v1.0: comprehensive refactor — security, concurrency, real streaming, multimodal, function calling, WebUI rewrite, 51 tests, CI

Major changes
=============

Security
--------
* Bearer Token auth middleware (verify_api_key) for /v1/*
* Admin cookie auth + /webui/login for the management UI
* Drop allow_origins=[*] + credentials=True CORS anti-pattern;
  opt-in via CORS_ORIGINS env var
* WebUI template XSS closed (html.escape on every dynamic value)
* Pydantic v2 strict mode with field validators (account name,
  role whitelist, temperature range, …)

Concurrency / stability
-----------------------
* Global _reqid counter replaced with per-request local base
* AccountPool wrapped in asyncio.Lock; select / record_* /
  release all atomic
* Per-account in-flight cap (max_concurrent) and per-minute
  rate limit (rate_limit_rpm) — LRU candidates skip saturated
  accounts
* Shared process-wide httpx.AsyncClient (no per-request TLS
  handshake)
* Per-stage httpx.Timeout (connect/read/write/pool)
* Exponential-backoff retry on 401/403/429/5xx with stream
  fallback to blocking call
* Background _env_watcher reloads .env on mtime change

Protocol
--------
* Real SSE incremental streaming (httpx.stream + server-side
  chunking with delta events)
* Token usage extracted from inner_data[2]._mtokenCount /
  _ttokenCount / _stokenCount
* Multipart image upload via Gemini UploadFile endpoint;
  supports data: URIs, HTTP(S) URLs, and local file paths
* Function calling: tools / tool_choice passthrough and
  response tool_calls detection
* parse_response picks the first non-empty candidate as the
  primary answer and the second as reasoning_content

Operations
----------
* Structured logger (logger.py) with rotating file handler;
  _SafeLogger adapter scrubs reserved LogRecord keys from
  any 'extra' dict
* SSE endpoint /api/events/stream pushes log events to the
  new /webui/logs page
* AccountPool.save_to_env() writes back atomically; WebUI
  mutations persist immediately
* Token-bucket rate limiter (rate_limit.py) with
  x-forwarded-for / x-real-ip awareness

WebUI
-----
* Full rewrite: dark theme, static/style.css + static/app.js
* New pages: /webui/login, /webui/logs
* Accounts page: inline add form, HAR paste → /api/accounts
  /from-har
* Dashboard: 5s auto-refresh of stats

Tooling
-------
* HAR parser (har_parser.py) simplified: dead code removed,
  reused by config_tool.py (no more duplicate implementation)
* config_tool.py: real-time stdout streaming into the GUI
  log box, port-conflict pre-flight check
* requirements.txt pinned to compatible ranges; pytest +
  pytest-asyncio added
* start.bat: 18000 -> 1800 typo fixed; where python pre-check

Tests
-----
* 51 pytest cases covering parse_response, build_jspb_header,
  build_request_body, account_pool (LRU/random/limit/concurrent
  /persist), auth, rate_limit, har_parser, and the /health,
  /v1/models, and /v1/chat/completions endpoints
* pytest.ini with asyncio_mode=auto
* CI workflow (.github/workflows/ci.yml) runs the full
  suite on Python 3.10/3.11/3.12 + pyflakes lint

Author: gemini2api bot

.env.example

@@ -1,41 +1,58 @@
 # ============================================
-# 服务器配置
+# 服务器配置 (gemini2api v1.0)
 # ============================================
 HOST=0.0.0.0
 PORT=1800
+
+# --- 鉴权 ---
+# 设置一个非占位值即可启用鉴权。留空 / 占位符 = 禁用鉴权(仅本地开发)。
 API_KEY=sk-web2api-placeholder
+# ADMIN_KEY 默认与 API_KEY 相同,可独立设置。
+ADMIN_KEY=
+# CORS 允许的来源(逗号分隔,留空 = 不启用 CORS)。
+# 警告:不要同时使用 * 与 credentials=true,浏览器会拒绝响应。
+CORS_ORIGINS=
+
+# --- 轮询与限流 ---
+# 可选: least-recently-used | round-robin | random | first
 ROTATION_STRATEGY=least-recently-used
 MAX_ERRORS_BEFORE_DISABLE=3
+# 全局每分钟请求数限流(0 = 关闭)
+GLOBAL_RATE_LIMIT_RPM=0
+
+# --- 日志 ---
+# 日志级别: DEBUG | INFO | WARNING | ERROR
+LOG_LEVEL=INFO
+# 日志文件名,设为空字符串可禁用文件日志
+LOG_FILE=gemini_proxy.log
+
+# --- 持久化 ---
+# 设为 0 可禁用 WebUI 自动写回 .env
+PERSIST_ACCOUNTS=1
 
