diff --git a/.env.example b/.env.example index 7d4c9ab2..c10a7638 100644 --- a/.env.example +++ b/.env.example @@ -4,6 +4,9 @@ # ============================================================ # HTTP 服务端口 +# Note: the standard compose files bind to all interfaces (0.0.0.0). +# Set BIND_HOST=127.0.0.1 in .env to restrict to localhost only. +# BIND_HOST=0.0.0.0 CODEX_PORT=8080 # 监听地址（默认 0.0.0.0，兼容 Docker 端口映射 / 反向代理 / 公网部署） diff --git a/.env.sqlite.example b/.env.sqlite.example index 1276fc7e..b388a0ad 100644 --- a/.env.sqlite.example +++ b/.env.sqlite.example @@ -4,6 +4,9 @@ # ============================================================ # HTTP 服务端口 +# Note: the SQLite compose files bind to 127.0.0.1 by default for security. +# Override with BIND_HOST=0.0.0.0 in .env if you need external access. +# BIND_HOST=127.0.0.1 CODEX_PORT=8080 # 监听地址（默认 0.0.0.0，兼容 Docker 端口映射 / 反向代理 / 公网部署） diff --git a/README.md b/README.md index eb088d46..2c647caf 100644 --- a/README.md +++ b/README.md @@ -23,9 +23,10 @@ Run it as a full **PostgreSQL + Redis** production stack or as a single-containe - - - + + + +
One compatible gatewayOpenAI-style Chat Completions / Responses / Images, Anthropic Messages, prefixless compatibility routes, and native Codex Responses forwarding are all exposed through one service.
Account-pool schedulerSelection is driven by account status, health tier, scheduler score, dynamic concurrency, cooldown recovery, and recent usage so unhealthy accounts are avoided automatically.
Visual admin consoleThe embedded React / Vite dashboard covers account import and testing, API keys, proxy pools, image studio, prompt filtering, usage analytics, operations, scheduler board, and system settings.
Two deployment shapesUse PostgreSQL + Redis for production or SQLite + Memory for lightweight single-node deployments; Docker images, source builds, local development, and the interactive deploy script are ready to use.
Account-pool schedulerSelection is driven by account status, health tier, scheduler score, dynamic concurrency, cooldown recovery, and recent usage so unhealthy accounts are avoided automatically. Supports round_robin and remaining_quota modes, with per-account credit billing flags.
Visual admin consoleThe embedded React / Vite dashboard covers account import and testing, API keys, proxy pools, image studio (text-to-image + image-to-image), prompt filtering, usage analytics, operations, scheduler board, and system settings.
Two deployment shapesUse PostgreSQL + Redis for production or SQLite + Memory for lightweight single-node deployments; Docker images, source builds, local development, and the interactive deploy script are ready to use. SQLite mode binds to 127.0.0.1 by default for security.
Billing and observabilityPer-account 5h/7d windowed USD cost tracking, credit quota support, API key usage tracking, OAuth PKCE token acquisition, prompt filtering, and a usage dashboard with request logs and trend charts.
--- @@ -164,6 +165,7 @@ Notes: - Standard and SQLite modes both read `.env`. - Before switching deployment modes, replace `.env` with the matching example file. - The SQLite lightweight mode runs a single `codex2api` container and stores data at `/data/codex2api.db`. +- **SQLite compose files bind to `127.0.0.1` by default for security.** To expose the SQLite service on all interfaces, set `BIND_HOST=0.0.0.0` in `.env` or override the port binding in the compose file. The standard compose files bind to `0.0.0.0` by default. - The image studio library is stored under `/data/images`; Docker configurations persist `/data`. - `docker compose down` does not delete named volumes by default. Data is removed only by commands such as `docker compose down -v`, `docker volume rm`, or `docker volume prune`. @@ -262,7 +264,7 @@ The standard `.env.example` declares `DATABASE_DRIVER=postgres` and `CACHE_DRIVE Runtime business settings are stored in the database `SystemSettings` table and can be updated from the admin settings page. -Examples include `MaxConcurrency`, `GlobalRPM`, `TestModel`, `TestConcurrency`, `ProxyURL`, `PgMaxConns`, `RedisPoolSize`, `AdminSecret`, and auto-cleanup switches. +Examples include `MaxConcurrency`, `GlobalRPM`, `TestModel`, `TestConcurrency`, `ProxyURL`, `PgMaxConns`, `RedisPoolSize`, `AdminSecret`, `SchedulerMode`, and auto-cleanup switches. Default settings are written automatically on first startup. @@ -284,9 +286,11 @@ Default settings are written automatically on first startup. | `POST /v1/responses` | Responses style endpoint | | `POST /v1/images/generations` | OpenAI Images generation endpoint | | `POST /v1/images/edits` | OpenAI Images edit endpoint | -| `GET /v1/models` | List available models | +| `GET /v1/models` | List available models (includes gpt-5.5, gpt-5.4, gpt-5.4-mini, gpt-5.3-codex, gpt-image-2, etc.) | | `GET /health` | Health check | +> **Pricing**: gpt-5.5 is billed at $5.00/M input and $30.00/M output (standard tier). Priority tier: $12.50/M input, $75.00/M output. Other models follow pricing rules in the billing engine. + See [API.md](docs/API.md) for full request formats, response formats, and error codes. ### Token Upload and Account Management @@ -349,6 +353,27 @@ curl -X POST http://localhost:8080/api/admin/accounts/import \ Import endpoints deduplicate tokens automatically. Existing tokens are not inserted again. +#### OAuth PKCE Authorization + +Codex2API supports acquiring Refresh Tokens through the OAuth PKCE flow, useful when manual token extraction is impractical: + +```bash +# Step 1: Generate an authorization URL +curl -X POST http://localhost:8080/api/admin/oauth/generate-auth-url \ + -H "X-Admin-Key: your-admin-secret" \ + -H "Content-Type: application/json" \ + -d '{}' + +# Step 2: Open the returned auth_url in a browser, complete authorization +# Step 3: Exchange the authorization code for a token (auto-creates account) +curl -X POST http://localhost:8080/api/admin/oauth/exchange-code \ + -H "X-Admin-Key: your-admin-secret" \ + -H "Content-Type: application/json" \ + -d '{"session_id": "...", "code": "...", "state": "..."