
def HANDLER_CLASS(self) -> type[CliMessageHandlerBase]:
|
||||
"""延迟导入 Handler 类避免循环依赖"""
|
||||
from src.api.handlers.gemini_cli.handler import GeminiCliMessageHandler
|
||||
|
||||
return GeminiCliMessageHandler
|
||||
|
||||
def __init__(self, allowed_api_formats: list[str] | None = None):
|
||||
super().__init__(allowed_api_formats)
|
||||
|
||||
def extract_api_key(self, request: Request) -> str | None:
|
||||
"""
|
||||
从请求中提取 API 密钥 - Gemini CLI 支持 header 和 query 两种方式
|
||||
|
||||
优先级（与 Google SDK 行为一致）：
|
||||
1. URL 参数 ?key=
|
||||
2. x-goog-api-key 请求头
|
||||
"""
|
||||
handler = get_auth_handler(AuthMethod.GOOG_API_KEY)
|
||||
return handler.extract_credentials(request)
|
||||
|
||||
def detect_capability_requirements(
|
||||
self,
|
||||
headers: dict[str, str],
|
||||
request_body: dict[str, Any] | None = None,
|
||||
) -> dict[str, bool]:
|
||||
"""从请求体检测 Gemini 能力需求（fileData.fileUri -> gemini_files）"""
|
||||
return GeminiCapabilityDetector.detect_from_request(headers, request_body)
|
||||
|
||||
def _merge_path_params(
|
||||
self, original_request_body: dict[str, Any], path_params: dict[str, Any] # noqa: ARG002
|
||||
) -> dict[str, Any]:
|
||||
"""
|
||||
合并 URL 路径参数到请求体 - Gemini CLI 特化版本
|
||||
|
||||
Gemini API 特点:
|
||||
- model 不合并到请求体（Gemini 原生请求体不含 model，通过 URL 路径传递）
|
||||
- stream 不合并到请求体（Gemini API 通过 URL 端点区分流式/非流式）
|
||||
|
||||
基类已经从 path_params 获取 model 和 stream 用于日志和路由判断。
|
||||
|
||||
Args:
|
||||
original_request_body: 原始请求体字典
|
||||
path_params: URL 路径参数字典（包含 model、stream 等）
|
||||
|
||||
Returns:
|
||||
原始请求体（不合并任何 path_params）
|
||||
"""
|
||||
# Gemini: 不合并任何 path_params 到请求体
|
||||
return original_request_body.copy()
|
||||
|
||||
def _extract_message_count(self, payload: dict[str, Any]) -> int:
|
||||
"""Gemini CLI 使用 contents 字段"""
|
||||
contents = payload.get("contents", [])
|
||||
return len(contents) if isinstance(contents, list) else 0
|
||||
|
||||
def _build_audit_metadata(
|
||||
self,
|
||||
payload: dict[str, Any],
|
||||
path_params: dict[str, Any] | None = None,
|
||||
) -> dict[str, Any]:
|
||||
"""Gemini CLI 特定的审计元数据"""
|
||||
# 从 path_params 获取 model（Gemini 请求体不含 model）
|
||||
model = path_params.get("model", "unknown") if path_params else "unknown"
|
||||
contents = payload.get("contents", [])
|
||||
generation_config = payload.get("generation_config", {}) or {}
|
||||
|
||||
role_counts: dict[str, int] = {}
|
||||
for content in contents:
|
||||
role = content.get("role", "unknown") if isinstance(content, dict) else "unknown"
|
||||
role_counts[role] = role_counts.get(role, 0) + 1
|