 # ============================================
-# Profiles — 模型配置（不含凭证）
-# 运行 config_tool.py 或通过 WebUI 添加账号
+# Profiles — 模型配置(不含凭证)
+# 通过 WebUI "账号管理"页或 config_tool.py 添加账号
 # ============================================
 PROFILES=gemini-3.5-flash,gemini-3.5-flash-adv,gemini-3.1-pro,gemini-3.1-pro-ext,gemini-3.1-flash-lite,gemini-3.1-flash-lite-adv
 DEFAULT_MODEL=gemini-3.5-flash
 
 MODEL_FAMILY_gemini-3.5-flash=1
 THINKING_MODE_gemini-3.5-flash=1
-
 MODEL_FAMILY_gemini-3.5-flash-adv=1
 THINKING_MODE_gemini-3.5-flash-adv=2
-
 MODEL_FAMILY_gemini-3.1-pro=3
 THINKING_MODE_gemini-3.1-pro=1
-
 MODEL_FAMILY_gemini-3.1-pro-ext=3
 THINKING_MODE_gemini-3.1-pro-ext=2
-
 MODEL_FAMILY_gemini-3.1-flash-lite=6
 THINKING_MODE_gemini-3.1-flash-lite=1
-
 MODEL_FAMILY_gemini-3.1-flash-lite-adv=6
 THINKING_MODE_gemini-3.1-flash-lite-adv=2
 
 # ============================================
 # Accounts — 账号凭证
-# 运行 config_tool.py 或通过 WebUI 添加
-# 格式: ACCOUNT_<name>_<FIELD>=<value>
+# 推荐使用 WebUI 或 config_tool.py 添加,系统会自动写回下方
 # ============================================
 # ACCOUNT_myaccount_F_SID=xxx
 # ACCOUNT_myaccount_AT=xxx
@@ -46,3 +63,5 @@ THINKING_MODE_gemini-3.1-flash-lite-adv=2
 # ACCOUNT_myaccount_HASH=
 # ACCOUNT_myaccount_ENABLED=true
 # ACCOUNT_myaccount_MODELS=gemini-3.5-flash,gemini-3.1-flash-lite
+# ACCOUNT_myaccount_RATE_LIMIT_RPM=60
+# ACCOUNT_myaccount_MAX_CONCURRENT=4

.github/workflows/ci.yml

@@ -0,0 +1,62 @@
+name: CI
+
+on:
+  push:
+    branches: [main, master]
+  pull_request:
+    branches: [main, master]
+  workflow_dispatch:
+
+jobs:
+  test:
+    name: pytest (ubuntu-latest, py${{ matrix.python-version }})
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ["3.10", "3.11", "3.12"]
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+          cache: pip
+          cache-dependency-path: requirements.txt
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install -r requirements.txt
+
+      - name: Verify module imports
+        run: |
+          python -c "import server, adapter, account_pool, auth, rate_limit, logger, har_parser, config_tool; print('all modules import OK')"
+
+      - name: Run tests
+        run: |
+          python -m pytest tests/ -v --tb=short --durations=10
+
+      - name: Collect coverage
+        run: |
+          pip install coverage[toml]==7.6.*
+          coverage run --source=adapter,account_pool,auth,rate_limit,logger,har_parser,server -m pytest tests/ -q
+          coverage report -m --skip-covered
+
+  lint:
+    name: pyflakes
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - run: pip install pyflakes
+      - name: Pyflakes
+        # Best-effort lint; we don't block on warnings, only on syntax errors.
+        run: |
+          python -m pyflakes adapter.py account_pool.py auth.py rate_limit.py logger.py har_parser.py server.py 2>&1 | tee lint.txt
+          echo "lint complete (see lint.txt)"

.gitignore

@@ -9,16 +9,29 @@ __pycache__/
 *.pyc
 *.pyo
 
-# 临时测试文件
-test_*.py
+# 临时分析脚本(用户级,不是仓库的 pytest)
 analyze_*.py
+
+# 日志 / 运行时产物
 server_out.log
 server_log.txt
 crash.log
 crash_err.log
 stdout.log
 stderr.log
 gemini_proxy.log
+*.log
+
+# pytest / coverage 产物
+.pytest_cache/
+.coverage
+coverage.xml
+htmlcov/
+.tox/
+
+# IDE
+.vscode/
+.idea/
 