}' +``` + +See [API.md](docs/API.md) for the full OAuth flow and all admin endpoints. + --- ## Admin Dashboard @@ -361,7 +386,7 @@ Open `/admin/` in a browser. | Accounts | `/admin/accounts` | Import, test, batch actions, scheduler state | | API Keys | `/admin/api-keys` | API key creation, inspection, deletion, and credential management | | Proxies | `/admin/proxies` | Proxy pool management, account proxy assignment, connectivity checks | -| Image Studio | `/admin/images/studio` | Text-to-image, prompt templates, task history, server-side image library | +| Image Studio | `/admin/images/studio` | Text-to-image, image-to-image, prompt templates, task history, server-side image library | | Prompt Filter | `/admin/prompt-filter/overview` | Rules, hit logs, testing, and handling mode configuration | | Usage | `/admin/usage` | Request logs, metric cards, charts, log cleanup | | Operations | `/admin/ops` | Runtime monitoring and system overview | @@ -431,6 +456,24 @@ Observability: - `GET /api/admin/ops/overview` shows runtime and connection pool state. - `/admin/ops/scheduler` provides the scheduler board. +**Scheduler mode** (`scheduler_mode`, via Admin Settings): + +| Mode | Behavior | +| --- | --- | +| `round_robin` (default) | Round-robin across available accounts per health tier, weighted by dispatch score | +| `remaining_quota` | Prioritizes accounts with lower usage percent; round-robin for ties | + +**Credit accounts** (per-account flags): + +When an account has a credit-based billing model instead of a usage-based Free/Pro plan, you can mark it so the scheduler skips usage-window penalties: + +| Field | Type | Effect | +| --- | --- | --- | +| `credit_enabled` | bool | Mark account as credit-based billing | +| `credit_skip_usage_window` | bool | When true, skip 7d/5h usage-window penalties for this account | + +**Windowed USD cost**: The accounts table displays per-account billed cost over two windows -- the past 5 hours and the past 7 days -- aligned with each account's usage reset boundaries. This shows actual spending per account rather than estimated token costs. + --- ## Project Structure diff --git a/README.zh-CN.md b/README.zh-CN.md index 8ad9fa9c..f751df4d 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -23,9 +23,10 @@ - - - + + + +
统一兼容入口同时覆盖 OpenAI 风格 Chat Completions / Responses / Images、Anthropic Messages、无前缀兼容路由和 Codex 原生 Responses 转发，客户端侧少改配置即可接入。
账号池调度核心围绕账号状态、健康层级、调度分、动态并发、冷却恢复和近期用量做选择，自动避开不可用账号，减少单账号打满和反复失败。
可视化管理后台内置 React / Vite 管理台，提供账号导入测试、API Key、代理池、生图、Prompt 检查、用量统计、运维概览、调度看板和系统设置。
两种部署形态生产环境用 PostgreSQL + Redis，单机测试用 SQLite + Memory；Docker 镜像、源码构建、本地开发和一键交互部署脚本都已准备好。
账号池调度核心围绕账号状态、健康层级、调度分、动态并发、冷却恢复和近期用量做选择，自动避开不可用账号，减少单账号打满和反复失败。支持 round_robin 和 remaining_quota 两种调度模式，以及单账号信用计费标记。
可视化管理后台内置 React / Vite 管理台，提供账号导入测试、API Key、代理池、生图（文生图 + 图生图）、Prompt 检查、用量统计、运维概览、调度看板和系统设置。
两种部署形态生产环境用 PostgreSQL + Redis，单机测试用 SQLite + Memory；Docker 镜像、源码构建、本地开发和一键交互部署脚本都已准备好。SQLite 模式默认绑定 127.0.0.1 以提升安全性。
计费与可观测性单账号 5h/7d 窗口化 USD 费用追踪、信用配额支持、API Key 用量追踪、OAuth PKCE 获取 Token、Prompt 过滤，以及含请求日志与趋势图表的用量仪表盘。
--- @@ -209,6 +210,7 @@ docker compose -f docker-compose.sqlite.local.yml logs -f codex2api - SQLite 镜像版容器名：`codex2api-sqlite` - SQLite 本地构建版容器名：`codex2api-sqlite-local` - SQLite 轻量版只启动 `codex2api` 单容器，数据保存在 `/data/codex2api.db` +- **SQLite compose 文件默认绑定 `127.0.0.1`，仅本机可访问。** 如需暴露给外部，请在 `.env` 中设置 `BIND_HOST=0.0.0.0` 或修改 compose 文件中的端口绑定。标准版 compose 文件默认绑定 `0.0.0.0`（所有网络接口）。 - 生图工作台图库默认保存在 `/data/images`，标准版和 SQLite 版 Docker 配置都会持久化 `/data` - `docker compose down` 默认不会删除命名卷；只有 `docker compose down -v`、`docker volume rm` 或 `docker volume prune` 才会删除持久化数据 - 不同部署模式的数据卷彼此隔离；切换 compose 文件后看到空数据，通常是切到了另一组卷，而不是原卷被自动删除 @@ -313,7 +315,7 @@ Vite 会自动代理 `/api` 和 `/health` 到后端，开发时访问 `http://lo 以下参数**保存在数据库 `SystemSettings` 中**，通过管理台设置页面修改： -`MaxConcurrency`、`GlobalRPM`、`TestModel`、`TestConcurrency`、`ProxyURL`、`PgMaxConns`、`RedisPoolSize`、`AdminSecret`、自动清理开关等。 +`MaxConcurrency`、`GlobalRPM`、`TestModel`、`TestConcurrency`、`ProxyURL`、`PgMaxConns`、`RedisPoolSize`、`AdminSecret`、`SchedulerMode`、自动清理开关等。 首次启动时程序会自动写入默认设置。 @@ -335,9 +337,11 @@ Vite 会自动代理 `/api` 和 `/health` 到后端，开发时访问 `http://lo | `POST /v1/responses` | Responses 风格入口 | | `POST /v1/images/generations` | OpenAI Images 生成入口 | | `POST /v1/images/edits` | OpenAI Images 编辑入口 | -| `GET /v1/models` | 返回可用模型列表 | +| `GET /v1/models` | 返回可用模型列表（含 gpt-5.5、gpt-5.4、gpt-5.4-mini、gpt-5.3-codex、gpt-image-2 等） | | `GET /health` | 健康检查 | +> **计费提示**：gpt-5.5 标准 tier 计费为 $5.00/M 输入 / $30.00/M 输出，priority tier 为 $12.50/M 输入 / $75.00/M 输出。其他模型按 billing 引擎规则计费。 + > 完整请求/响应格式、错误码参见 [API 文档](docs/API.md)。 ### Token 上传与账号管理 @@ -400,6 +404,27 @@ curl -X POST http://localhost:8080/api/admin/accounts/import \ > 所有导入接口自动去重，已存在的 Token 不会重复写入。更多管理接口（导出、迁移、OAuth 授权等）参见 [API 文档](docs/API.md)。 +#### OAuth PKCE 授权 + +Codex2API 支持通过 OAuth PKCE 流程获取 Refresh Token，适用于无法手动提取 Token 的场景： + +```bash +# 步骤 1：生成授权 URL +curl -X POST http://localhost:8080/api/admin/oauth/generate-auth-url \ + -H "X-Admin-Key: your-admin-secret" \ + -H "Content-Type: application/json" \ + -d '{}' + +# 步骤 2：在浏览器中打开返回的 auth_url，完成授权 +# 步骤 3：用授权码兑换 Token（自动创建账号） +curl -X POST http://localhost:8080/api/admin/oauth/exchange-code \ + -H "X-Admin-Key: your-admin-secret" \ + -H "Content-Type: application/json" \ + -d '{"session_id": "...", "code": "...", "state": "..."}' +``` + +> 完整 OAuth 流程及所有管理接口参见 [API 文档](docs/API.md)。 + --- ## 管理后台 @@ -412,7 +437,7 @@ curl -X POST http://localhost:8080/api/admin/accounts/import \ | 账号管理 | `/admin/accounts` | 导入、测试、批量处理、调度信息查看 | | API 密钥 | `/admin/api-keys` | API Key 创建、查看、删除与调用凭据管理 | | 代理管理 | `/admin/proxies` | 代理池维护、账号代理分配与连通性管理 | -| 生图工作台 | `/admin/images/studio` | 文生图、提示词模板、任务历史和服务器图库 | +| 生图工作台 | `/admin/images/studio` | 文生图、图生图、提示词模板、任务历史和服务器图库 | | Prompt 检查 | `/admin/prompt-filter/overview` | Prompt 规则、触发日志、测试和处理模式配置 | | 使用统计 | `/admin/usage` | 请求日志、统计卡片、图表、日志清空 | | 运维概览 | `/admin/ops` | 运行态监控与系统概览 | @@ -495,6 +520,24 @@ curl -X POST http://localhost:8080/api/admin/accounts/import \ - `GET /api/admin/ops/overview` — 系统运行态与连接池概览 - `/admin/ops/scheduler` — 前端调度看板 +**调度模式**（`scheduler_mode`，通过管理后台设置）： + +| 模式 | 行为 | +| --- | --- | +| `round_robin`（默认） | 按健康层级轮询可用账号，权重按调度分排序 | +| `remaining_quota` | 优先使用用量较低的账号；用量相同时轮询 | + +**信用账号**（单账号标记）： + +对采用信用计费而非 Free/Pro 用量计费的账号，可标记为信用账号以跳过用量窗口惩罚： + +| 字段 | 类型 | 作用 | +| --- | --- | --- | +| `credit_enabled` | bool | 标记账号为信用计费模式 | +| `credit_skip_usage_window` | bool | 开启后跳过 7 天/5 小时用量窗口惩罚 | + +**窗口化 USD 费用**：账号列表展示每个账号在两个时间窗口内的累计计费金额——过去 5 小时和过去 7 天，窗口对齐各账号的用量重置边界。这反映的是实际扣费金额而非估算的 Token 费用。 + --- ## 目录结构 diff --git a/admin/handler.go b/admin/handler.go index c6bdb835..b076a3f6 100644 --- a/admin/handler.go +++ b/admin/handler.go @@ -203,6 +203,7 @@ func (h *Handler) RegisterRoutes(r *gin.Engine) { api.GET("/accounts/:id/test", h.TestConnection) api.GET("/accounts/:id/usage", h.GetAccountUsage) api.GET("/accounts/:id/auth-json", h.GetAccountAuthJSON) + api.PATCH("/accounts/:id/credit", h.UpdateAccountCredit) api.POST("/accounts/batch-test", h.BatchTest) api.POST("/accounts/batch-reset-status", h.BatchResetStatus) api.POST("/accounts/clean-banned", h.CleanBanned) @@ -242,6 +243,7 @@ func (h *Handler) RegisterRoutes(r *gin.Engine) { api.PATCH("/image-prompts/:id", h.UpdateImagePromptTemplate) api.DELETE("/image-prompts/:id", h.DeleteImagePromptTemplate) api.POST("/images/jobs", h.CreateImageGenerationJob) + api.POST("/images/edit-jobs", h.CreateImageEditJob) api.GET("/images/jobs", h.ListImageGenerationJobs) api.GET("/images/jobs/:id", h.GetImageGenerationJob) api.DELETE("/images/jobs/:id", h.DeleteImageGenerationJob) @@ -381,6 +383,8 @@ type accountResponse struct { Status string `json:"status"` ErrorMessage string `json:"error_message,omitempty"` ATOnly bool `json:"at_only"` + CreditEnabled bool `json:"credit_enabled"` + CreditSkipUsageWindow bool `json:"credit_skip_usage_window"` AccountType string `json:"account_type,omitempty"` OpenAIResponsesAPI bool `json:"openai_responses_api,omitempty"` BaseURL string `json:"base_url,omitempty"` @@ -409,6 +413,8 @@ type accountResponse struct { Usage7dDetail *accountUsageWindow `json:"usage_7d_detail,omitempty"` Reset5hAt string `json:"reset_5h_at,omitempty"` Reset7dAt string `json:"reset_7d_at,omitempty"` + Billed5h *float64 `json:"billed_5h"` + Billed7d *float64 `json:"billed_7d"` ScoreBreakdown schedulerBreakdownResponse `json:"scheduler_breakdown"` LastUnauthorizedAt string `json:"last_unauthorized_at,omitempty"` LastRateLimitedAt string `json:"last_rate_limited_at,omitempty"` @@ -501,6 +507,8 @@ func (h *Handler) ListAccounts(c *gin.Context) { Status: row.Status, ErrorMessage: row.ErrorMessage, ATOnly: !isOpenAIResponsesAccount && row.GetCredential("refresh_token") == "" && row.GetCredential("access_token") != "", + CreditEnabled: row.CreditEnabled, + CreditSkipUsageWindow: row.CreditSkipUsageWindow, AccountType: row.Type, OpenAIResponsesAPI: isOpenAIResponsesAccount, BaseURL: baseURL, @@ -625,6 +633,26 @@ func (h *Handler) ListAccounts(c *gin.Context) { accounts = append(accounts, resp) } + // 批量查询各账号 5h / 7d 窗口内累计 account_billed + for i := range accounts { + acc, ok := accountMap[accounts[i].ID] + if !ok { + continue + } + if t := acc.GetReset5hAt(); !t.IsZero() { + billed, err := h.db.GetAccountBilledSince(ctx, accounts[i].ID, t.Add(-5*time.Hour)) + if err == nil { + accounts[i].Billed5h = &billed + } + } + if t := acc.GetReset7dAt(); !t.IsZero() { + billed, err := h.db.GetAccountBilledSince(ctx, accounts[i].ID, t.AddDate(0, 0, -7)) + if err == nil { + accounts[i].Billed7d = &billed + } + } + } + c.JSON(http.StatusOK, accountsResponse{Accounts: accounts}) } @@ -638,6 +666,45 @@ type updateAccountSchedulerReq struct { } // UpdateAccountScheduler 更新账号调度配置。 +// UpdateAccountCredit 更新账号信用设置 +func (h *Handler) UpdateAccountCredit(c *gin.Context) { + id, err := strconv.ParseInt(c.Param("id"), 10, 64) + if err != nil { + writeError(c, http.StatusBadRequest, "无效的账号 ID") + return + } + + var req struct { + CreditEnabled *bool `json:"credit_enabled"` + CreditSkipUsageWindow *bool `json:"credit_skip_usage_window"` + } + decoder := json.NewDecoder(c.Request.Body) + decoder.DisallowUnknownFields() + if err := decoder.Decode(&req); err != nil { + writeError(c, http.StatusBadRequest, "请求格式错误") + return + } + + acc := h.store.FindByID(id) + if acc == nil { + writeError(c, http.StatusNotFound, "账号不存在") + return + } + + // 传入 *bool：nil = 不修改该字段 + if err := h.store.UpdateAccountCredit(id, req.CreditEnabled, req.CreditSkipUsageWindow); err != nil { + writeError(c, http.StatusInternalServerError, "更新信用设置失败: "+err.Error()) + return + } + + acc = h.store.FindByID(id) + if acc != nil { + c.JSON(http.StatusOK, gin.H{"message": "信用设置已更新", "credit_enabled": acc.CreditEnabled, "credit_skip_usage_window": acc.CreditSkipUsageWindow}) + } else { + c.JSON(http.StatusOK, gin.H{"message": "信用设置已更新"}) + } +} + func (h *Handler) UpdateAccountScheduler(c *gin.Context) { id, err := strconv.ParseInt(c.Param("id"), 10, 64) if err != nil { @@ -3496,6 +3563,7 @@ type settingsResponse struct { AutoCleanExpired bool `json:"auto_clean_expired"` ProxyPoolEnabled bool `json:"proxy_pool_enabled"` FastSchedulerEnabled bool `json:"fast_scheduler_enabled"` + SchedulerMode string `json:"scheduler_mode"` MaxRetries int `json:"max_retries"` MaxRateLimitRetries int `json:"max_rate_limit_retries"` AllowRemoteMigration bool `json:"allow_remote_migration"` @@ -3554,6 +3622,7 @@ type updateSettingsReq struct { AutoCleanExpired *bool `json:"auto_clean_expired"` ProxyPoolEnabled *bool `json:"proxy_pool_enabled"` FastSchedulerEnabled *bool `json:"fast_scheduler_enabled"` + SchedulerMode *string `json:"scheduler_mode"` MaxRetries *int `json:"max_retries"` MaxRateLimitRetries *int `json:"max_rate_limit_retries"` AllowRemoteMigration *bool `json:"allow_remote_migration"` @@ -3694,6 +3763,7 @@ func (h *Handler) GetSettings(c *gin.Context) { AutoCleanExpired: h.store.GetAutoCleanExpired(), ProxyPoolEnabled: h.store.GetProxyPoolEnabled(), FastSchedulerEnabled: h.store.FastSchedulerEnabled(), + SchedulerMode: h.store.GetSchedulerMode(), MaxRetries: h.store.GetMaxRetries(), MaxRateLimitRetries: h.store.GetMaxRateLimitRetries(), AllowRemoteMigration: h.store.GetAllowRemoteMigration() && adminAuthSource != "disabled", @@ -3926,6 +3996,11 @@ func (h *Handler) UpdateSettings(c *gin.Context) { log.Printf("设置已更新: fast_scheduler_enabled = %t", *req.FastSchedulerEnabled) } + if req.SchedulerMode != nil { + h.store.SetSchedulerMode(*req.SchedulerMode) + log.Printf("设置已更新: scheduler_mode = %s", *req.SchedulerMode) + } + if req.MaxRetries != nil { v := *req.MaxRetries if v < 0 { @@ -4167,6 +4242,7 @@ func (h *Handler) UpdateSettings(c *gin.Context) { AutoCleanExpired: h.store.GetAutoCleanExpired(), ProxyPoolEnabled: h.store.GetProxyPoolEnabled(), FastSchedulerEnabled: h.store.FastSchedulerEnabled(), + SchedulerMode: h.store.GetSchedulerMode(), MaxRetries: h.store.GetMaxRetries(), MaxRateLimitRetries: h.store.GetMaxRateLimitRetries(), AllowRemoteMigration: h.store.GetAllowRemoteMigration() && hasAdminSecret, @@ -4230,6 +4306,7 @@ func (h *Handler) UpdateSettings(c *gin.Context) { AutoCleanExpired: h.store.GetAutoCleanExpired(), ProxyPoolEnabled: h.store.GetProxyPoolEnabled(), FastSchedulerEnabled: h.store.FastSchedulerEnabled(), + SchedulerMode: h.store.GetSchedulerMode(), MaxRetries: h.store.GetMaxRetries(), MaxRateLimitRetries: h.store.GetMaxRateLimitRetries(), AllowRemoteMigration: h.store.GetAllowRemoteMigration() && adminAuthSource != "disabled", diff --git a/admin/image_studio.go b/admin/image_studio.go index 47186da1..dcd6912d 100644 --- a/admin/image_studio.go +++ b/admin/image_studio.go @@ -63,7 +63,8 @@ type imageGenerationJobPayload struct { Style string `json:"style"` Upscale string `json:"upscale"` APIKeyID