||||
|
||||
return {
|
||||
"action": "gemini_cli_request",
|
||||
"model": model,
|
||||
"stream": bool(payload.get("stream", False)),
|
||||
"max_output_tokens": generation_config.get("max_output_tokens"),
|
||||
"contents_count": len(contents),
|
||||
"content_roles": role_counts,
|
||||
"temperature": generation_config.get("temperature"),
|
||||
"top_p": generation_config.get("top_p"),
|
||||
"top_k": generation_config.get("top_k"),
|
||||
"tools_count": len(payload.get("tools") or []),
|
||||
"system_instruction_present": bool(payload.get("system_instruction")),
|
||||
"safety_settings_count": len(payload.get("safety_settings") or []),
|
||||
}
|
||||
|
||||
@classmethod
|
||||
def build_endpoint_url(
|
||||
cls,
|
||||
base_url: str,
|
||||
request_data: dict[str, Any],
|
||||
model_name: str | None = None,
|
||||
*,
|
||||
provider_type: str | None = None,
|
||||
) -> str:
|
||||
"""构建Gemini CLI API端点URL"""
|
||||
effective_model_name = model_name or request_data.get("model", "")
|
||||
if not effective_model_name:
|
||||
raise ValueError("Model name is required for Gemini API")
|
||||
|
||||
base_url = base_url.rstrip("/")
|
||||
if base_url.endswith("/v1beta"):
|
||||
prefix = base_url
|
||||
else:
|
||||
prefix = f"{base_url}/v1beta"
|
||||
return f"{prefix}/models/{effective_model_name}:generateContent"
|
||||
|
||||
# build_request_body 使用基类实现，通过 format_conversion_registry 自动转换 OPENAI -> GEMINI_CLI
|
||||
|
||||
@classmethod
|
||||
def get_cli_user_agent(cls) -> str | None:
|
||||
"""获取Gemini CLI User-Agent"""
|
||||
return config.internal_user_agent_gemini_cli
|
||||
|
||||
@classmethod
|
||||
def get_cli_extra_headers(
|
||||
cls, *, base_url: str | None = None, provider_type: str | None = None
|
||||
) -> dict[str, str]:
|
||||
"""获取Gemini CLI额外请求头，包含 x-app: cli 标识"""
|
||||
headers = super().get_cli_extra_headers(base_url=base_url, provider_type=provider_type)
|
||||
headers["x-app"] = "cli" # 标识 CLI 模式，让上游使用正确的 adapter
|
||||
return headers
|
||||
|
||||
|
||||
def build_gemini_cli_adapter(x_app_header: str = "") -> GeminiCliAdapter:
|
||||
"""
|
||||
构建 Gemini CLI 适配器
|
||||
|
||||
Args:
|
||||
x_app_header: X-App 请求头值（预留扩展）
|
||||
|
||||
Returns:
|
||||
GeminiCliAdapter 实例
|
||||
"""
|
||||
return GeminiCliAdapter()
|
||||
|
||||
|
||||
__all__ = ["GeminiCliAdapter", "build_gemini_cli_adapter"]
|
||||
291
_deprecated_py_src/api/handlers/gemini_cli/handler.py
Normal file
291
_deprecated_py_src/api/handlers/gemini_cli/handler.py
Normal file
@@ -0,0 +1,291 @@
|
||||
"""
|
||||
Gemini CLI Message Handler - 基于通用 CLI Handler 基类的实现