 # 系统生成
 start_proxy.bat

CHANGELOG.md

@@ -0,0 +1,46 @@
+# Changelog
+
+## [1.0.0] - 2026-06-27
+
+### 🐛 Bug fixes
+- **API_KEY 鉴权缺失** — 新增 `verify_api_key` 中间件保护 `/v1/*`。占位符值(空、`sk-web2api-placeholder` 等)即禁用,生产环境必须设非占位值。
+- **全局 `_reqid` 竞态** — 改为每请求局部 `reqid`,并发请求不再互相覆盖。
+- **账号池无锁** — `AccountPool.select` / `record_*` 全部加 `asyncio.Lock`,并发下 `last_used` / `error_count` / `inflight` 不再失序。
+- **WebUI XSS** — 账号名、profile 名、状态文本经 `html.escape` 注入到模板。
+- **CORS 反模式** — 移除 `allow_origins=["*"]` + `allow_credentials=True`,改为可选 `CORS_ORIGINS` 环境变量。
+- **`parse_response` 取最长错位** — 改为取第一个非空候选(主答案),第二候选进 `reasoning_content`。
+- **`stream_request` 假流式** — 用 `httpx.AsyncClient.stream` 接收 + 服务端分块到 SSE 客户端。
+- **每次新建 httpx client** — 进程级共享 client,省去每次 TLS 握手。
+- **timeout 不分项** — `httpx.Timeout(connect=10, read=90, write=30, pool=10)`,流式 read 180s。
+- **`except: pass` 吞噬异常** — 所有 `except` 改为具体类型,警告可见。
+- **重试无退避** — 5xx / 401 / 403 / 429 加 1s/2s/4s/8s 指数退避。
+- **流式无重试** — 流式路径预选账号失败时回退到 `send_request`。
+- **HAR 解析死代码** — 删 `_find_content_path`、`template_params`、`all_endpoints`。
+- **start.bat 端口 typo** — `18000` → `1800`,并加 `where python` 预检。
+- **Pydantic 别名** — 删 `PydanticBase` 别名,统一 v2 strict 模式。
+- **requirements.txt 版本** — 全部 pin 到兼容区间,新增 `pytest` / `pytest-asyncio`。
+- **`_reqid` 初值 7 位回绕** — 改用 28 位微秒时间窗。
+
+### ✨ New features
+- **Bearer Token 鉴权中间件** — `Authorization: Bearer <key>` 或 `x-api-key` header;admin 路径还支持 cookie。
+- **Admin 登录页** — `/webui/login` 浏览器登录,1 小时 cookie。
+- **真正流式** — 服务端分块 `delta` 事件,OpenAI 协议兼容;附带尾部 `usage` chunk。
+- **Token usage 统计** — 从 `inner_data[2]._mtokenCount/_ttokenCount` 提取,`response.usage` 不再全是 0。
+- **多模态(图片)输入** — `image_url` 支持 `data:` URI、HTTP(S) URL、本地路径;走 Gemini `UploadFile` 端点。
+- **Function calling / Tools** — `tools` / `tool_choice` 字段;响应解析时检测 function_call 转 OpenAI `tool_calls`。
+- **账号池限流** — 每账号 `rate_limit_rpm` + `max_concurrent`,LRU 候选自动排除超限账号。
+- **账号池持久化** — WebUI 增删改后写回 `.env`(原子 rename);后台 `_env_watcher` 每 5s 检测外部编辑并热加载。
+- **结构化 logging** — `logger.py` 统一,DEBUG/INFO/WARNING/ERROR 分级,输出 stderr + `gemini_proxy.log` 滚动文件;`extra` 自动 namespace-prefix。
+- **SSE 实时日志** — `/api/events/stream` 推送日志事件,`/webui/logs` 页面实时显示。
+- **HAR 导入端点** — `/api/accounts/from-har` 接受 HAR 文本,返回解析结果,前端再确认保存。
+- **Token Bucket 限流中间件** — `GLOBAL_RATE_LIMIT_RPM` 控制单 IP 全局速率;支持 `x-forwarded-for`。
+- **WebUI 全面重做** — 统一 `static/style.css` + `static/app.js`,深色主题,响应式,5s 自动刷新统计。
+- **Pydantic v2 严格** — `extra="ignore"`,字段 validator(账号名 `^[A-Za-z0-9_.-]{1,64}$`、role 白名单、温度区间)。
+- **51 个单元/集成测试** — 覆盖 `parse_response`、`build_jspb_header`、`build_request_body`、`account_pool.select/release/限流/持久化`、`auth.verify_*`、`rate_limit`、`har_parser`、`/health`、`/v1/models`、`/v1/chat/completions` 鉴权+快乐路径。
+- **HAR 解析合并** — GUI `config_tool.py` 不再重复实现,统一调用 `har_parser.parse_har`。
+- **GUI 启动 stdout 实时显示** — `subprocess.Popen(stdout=PIPE)` + 线程读取 → Tkinter Text 实时显示;启动前 socket 端口占用预检。
+
+### ⚠️ Notes
+- 协议仍基于逆向工程的 `/_/BardChatUi/data/assistant.lamda.BardFrontendService/StreamGenerate`,Google 改版可能失效。
+- 多模态和 Function calling 是 best-effort;Gemini Web 端不保证稳定支持。
+- 默认 0 账号启动,首次使用必须通过 WebUI 或 `config_tool.py` 导入 HAR 凭证。