int64 `json:"api_key_id"` - TemplateID int64 `json:"template_id"` + TemplateID int64 `json:"template_id"` + InputImages []string `json:"input_images"` } type imageJobResponse struct { @@ -322,6 +323,89 @@ func (h *Handler) CreateImageGenerationJob(c *gin.Context) { c.JSON(http.StatusOK, imageJobResponse{Job: job}) } +func (h *Handler) CreateImageEditJob(c *gin.Context) { + var req imageGenerationJobPayload + if err := c.ShouldBindJSON(&req); err != nil { + writeError(c, http.StatusBadRequest, "请求体无效") + return + } + req.Prompt = strings.TrimSpace(req.Prompt) + if req.Prompt == "" { + writeError(c, http.StatusBadRequest, "提示词不能为空") + return + } + if len(req.InputImages) == 0 { + writeError(c, http.StatusBadRequest, "图生图需要上传参考图片") + return + } + if len(req.InputImages) > proxy.MaxImageEditInputCount { + writeError(c, http.StatusBadRequest, fmt.Sprintf("参考图片数量超过限制 (%d, 最多 %d)", len(req.InputImages), proxy.MaxImageEditInputCount)) + return + } + if len([]rune(req.Prompt)) > 8000 { + writeError(c, http.StatusBadRequest, "提示词不能超过 8000 个字符") + return + } + req.Model = normalizeImageStudioModel(req.Model) + if req.Model == "" { + req.Model = "gpt-image-2" + } + req.Size = normalizeOptionalImageParam(req.Size) + req.Quality = normalizeOptionalImageParam(req.Quality) + req.OutputFormat = normalizeOptionalImageParam(req.OutputFormat) + if req.OutputFormat == "" { + req.OutputFormat = "png" + } + req.Background = normalizeOptionalImageParam(req.Background) + req.Style = normalizeOptionalImageParam(req.Style) + req.Upscale = imageproc.NormalizeUpscale(req.Upscale) + + ctx, cancel := context.WithTimeout(c.Request.Context(), 10*time.Second) + defer cancel() + apiKey, err := h.resolveImageJobAPIKey(ctx, req.APIKeyID) + if err != nil { + writeError(c, http.StatusBadRequest, err.Error()) + return + } + paramsJSON, _ := json.Marshal(req) + keyID, keyName, keyMasked := imageJobAPIKeyMeta(apiKey) + if h.inspectImageStudioPromptFilter(c, proxy.AppendImageStyleToPrompt(req.Prompt, req.Style), req.Model, keyID, keyName, keyMasked) { + return + } + jobID, err := h.db.InsertImageGenerationJob(ctx, database.ImageGenerationJobInput{ + Prompt: req.Prompt, + ParamsJSON: string(paramsJSON), + APIKeyID: keyID, + APIKeyName: keyName, + APIKeyMasked: keyMasked, + }) + if err != nil { + writeInternalError(c, err) + return + } + if req.TemplateID > 0 { + _ = h.db.IncrementImagePromptTemplateUsage(ctx, req.TemplateID) + } + job, err := h.db.GetImageGenerationJob(ctx, jobID) + if err != nil { + writeInternalError(c, err) + return + } + log.Printf("[image-studio] job=%d queued mode=edit model=%s size=%s quality=%s format=%s image_count=%d api_key=%s template=%d prompt=%q", + jobID, + imageLogValue(req.Model), + imageLogValue(req.Size), + imageLogValue(req.Quality), + imageLogValue(req.OutputFormat), + len(req.InputImages), + imageLogAPIKeyLabel(keyID, keyName, keyMasked), + req.TemplateID, + imageLogPromptPreview(req.Prompt), + ) + go h.runImageEditJob(jobID, req, apiKey) + c.JSON(http.StatusOK, imageJobResponse{Job: job}) +} + func (h *Handler) ListImageGenerationJobs(c *gin.Context) { page, pageSize := paginationParams(c, 20) ctx, cancel := context.WithTimeout(c.Request.Context(), 10*time.Second) @@ -689,7 +773,7 @@ func (h *Handler) runImageGenerationJob(jobID int64, req imageGenerationJobPaylo if err != nil { durationMs := int(time.Since(start).Milliseconds()) log.Printf("[image-studio] job=%d failed duration=%s stage=build_request error=%s", jobID, imageLogDuration(durationMs), security.SanitizeLog(err.Error())) - _ = h.db.MarkImageJobFailed(context.Background(), jobID, err.Error(), durationMs) + _ = h.db.MarkImageJobFailed(ctx, jobID, err.Error(), durationMs) return } log.Printf("[image-studio] job=%d upstream request model=%s size=%s quality=%s format=%s body_bytes=%d prompt_chars=%d prompt=%q", @@ -715,7 +799,7 @@ func (h *Handler) runImageGenerationJob(jobID int64, req imageGenerationJobPaylo if buildErr != nil { durationMs := int(time.Since(start).Milliseconds()) log.Printf("[image-studio] job=%d failed duration=%s stage=build_jpeg_fallback error=%s", jobID, imageLogDuration(durationMs), security.SanitizeLog(buildErr.Error())) - _ = h.db.MarkImageJobFailed(context.Background(), jobID, buildErr.Error(), durationMs) + _ = h.db.MarkImageJobFailed(ctx, jobID, buildErr.Error(), durationMs) return } fallbackStyledPrompt := proxy.AppendImageStyleToPrompt(fallbackReq.Prompt, fallbackReq.Style) @@ -734,7 +818,7 @@ func (h *Handler) runImageGenerationJob(jobID int64, req imageGenerationJobPaylo req = fallbackReq rawBody = fallbackBody if paramsJSON, marshalErr := json.Marshal(fallbackReq); marshalErr == nil { - if updateErr := h.db.UpdateImageGenerationJobParamsJSON(context.Background(), jobID, string(paramsJSON)); updateErr != nil { + if updateErr := h.db.UpdateImageGenerationJobParamsJSON(ctx, jobID, string(paramsJSON)); updateErr != nil { logImageJobError(jobID, updateErr) } } @@ -756,7 +840,7 @@ func (h *Handler) runImageGenerationJob(jobID int64, req imageGenerationJobPaylo upstreamStatus, security.SanitizeLog(err.Error()), ) - _ = h.db.MarkImageJobFailed(context.Background(), jobID, err.Error(), durationMs) + _ = h.db.MarkImageJobFailed(ctx, jobID, err.Error(), durationMs) return } log.Printf("[image-studio] job=%d upstream completed duration=%s upstream_status=%d response_bytes=%d", @@ -766,18 +850,18 @@ func (h *Handler) runImageGenerationJob(jobID int64, req imageGenerationJobPaylo len(responseJSON), ) - assets, err := h.saveImageJobAssets(context.Background(), jobID, req, responseJSON) + assets, err := h.saveImageJobAssets(ctx, jobID, req, responseJSON) if err != nil { log.Printf("[image-studio] job=%d failed duration=%s stage=save_assets error=%s", jobID, imageLogDuration(durationMs), security.SanitizeLog(err.Error())) - _ = h.db.MarkImageJobFailed(context.Background(), jobID, err.Error(), durationMs) + _ = h.db.MarkImageJobFailed(ctx, jobID, err.Error(), durationMs) return } if len(assets) == 0 { log.Printf("[image-studio] job=%d failed duration=%s stage=save_assets error=%s", jobID, imageLogDuration(durationMs), "上游未返回图片") - _ = h.db.MarkImageJobFailed(context.Background(), jobID, "上游未返回图片", durationMs) + _ = h.db.MarkImageJobFailed(ctx, jobID, "上游未返回图片", durationMs) return } - if err := h.db.MarkImageJobSucceeded(context.Background(), jobID, durationMs); err != nil { + if err := h.db.MarkImageJobSucceeded(ctx, jobID, durationMs); err != nil { logImageJobError(jobID, err) } log.Printf("[image-studio] job=%d succeeded duration=%s assets=%d total_bytes=%d first_size=%s dir=%s", @@ -790,6 +874,142 @@ func (h *Handler) runImageGenerationJob(jobID