|
||||
|
||||
继承 CliMessageHandlerBase，处理 Gemini CLI API 格式的请求。
|
||||
"""
|
||||
|
||||
from typing import Any
|
||||
|
||||
from src.api.handlers.base.cli_handler_base import (
|
||||
CliMessageHandlerBase,
|
||||
StreamContext,
|
||||
)
|
||||
from src.core.api_format import ApiFamily, EndpointKind
|
||||
|
||||
|
||||
class GeminiCliMessageHandler(CliMessageHandlerBase):
|
||||
"""
|
||||
Gemini CLI Message Handler - 处理 Gemini CLI API 格式
|
||||
|
||||
使用新三层架构 (Provider -> ProviderEndpoint -> ProviderAPIKey)
|
||||
通过 TaskService/FailoverEngine 实现自动故障转移、健康监控和并发控制
|
||||
|
||||
响应格式特点：
|
||||
- Gemini 使用 JSON 数组格式流式响应（非 SSE）
|
||||
- 每个 chunk 包含 candidates、usageMetadata 等字段
|
||||
- finish_reason: STOP, MAX_TOKENS, SAFETY, RECITATION, OTHER
|
||||
- Token 使用: promptTokenCount (输入), thoughtsTokenCount + candidatesTokenCount (输出), cachedContentTokenCount (缓存)
|
||||
|
||||
Gemini API 特殊处理：
|
||||
- model 在 URL 路径中而非请求体，如 /v1beta/models/{model}:generateContent
|
||||
- 请求体中的 model 字段用于内部路由，不发送给 API
|
||||
"""
|
||||
|
||||
FORMAT_ID = "gemini:cli"
|
||||
API_FAMILY = ApiFamily.GEMINI
|
||||
ENDPOINT_KIND = EndpointKind.CLI
|
||||
|
||||
def extract_model_from_request(
|
||||
self,
|
||||
request_body: dict[str, Any], # noqa: ARG002 - 基类签名要求
|
||||
path_params: dict[str, Any] | None = None,
|
||||
) -> str:
|
||||
"""
|
||||
从请求中提取模型名 - Gemini 格式实现
|
||||
|
||||
Gemini API 的 model 在 URL 路径中而非请求体：
|
||||
/v1beta/models/{model}:generateContent
|
||||
|
||||
Args:
|
||||
request_body: 请求体（Gemini 不包含 model）
|
||||
path_params: URL 路径参数（包含 model）
|
||||
|
||||
Returns:
|
||||
模型名，如果无法提取则返回 "unknown"
|
||||
"""
|
||||
# Gemini: model 从 URL 路径参数获取
|
||||
if path_params and "model" in path_params:
|
||||
return str(path_params["model"])
|
||||
return "unknown"
|
||||
|
||||
def prepare_provider_request_body(
|
||||
self,
|
||||
request_body: dict[str, Any],
|
||||
) -> dict[str, Any]:
|
||||
"""
|
||||
准备发送给 Gemini API 的请求体 - 移除 model 字段
|
||||
|
||||
Gemini API 要求 model 只在 URL 路径中，请求体中的 model 字段
|
||||
会导致某些代理返回 404 错误。
|
||||
|
||||
Args:
|
||||
request_body: 请求体
|
||||
|
||||
Returns:
|
||||
不含 model 字段的请求体
|
||||
"""
|
||||
result = dict(request_body)
|
||||
result.pop("model", None)
|
||||
return result
|
||||
|
||||
def finalize_provider_request(
|
||||
self,
|
||||
request_body: dict[str, Any],
|
||||
*,
|
||||
mapped_model: str | None,
|
||||
provider_api_format: str | None, # noqa: ARG002
|
||||
) -> dict[str, Any]:
|
||||
from src.api.handlers.gemini.image_gen import (
|
||||
adapt_request_for_image_gen,
|
||||
is_image_gen_model,
|
||||
)