int64, req imageGenerationJobPaylo ) } +func buildAdminImageEditRequest(req imageGenerationJobPayload) ([]byte, error) { + if len(req.InputImages) == 0 { + return nil, fmt.Errorf("input_images is required for image edit") + } + if len(req.InputImages) > proxy.MaxImageEditInputCount { + return nil, fmt.Errorf("too many input images (%d, max %d)", len(req.InputImages), proxy.MaxImageEditInputCount) + } + body := map[string]any{ + "model": req.Model, + "prompt": proxy.AppendImageStyleToPrompt(req.Prompt, req.Style), + "response_format": "b64_json", + } + images := make([]map[string]string, len(req.InputImages)) + for i, img := range req.InputImages { + images[i] = map[string]string{"image_url": img} + } + body["images"] = images + if req.Size != "" && req.Size != "auto" { + body["size"] = req.Size + } + if req.Quality != "" && req.Quality != "auto" { + body["quality"] = req.Quality + } + if req.OutputFormat != "" { + body["output_format"] = req.OutputFormat + } + if req.Background != "" && req.Background != "auto" { + body["background"] = req.Background + } + return json.Marshal(body) +} + +func (h *Handler) runImageEditJob(jobID int64, req imageGenerationJobPayload, apiKey *database.APIKeyRow) { + ctx, cancel := context.WithTimeout(context.Background(), 12*time.Minute) + defer cancel() + start := time.Now() + if err := h.db.MarkImageJobRunning(ctx, jobID); err != nil { + logImageJobError(jobID, err) + return + } + log.Printf("[image-studio] job=%d started mode=edit model=%s image_count=%d prompt_chars=%d", + jobID, + imageLogValue(req.Model), + len(req.InputImages), + len([]rune(req.Prompt)), + ) + + rawBody, err := buildAdminImageEditRequest(req) + if err != nil { + durationMs := int(time.Since(start).Milliseconds()) + log.Printf("[image-studio] job=%d failed mode=edit duration=%s stage=build_request error=%s", jobID, imageLogDuration(durationMs), security.SanitizeLog(err.Error())) + _ = h.db.MarkImageJobFailed(ctx, jobID, err.Error(), durationMs) + return + } + + imageProxy := h.imageProxy + if imageProxy == nil { + imageProxy = proxy.NewHandler(h.store, h.db, nil, nil) + } + responseJSON, upstreamStatus, err := imageProxy.GenerateImageEditForAdmin(ctx, rawBody, apiKey) + if shouldFallbackImageJobToJPEG(req, upstreamStatus, err) { + pngErr := err + pngStatus := upstreamStatus + fallbackReq := jpegFallbackImageJobRequest(req) + fallbackBody, buildErr := buildAdminImageEditRequest(fallbackReq) + if buildErr != nil { + durationMs := int(time.Since(start).Milliseconds()) + log.Printf("[image-studio] job=%d failed mode=edit duration=%s stage=build_jpeg_fallback error=%s", jobID, imageLogDuration(durationMs), security.SanitizeLog(buildErr.Error())) + _ = h.db.MarkImageJobFailed(ctx, jobID, buildErr.Error(), durationMs) + return + } + log.Printf("[image-studio] job=%d png_failed_retrying_jpeg mode=edit upstream_status=%d error=%s", + jobID, + pngStatus, + security.SanitizeLog(pngErr.Error()), + ) + responseJSON, upstreamStatus, err = imageProxy.GenerateImageEditForAdmin(ctx, fallbackBody, apiKey) + if err == nil { + req = fallbackReq + rawBody = fallbackBody + if paramsJSON, marshalErr := json.Marshal(fallbackReq); marshalErr == nil { + if updateErr := h.db.UpdateImageGenerationJobParamsJSON(ctx, jobID, string(paramsJSON)); updateErr != nil { + logImageJobError(jobID, updateErr) + } + } + log.Printf("[image-studio] job=%d jpeg_fallback_succeeded mode=edit original_status=%d upstream_status=%d body_bytes=%d", + jobID, + pngStatus, + upstreamStatus, + len(rawBody), + ) + } else { + err = fmt.Errorf("PNG 生成失败: %s；自动改用 JPEG 重试仍失败: %w", pngErr.Error(), err) + } + } + durationMs := int(time.Since(start).Milliseconds()) + if err != nil { + log.Printf("[image-studio] job=%d failed mode=edit duration=%s upstream_status=%d error=%s", + jobID, + imageLogDuration(durationMs), + upstreamStatus, + security.SanitizeLog(err.Error()), + ) + _ = h.db.MarkImageJobFailed(ctx, jobID, err.Error(), durationMs) + return + } + log.Printf("[image-studio] job=%d upstream completed mode=edit duration=%s upstream_status=%d response_bytes=%d", + jobID, + imageLogDuration(durationMs), + upstreamStatus, + len(responseJSON), + ) + + assets, err := h.saveImageJobAssets(ctx, jobID, req, responseJSON) + if err != nil { + log.Printf("[image-studio] job=%d failed mode=edit duration=%s stage=save_assets error=%s", jobID, imageLogDuration(durationMs), security.SanitizeLog(err.Error())) + _ = h.db.MarkImageJobFailed(ctx, jobID, err.Error(), durationMs) + return + } + if len(assets) == 0 { + log.Printf("[image-studio] job=%d failed mode=edit duration=%s stage=save_assets error=%s", jobID, imageLogDuration(durationMs), "上游未返回图片") + _ = h.db.MarkImageJobFailed(ctx, jobID, "上游未返回图片", durationMs) + return + } + if err := h.db.MarkImageJobSucceeded(ctx, jobID, durationMs); err != nil { + logImageJobError(jobID, err) + } + log.Printf("[image-studio] job=%d succeeded mode=edit duration=%s assets=%d total_bytes=%d first_size=%s", + jobID, + imageLogDuration(durationMs), + len(assets), + imageAssetsTotalBytes(assets), + imageLogFirstAssetSize(assets), + ) +} + func buildAdminImageGenerationRequest(req imageGenerationJobPayload) ([]byte, error) { body := map[string]any{ "model": req.Model, diff --git a/admin/oauth.go b/admin/oauth.go index c5cf25e0..4df3532b 100644 --- a/admin/oauth.go +++ b/admin/oauth.go @@ -224,6 +224,9 @@ func (h *Handler) ExchangeOAuthCode(c *gin.Context) { if trimmed := strings.TrimSpace(req.ProxyURL); trimmed != "" { proxyURL = trimmed } + if proxyURL == "" { + proxyURL = h.store.GetProxyURL() + } // Resin 临时身份用于 OAuth 兑换（新账号尚无 DBID） resinTempID := "oauth-" + req.SessionID @@ -391,8 +394,12 @@ func (h *Handler) OAuthCallback(c *gin.Context) { sess.CallbackAt = time.Now() // 执行 code exchange（Resin 临时身份） + proxyURL := sess.ProxyURL + if proxyURL == "" { + proxyURL = h.store.GetProxyURL() + } resinTempID := "oauth-" + sessionID - tokenResp, accountInfo, err := doOAuthCodeExchange(c.Request.Context(), code, sess.CodeVerifier, sess.RedirectURI, sess.ProxyURL, resinTempID) + tokenResp, accountInfo, err := doOAuthCodeExchange(c.Request.Context(), code, sess.CodeVerifier, sess.RedirectURI, proxyURL, resinTempID) if err != nil { sess.ExchangeResult = &oauthExchangeResult{ Success: false, @@ -429,7 +436,7 @@ func (h *Handler) OAuthCallback(c *gin.Context) { ctx, cancel := context.WithTimeout(c.Request.Context(), 30*time.Second) defer cancel() - id, err := h.db.InsertAccount(ctx, name, tokenResp.RefreshToken, sess.ProxyURL) + id, err := h.db.InsertAccount(ctx, name, tokenResp.RefreshToken, proxyURL) if err != nil { sess.ExchangeResult = &oauthExchangeResult{ Success: false, @@ -453,7 +460,7 @@ func (h *Handler) OAuthCallback(c *gin.Context) { go proxy.InheritLease(resinTempID, fmt.Sprintf("%d", id)) } - newAcc := accountFromCredentialSeed(id, sess.ProxyURL, seed) + newAcc := accountFromCredentialSeed(id, proxyURL, seed) h.store.AddAccount(newAcc) email := "" diff --git a/auth/fast_scheduler.go b/auth/fast_scheduler.go index 992644c8..76a0094b 100644 --- a/auth/fast_scheduler.go +++ b/auth/fast_scheduler.go @@ -33,26 +33,31 @@ type fastSchedulerPosition struct { // 调度策略：按健康层级分桶，桶内按调度分排序后 round-robin。 // 验证过的账号只作为同分 tie-breaker，避免历史请求量盖过额度快重置优先级。 type FastScheduler struct { - mu sync.RWMutex - baseLimit int64 - buckets map[AccountHealthTier][]fastSchedulerEntry - positions map[int64]fastSchedulerPosition - cursors [3]atomic.Uint64 - groupCheck func(apiKeyID int64, account *Account) bool + mu sync.RWMutex + baseLimit int64 + schedulerMode string + buckets map[AccountHealthTier][]fastSchedulerEntry + positions map[int64]fastSchedulerPosition + cursors [3]atomic.Uint64 + groupCheck func(apiKeyID int64, account *Account) bool } -func NewFastScheduler(baseLimit int64) *FastScheduler { +func NewFastScheduler(baseLimit int64, schedulerMode string) *FastScheduler { if baseLimit <= 0 { baseLimit = 1 } + if schedulerMode == "" { + schedulerMode = "round_robin" + } return &FastScheduler{ - baseLimit: baseLimit, + baseLimit: baseLimit, + schedulerMode: schedulerMode, buckets: map[AccountHealthTier][]fastSchedulerEntry{ HealthTierHealthy: nil, HealthTierWarm: nil, HealthTierRisky: nil, }, - positions: make(map[int64]fastSchedulerPosition), + positions: map[int64]fastSchedulerPosition{}, } } @@ -65,13 +70,67 @@ func (s *FastScheduler) SetGroupCheck(check func(apiKeyID int64, account *Accoun s.mu.Unlock() } +func (s *FastScheduler) SetSchedulerMode(mode string) { + if s == nil { + return + } + s.mu.Lock() + defer s.mu.Unlock() + if mode == "" { + mode = "round_robin" + } + s.schedulerMode = mode + + // Re-sort all tier buckets according to the new mode. + for _, tier := range fastSchedulerTierOrder { + entries := s.buckets[tier] + if len(entries) == 0 { + continue + } + if mode == "remaining_quota" { + sort.SliceStable(entries, func(i, j int) bool { + usageI := entries[i].acc.usagePercentForScheduling() + usageJ := entries[j].acc.usagePercentForScheduling() + if usageI == usageJ { + if entries[i].proven != entries[j].proven { + return entries[i].proven + } + return entries[i].dbID < entries[j].dbID + } + return usageI < usageJ + }) + } else { + sort.SliceStable(entries, func(i, j int) bool { + if entries[i].dispatchScore == entries[j].dispatchScore { + if entries[i].proven != entries[j].proven { + return entries[i].proven + } + return entries[i].dbID < entries[j].dbID + } + return entries[i].dispatchScore > entries[j].dispatchScore + }) + } + s.buckets[tier] = entries + s.rebuildPositionsLocked(tier) + } +} + +func (s *FastScheduler) SchedulerMode() string { + if s == nil { + return "round_robin" + } + s.mu.RLock() + defer s.mu.RUnlock() + return s.schedulerMode +} + // BuildFastScheduler 用当前 Store 快照构建一个独立 scheduler。 // 该方法不会影响现有生产流量路径，只用于 POC/benchmark/灰度验证。 func (s *Store) BuildFastScheduler() *FastScheduler { if s == nil { - return NewFastScheduler(1) + return NewFastScheduler(1, "round_robin") } - scheduler := NewFastScheduler(atomic.LoadInt64(&s.maxConcurrency)) + scheduler := NewFastScheduler(atomic.LoadInt64(&s.maxConcurrency), s.GetSchedulerMode()) s.mu.RLock() accounts := make([]*Account, len(s.accounts)) @@ -124,15 +183,29 @@ func (s *FastScheduler) Rebuild(accounts []*Account) { if len(entries) == 0 { continue } - sort.SliceStable(entries, func(i, j int) bool { - if entries[i].dispatchScore == entries[j].dispatchScore { - if entries[i].proven != entries[j].proven { - return entries[i].proven + if s.schedulerMode == "remaining_quota" { + sort.SliceStable(entries, func(i, j int) bool { + usageI := entries[i].acc.usagePercentForScheduling() + usageJ := entries[j].acc.usagePercentForScheduling() + if usageI == usageJ { + if entries[i].proven != entries[j].proven { + return entries[i].proven + } + return entries[i].dbID < entries[j].dbID } - return entries[i].dbID < entries[j].dbID - } - return entries[i].dispatchScore > entries[j].dispatchScore - }) + return usageI < usageJ + }) + } else { + sort.SliceStable(entries, func(i, j int) bool { + if entries[i].dispatchScore == entries[j].dispatchScore { + if entries[i].proven != entries[j].proven { + return entries[i].proven + } + return entries[i].dbID < entries[j].dbID + } + return entries[i].dispatchScore > entries[j].dispatchScore + }) + } s.buckets[tier] = entries s.rebuildPositionsLocked(tier) } @@ -193,6 +266,7 @@ func (s *FastScheduler) AcquireExcludingWithFilter(apiKeyID int64, exclude map[i defer s.mu.Unlock() baseLimit := s.baseLimit + var zeroCursor atomic.Uint64 for { changed := false for tierIdx, tier := range fastSchedulerTierOrder { @@ -201,7 +275,12 @@ func (s *FastScheduler) AcquireExcludingWithFilter(apiKeyID int64, exclude map[i continue } - acc, stale := s.scanRangeLocked(tier, 0, len(bucket), &s.cursors[tierIdx], baseLimit, now, apiKeyID, exclude, filter) + cursor := &s.cursors[tierIdx] + if s.schedulerMode == "remaining_quota" { + zeroCursor.Store(0) + cursor = &zeroCursor + } + acc, stale := s.scanRangeLocked(tier, 0, len(bucket), cursor, baseLimit, now, apiKeyID, exclude, filter) if acc != nil { return acc } @@ -301,15 +380,29 @@ func (s *FastScheduler) insertLocked(acc *Account, now time.Time) { dispatchScore: dispatchScore, proven: proven, }) - sort.SliceStable(entries, func(i, j int) bool { - if entries[i].dispatchScore == entries[j].dispatchScore { - if entries[i].proven != entries[j].proven { - return entries[i].proven + if s.schedulerMode == "remaining_quota" { + sort.SliceStable(entries, func(i, j int) bool { + usageI := entries[i].acc.usagePercentForScheduling() + usageJ := entries[j].acc.usagePercentForScheduling() + if usageI == usageJ { + if entries[i].proven != entries[j].proven { + return entries[i].proven + } + return entries[i].dbID < entries[j].dbID } - return entries[i].dbID < entries[j].dbID - } - return entries[i].dispatchScore > entries[j].dispatchScore - }) + return usageI < usageJ + }) + } else { + sort.SliceStable(entries, func(i, j int) bool { + if entries[i].dispatchScore == entries[j].dispatchScore { + if entries[i].proven != entries[j].proven { + return entries[i].proven + } + return entries[i].dbID < entries[j].dbID + } + return entries[i].dispatchScore > entries[j].dispatchScore + }) + } s.buckets[tier] = entries s.rebuildPositionsLocked(tier) } diff --git a/auth/fast_scheduler_test.go