|
||||
|
||||
# Sanitize Gemini contents: strip parts without a valid data-oneof
|
||||
# field and merge consecutive same-role entries. This catches cases
|
||||
# missed by the normalizer (passthrough) or the antigravity envelope.
|
||||
from src.core.api_format.conversion.normalizers.gemini import (
|
||||
compact_gemini_contents,
|
||||
)
|
||||
|
||||
contents = request_body.get("contents")
|
||||
if isinstance(contents, list):
|
||||
request_body["contents"] = compact_gemini_contents(contents)
|
||||
|
||||
if not is_image_gen_model(mapped_model):
|
||||
return request_body
|
||||
return adapt_request_for_image_gen(request_body)
|
||||
|
||||
def get_model_for_url(
|
||||
self,
|
||||
request_body: dict[str, Any],
|
||||
mapped_model: str | None,
|
||||
) -> str | None:
|
||||
"""
|
||||
Gemini 需要将 model 放入 URL 路径中
|
||||
|
||||
Args:
|
||||
request_body: 请求体
|
||||
mapped_model: 映射后的模型名（如果有）
|
||||
|
||||
Returns:
|
||||
用于 URL 路径的模型名
|
||||
"""
|
||||
# 优先使用映射后的模型名，否则使用请求体中的
|
||||
return mapped_model or request_body.get("model")
|
||||
|
||||
def _extract_usage_from_event(
|
||||
self,
|
||||
event: dict[str, Any],
|
||||
*,
|
||||
provider_type: str | None = None,
|
||||
) -> dict[str, int]:
|
||||
"""
|
||||
从 Gemini 事件中提取 token 使用情况
|
||||
|
||||
调用 GeminiStreamParser.extract_usage 作为单一实现源
|
||||
|
||||
Args:
|
||||
event: Gemini 流式响应事件
|
||||
provider_type: Provider 类型（用于 Antigravity 特判）
|
||||
|
||||
Returns:
|
||||
包含 input_tokens, output_tokens, cached_tokens 的字典
|
||||
"""
|
||||
from src.core.provider_types import ProviderType
|
||||
|
||||
if str(provider_type or "").lower() == ProviderType.ANTIGRAVITY:
|
||||
return self._extract_antigravity_usage(event)
|
||||
|
||||
from src.api.handlers.gemini.stream_parser import GeminiStreamParser
|
||||
|
||||
usage = GeminiStreamParser().extract_usage(event)
|
||||
|
||||
if not usage:
|
||||
return {
|
||||
"input_tokens": 0,
|
||||
"output_tokens": 0,
|
||||
"cached_tokens": 0,
|
||||
}
|
||||
|
||||
return {
|
||||
"input_tokens": usage.get("input_tokens", 0),
|
||||
"output_tokens": usage.get("output_tokens", 0),
|
||||
"cached_tokens": usage.get("cached_tokens", 0),
|
||||
}
|
||||
|
||||
def _extract_antigravity_usage(self, event: dict[str, Any]) -> dict[str, int]:
|
||||
"""Antigravity 专用 usage 提取（宽松 + 边界保护）。
|
||||
|
||||
Antigravity 的 usageMetadata 可能缺少 totalTokenCount，因此不能依赖
|
||||
GeminiStreamParser.extract_usage 的“totalTokenCount 必须存在”的严格判断。
|
||||
"""
|
||||
usage_metadata = event.get("usageMetadata", {})
|
||||
if not isinstance(usage_metadata, dict) or not usage_metadata:
|
||||
return {"input_tokens": 0, "output_tokens": 0, "cached_tokens": 0}
|
||||
|
||||
def _as_int(v: Any) -> int:
|
||||
try:
|
||||
return int(v or 0)
|
||||
except Exception:
|
||||
return 0
|
||||
|
||||
prompt = _as_int(usage_metadata.get("promptTokenCount"))
|
||||
cached = _as_int(usage_metadata.get("cachedContentTokenCount"))
|
||||
candidates = _as_int(usage_metadata.get("candidatesTokenCount"))
|
||||
thoughts = _as_int(usage_metadata.get("thoughtsTokenCount"))
|
||||
|
||||
return {
|
||||
# 注意：计费层会根据 api_family(GEMINI) 扣除 cache_read_tokens，
|
||||
# 因此这里保持 Gemini 口径：input_tokens=promptTokenCount（含缓存）。
|
||||
"input_tokens": max(0, prompt),
|
||||
"output_tokens": max(0, candidates + thoughts),
|
||||
"cached_tokens": max(0, cached),
|
||||
}
|
||||
|
||||
def _process_event_data(
|
||||
self,
|
||||
ctx: StreamContext,
|
||||
_event_type: str,
|
||||
data: dict[str, Any],
|
||||
) -> None:
|
||||
"""
|
||||
处理 Gemini CLI 格式的流式事件
|
||||
|
||||
Gemini 的流式响应是 JSON 数组格式，每个元素结构如下:
|
||||
{
|
||||
"candidates": [{
|
||||
"content": {"parts": [{"text": "..."}], "role": "model"},
|
||||
"finishReason": "STOP",
|
||||
"safetyRatings": [...]