b/auth/fast_scheduler_test.go index 915f5f2d..440b5ee6 100644 --- a/auth/fast_scheduler_test.go +++ b/auth/fast_scheduler_test.go @@ -24,7 +24,7 @@ func TestFastSchedulerAcquirePrefersHealthyTier(t *testing.T) { warm := newFastSchedulerTestAccount(1, HealthTierWarm, 90, 2) healthy := newFastSchedulerTestAccount(2, HealthTierHealthy, 80, 2) - scheduler := NewFastScheduler(2) + scheduler := NewFastScheduler(2, "round_robin") scheduler.Rebuild([]*Account{warm, healthy}) got := scheduler.Acquire() @@ -43,7 +43,7 @@ func TestFastSchedulerSkipsDispatchPausedAccount(t *testing.T) { atomic.StoreInt32(&paused.DispatchPaused, 1) fallback := newFastSchedulerTestAccount(2, HealthTierHealthy, 80, 2) - scheduler := NewFastScheduler(2) + scheduler := NewFastScheduler(2, "round_robin") scheduler.Rebuild([]*Account{paused, fallback}) got := scheduler.Acquire() @@ -62,7 +62,7 @@ func TestFastSchedulerSkipsErrorAccount(t *testing.T) { errored.Status = StatusError fallback := newFastSchedulerTestAccount(2, HealthTierHealthy, 80, 2) - scheduler := NewFastScheduler(2) + scheduler := NewFastScheduler(2, "round_robin") scheduler.Rebuild([]*Account{errored, fallback}) got := scheduler.Acquire() @@ -79,7 +79,7 @@ func TestFastSchedulerSkipsErrorAccount(t *testing.T) { func TestFastSchedulerRespectsConcurrencyLimit(t *testing.T) { acc := newFastSchedulerTestAccount(1, HealthTierHealthy, 100, 1) - scheduler := NewFastScheduler(1) + scheduler := NewFastScheduler(1, "round_robin") scheduler.Rebuild([]*Account{acc}) first := scheduler.Acquire() @@ -105,7 +105,7 @@ func TestFastSchedulerRoundRobinWithinTier(t *testing.T) { a2 := newFastSchedulerTestAccount(2, HealthTierHealthy, 100, 4) a3 := newFastSchedulerTestAccount(3, HealthTierHealthy, 100, 4) - scheduler := NewFastScheduler(4) + scheduler := NewFastScheduler(4, "round_robin") scheduler.Rebuild([]*Account{a1, a2, a3}) var got []int64 @@ -287,7 +287,7 @@ func TestFastSchedulerAcquireExcludingRespectsAPIKeyWhitelist(t *testing.T) { restricted.SetAllowedAPIKeyIDs([]int64{2}) fallback := newFastSchedulerTestAccount(2, HealthTierHealthy, 80, 1) - scheduler := NewFastScheduler(1) + scheduler := NewFastScheduler(1, "round_robin") scheduler.Rebuild([]*Account{restricted, fallback}) got := scheduler.AcquireExcluding(1, nil) @@ -307,7 +307,7 @@ func TestFastSchedulerAcquireExcludingWithFilterRespectsPlanFilter(t *testing.T) pro := newFastSchedulerTestAccount(2, HealthTierHealthy, 80, 1) pro.PlanType = "pro" - scheduler := NewFastScheduler(1) + scheduler := NewFastScheduler(1, "round_robin") scheduler.Rebuild([]*Account{plus, pro}) got := scheduler.AcquireExcludingWithFilter(0, nil, func(acc *Account) bool { @@ -325,7 +325,7 @@ func TestFastSchedulerAcquireExcludingWithFilterRespectsPlanFilter(t *testing.T) func TestFastSchedulerUpdateMovesAccountBetweenBuckets(t *testing.T) { acc := newFastSchedulerTestAccount(1, HealthTierHealthy, 100, 2) - scheduler := NewFastScheduler(2) + scheduler := NewFastScheduler(2, "round_robin") scheduler.Rebuild([]*Account{acc}) sizes := scheduler.BucketSizes() @@ -358,7 +358,7 @@ func TestFastSchedulerUpdateMovesAccountBetweenBuckets(t *testing.T) { func TestFastSchedulerSkipsStaleBucketEntryWithoutUpdate(t *testing.T) { acc := newFastSchedulerTestAccount(1, HealthTierHealthy, 100, 1) - scheduler := NewFastScheduler(1) + scheduler := NewFastScheduler(1, "round_robin") scheduler.Rebuild([]*Account{acc}) acc.SetCooldownUntil(time.Now().Add(5*time.Minute), "rate_limited") @@ -392,7 +392,7 @@ func TestFastSchedulerDispatchScoreOutranksProvenHistory(t *testing.T) { proven := newFastSchedulerTestAccount(2, HealthTierHealthy, 100, 1) atomic.StoreInt64(&proven.TotalRequests, 11) - scheduler := NewFastScheduler(1) + scheduler := NewFastScheduler(1, "round_robin") scheduler.Rebuild([]*Account{highScore, proven}) got := scheduler.Acquire() @@ -413,7 +413,7 @@ func TestFastSchedulerProvenHistoryBreaksDispatchScoreTies(t *testing.T) { proven := newFastSchedulerTestAccount(2, HealthTierHealthy, 100, 1) atomic.StoreInt64(&proven.TotalRequests, 11) - scheduler := NewFastScheduler(1) + scheduler := NewFastScheduler(1, "round_robin") scheduler.Rebuild([]*Account{unproven, proven}) got := scheduler.Acquire() @@ -442,7 +442,7 @@ func TestFastSchedulerPrefersPremium7dResetSoonOverProvenAccount(t *testing.T) { soon.UsagePercent7dValid = true soon.Reset7dAt = now.Add(36 * time.Hour) - scheduler := NewFastScheduler(1) + scheduler := NewFastScheduler(1, "round_robin") scheduler.Rebuild([]*Account{later, soon}) got := scheduler.Acquire() @@ -470,7 +470,7 @@ func TestFastSchedulerPrefersPremium5hResetSoonWithinTier(t *testing.T) { soon.UsagePercent5hValid = true soon.Reset5hAt = now.Add(30 * time.Minute) - scheduler := NewFastScheduler(1) + scheduler := NewFastScheduler(1, "round_robin") scheduler.Rebuild([]*Account{later, soon}) got := scheduler.Acquire() @@ -600,7 +600,7 @@ func TestFastSchedulerPremium5hRateLimitIsFencedAndRecoversAfterReset(t *testing Reset5hAt: time.Now().Add(30 * time.Minute), } - scheduler := NewFastScheduler(4) + scheduler := NewFastScheduler(4, "round_robin") scheduler.Rebuild([]*Account{acc}) sizes := scheduler.BucketSizes() @@ -784,10 +784,189 @@ func TestFastSchedulerRelease(t *testing.T) { acc := newFastSchedulerTestAccount(1, HealthTierHealthy, 100, 2) atomic.StoreInt64(&acc.ActiveRequests, 1) - scheduler := NewFastScheduler(2) + scheduler := NewFastScheduler(2, "round_robin") scheduler.Release(acc) if got := atomic.LoadInt64(&acc.ActiveRequests); got != 0 { t.Fatalf("ActiveRequests after Release() = %d, want 0", got) } } + +func TestFastSchedulerRemainingQuotaPicksLowestUsage(t *testing.T) { + highUsage := &Account{ + DBID: 1, + AccessToken: "token", + Status: StatusReady, + HealthTier: HealthTierHealthy, + UsagePercent7d: 90, + UsagePercent7dValid: true, + BaseConcurrencyEffective: 4, + DynamicConcurrencyLimit: 4, + } + lowUsage := &Account{ + DBID: 2, + AccessToken: "token", + Status: StatusReady, + HealthTier: HealthTierHealthy, + UsagePercent7d: 10, + UsagePercent7dValid: true, + BaseConcurrencyEffective: 4, + DynamicConcurrencyLimit: 4, + } + midUsage := &Account{ + DBID: 3, + AccessToken: "token", + Status: StatusReady, + HealthTier: HealthTierHealthy, + UsagePercent7d: 50, + UsagePercent7dValid: true, + BaseConcurrencyEffective: 4, + DynamicConcurrencyLimit: 4, + } + + scheduler := NewFastScheduler(4, "remaining_quota") + scheduler.Rebuild([]*Account{highUsage, lowUsage, midUsage}) + + got := scheduler.Acquire() + if got == nil { + t.Fatal("Acquire() returned nil") + } + defer scheduler.Release(got) + + if got.DBID != lowUsage.DBID { + t.Fatalf("Acquire() picked