|
||||
}],
|
||||
"usageMetadata": {
|
||||
"promptTokenCount": 10,
|
||||
"candidatesTokenCount": 20,
|
||||
"totalTokenCount": 30,
|
||||
"cachedContentTokenCount": 5
|
||||
},
|
||||
"modelVersion": "gemini-1.5-pro"
|
||||
}
|
||||
|
||||
注意: Gemini 流解析器会将每个 JSON 对象作为一个"事件"传递
|
||||
event_type 在这里可能为空或是自定义的标记
|
||||
|
||||
跨格式转换时（如 provider=claude:chat），原始事件数据是 Provider 格式而非 Gemini 格式。
|
||||
此时委托基类方法通过 Provider 格式解析器提取 usage。
|
||||
"""
|
||||
# 跨格式转换时：原始事件是 Provider 格式，
|
||||
# 基类 _process_event_data 会自动选择正确的 Provider 解析器提取 usage/text
|
||||
if ctx.provider_api_format and ctx.provider_api_format != ctx.client_api_format:
|
||||
super()._process_event_data(ctx, _event_type, data)
|
||||
return
|
||||
|
||||
# 以下是同格式（gemini:cli / gemini:chat）的处理逻辑
|
||||
|
||||
# 提取候选响应
|
||||
candidates = data.get("candidates", [])
|
||||
if candidates:
|
||||
candidate = candidates[0]
|
||||
content = candidate.get("content", {})
|
||||
|
||||
# 提取文本内容
|
||||
parts = content.get("parts", [])
|
||||
for part in parts:
|
||||
if "text" in part:
|
||||
ctx.append_text(part["text"])
|
||||
|
||||
# 检查结束原因
|
||||
finish_reason = candidate.get("finishReason")
|
||||
if finish_reason in ("STOP", "MAX_TOKENS", "SAFETY", "RECITATION", "OTHER"):
|
||||
ctx.has_completion = True
|
||||
ctx.final_response = data
|
||||
|
||||
# 提取使用量信息（复用 GeminiStreamParser.extract_usage）
|
||||
usage = self._extract_usage_from_event(data, provider_type=ctx.provider_type)
|
||||
if usage["input_tokens"] > 0 or usage["output_tokens"] > 0:
|
||||
ctx.input_tokens = usage["input_tokens"]
|
||||
ctx.output_tokens = usage["output_tokens"]
|
||||
ctx.cached_tokens = usage["cached_tokens"]
|
||||
|
||||
# 提取模型版本作为响应 ID
|
||||
model_version = data.get("modelVersion")
|
||||
if model_version:
|
||||
if not ctx.response_id:
|
||||
ctx.response_id = f"gemini-{model_version}"
|
||||
# 存储到 response_metadata 供 Usage 记录使用
|
||||
ctx.response_metadata["model_version"] = model_version
|
||||
|
||||
# 检查错误
|
||||
if "error" in data:
|
||||
ctx.has_completion = True
|
||||
ctx.final_response = data
|
||||
|
||||
def _extract_response_metadata(
|
||||
self,
|
||||
response: dict[str, Any],
|
||||
) -> dict[str, Any]:
|
||||
"""
|
||||
从 Gemini 响应中提取元数据
|
||||
|
||||
提取 modelVersion 字段，记录实际使用的模型版本。
|
||||
|
||||
Args:
|
||||
response: Gemini API 响应
|
||||
|
||||
Returns:
|
||||
包含 model_version 的元数据字典
|
||||
"""
|
||||
metadata: dict[str, Any] = {}
|
||||
model_version = response.get("modelVersion")
|
||||
if model_version:
|
||||
metadata["model_version"] = model_version
|
||||
return metadata
|
||||
Reference in New Issue
Block a user