dbID=%d, want lowest-usage account %d", got.DBID, lowUsage.DBID) + } +} + +func TestFastSchedulerRemainingQuotaSortOrder(t *testing.T) { + a1 := &Account{ + DBID: 1, + AccessToken: "token", + Status: StatusReady, + HealthTier: HealthTierHealthy, + UsagePercent7d: 70, + UsagePercent7dValid: true, + BaseConcurrencyEffective: 1, + DynamicConcurrencyLimit: 1, + } + a2 := &Account{ + DBID: 2, + AccessToken: "token", + Status: StatusReady, + HealthTier: HealthTierHealthy, + UsagePercent7d: 30, + UsagePercent7dValid: true, + BaseConcurrencyEffective: 1, + DynamicConcurrencyLimit: 1, + } + a3 := &Account{ + DBID: 3, + AccessToken: "token", + Status: StatusReady, + HealthTier: HealthTierHealthy, + UsagePercent7d: 90, + UsagePercent7dValid: true, + BaseConcurrencyEffective: 1, + DynamicConcurrencyLimit: 1, + } + + scheduler := NewFastScheduler(1, "remaining_quota") + scheduler.Rebuild([]*Account{a1, a2, a3}) + + // Acquire all without releasing; concurrency limit 1 forces + // iteration through the usage-ascending sorted order. + var got []int64 + for i := 0; i < 3; i++ { + acc := scheduler.Acquire() + if acc == nil { + t.Fatalf("Acquire() returned nil at iteration %d", i) + } + got = append(got, acc.DBID) + } + + // Expect ascending usage: a2 (30%), a1 (70%), a3 (90%) + want := []int64{2, 1, 3} + for i := range want { + if got[i] != want[i] { + t.Fatalf("remaining_quota sort order mismatch: got=%v want=%v", got, want) + } + } +} + +func TestFastSchedulerRemainingQuotaTieBreakProvenThenDBID(t *testing.T) { + unproven := &Account{ + DBID: 1, + AccessToken: "token", + Status: StatusReady, + HealthTier: HealthTierHealthy, + UsagePercent7d: 50, + UsagePercent7dValid: true, + BaseConcurrencyEffective: 2, + DynamicConcurrencyLimit: 2, + } + unproven.TotalRequests = 0 // not proven + + proven := &Account{ + DBID: 2, + AccessToken: "token", + Status: StatusReady, + HealthTier: HealthTierHealthy, + UsagePercent7d: 50, + UsagePercent7dValid: true, + BaseConcurrencyEffective: 2, + DynamicConcurrencyLimit: 2, + } + proven.TotalRequests = 11 // proven (>10) + + scheduler := NewFastScheduler(4, "remaining_quota") + scheduler.Rebuild([]*Account{unproven, proven}) + + // Both same usage; proven account should sort first. + got := scheduler.Acquire() + if got == nil { + t.Fatal("Acquire() returned nil") + } + defer scheduler.Release(got) + + if got.DBID != proven.DBID { + t.Fatalf("Acquire() picked dbID=%d, want proven tie-breaker account %d", got.DBID, proven.DBID) + } +} + +func TestFastSchedulerSetSchedulerModeEmptyDefaultsToRoundRobin(t *testing.T) { + s := NewFastScheduler(4, "remaining_quota") + if s.SchedulerMode() != "remaining_quota" { + t.Fatalf("initial mode = %q, want remaining_quota", s.SchedulerMode()) + } + s.SetSchedulerMode("") + if s.SchedulerMode() != "round_robin" { + t.Fatalf("after empty mode: got %q, want round_robin", s.SchedulerMode()) + } +} + +func TestFastSchedulerSetSchedulerModeResortsBuckets(t *testing.T) { + highUsage := newFastSchedulerTestAccount(1, HealthTierHealthy, 90, 2) + highUsage.UsagePercent7d = 90 + highUsage.UsagePercent7dValid = true + lowUsage := newFastSchedulerTestAccount(2, HealthTierHealthy, 10, 2) + lowUsage.UsagePercent7d = 10 + lowUsage.UsagePercent7dValid = true + + s := NewFastScheduler(4, "round_robin") + s.Rebuild([]*Account{highUsage, lowUsage}) + + // In round_robin mode, acquires are cursor-based, not usage-based + first := s.Acquire() + s.Release(first) + + s.SetSchedulerMode("remaining_quota") + // After re-sort + zero cursor, should pick lowest usage first + first = s.Acquire() + if first == nil { + t.Fatal("Acquire() returned nil after mode switch") + } + if first.DBID != lowUsage.DBID { + t.Fatalf("after remaining_quota switch, Acquire() picked dbID=%d, want lowest-usage account %d", first.DBID, lowUsage.DBID) + } + s.Release(first) +} diff --git a/auth/store.go b/auth/store.go index 4420b744..b1ed11d8 100644 --- a/auth/store.go +++ b/auth/store.go @@ -110,6 +110,8 @@ type Account struct { // per-account 调度配置（nil = 跟随默认） ScoreBiasOverride *int64 BaseConcurrencyOverride *int64 + CreditEnabled bool // 信用账号标记 + CreditSkipUsageWindow bool // 跳过用量窗口惩罚 AllowedAPIKeyIDs []int64 allowedAPIKeySet map[int64]struct{} Tags []string @@ -617,7 +619,7 @@ func (a *Account) schedulerBreakdownLocked(now time.Time) SchedulerBreakdown { } } - if a.UsagePercent7dValid && strings.EqualFold(a.PlanType, "free") { + if !(a.CreditEnabled && a.CreditSkipUsageWindow) && a.UsagePercent7dValid && strings.EqualFold(a.PlanType, "free") { switch { case a.UsagePercent7d >= 100: breakdown.UsagePenalty7d = 40 @@ -801,7 +803,7 @@ func (a *Account) recomputeSchedulerLocked(baseLimit int64) { if !a.LastUnauthorizedAt.IsZero() && now.Sub(a.LastUnauthorizedAt) < 24*time.Hour && tier == HealthTierHealthy { tier = HealthTierWarm } - if a.UsagePercent7dValid && strings.EqualFold(a.PlanType, "free") { + if !(a.CreditEnabled && a.CreditSkipUsageWindow) && a.UsagePercent7dValid && strings.EqualFold(a.PlanType, "free") { switch { case a.UsagePercent7d >= 95: tier = HealthTierRisky @@ -880,6 +882,9 @@ func (a *Account) IsAvailable() bool { // usageExhaustedLocked 判断 Free 账号 7d 用量是否已耗尽（需持有 mu 读锁） func (a *Account) usageExhaustedLocked() bool { + if a.CreditEnabled && a.CreditSkipUsageWindow { + return false + } return a.UsagePercent7dValid && strings.EqualFold(a.PlanType, "free") && a.UsagePercent7d >= 100 } @@ -1014,6 +1019,16 @@ func (a *Account) GetUsagePercent7d() (float64, bool) { return a.UsagePercent7d, a.UsagePercent7dValid } +// usagePercentForScheduling 返回调度排序用的用量百分比（7d 窗口有效则返回，否则 0）。 +func (a *Account) usagePercentForScheduling() float64 { + a.mu.RLock() + defer a.mu.RUnlock() + if a.UsagePercent7dValid { + return a.UsagePercent7d + } + return 0 +} + // SetUsageSnapshot5h 更新 5h 用量快照 func (a *Account) SetUsageSnapshot5h(pct float64, resetAt time.Time) { a.mu.Lock() @@ -1491,6 +1506,7 @@ type Store struct { allowRemoteMigration atomic.Bool // 是否允许远程迁移拉取账号 modelMapping atomic.Value // 模型映射 JSON 字符串 + schedulerMode atomic.Value // string: "round_robin" or "remaining_quota" promptFilterConfig atomic.Value // promptfilter.Config sessionMu sync.RWMutex sessionBindings map[string]sessionAffinity @@ -1841,6 +1857,7 @@ func NewStore(db *database.DB, tc cache.TokenCache, settings *database.SystemSet RecoveryProbeIntervalMinutes: 30, ProxyURL: "", MaxRateLimitRetries: 1, + SchedulerMode: "round_robin", } } s := &Store{ @@ -1874,6 +1891,7 @@ func NewStore(db *database.DB, tc cache.TokenCache, settings *database.SystemSet } atomic.StoreInt64(&s.maxRateLimitRetries, rateLimitRetries) s.allowRemoteMigration.Store(settings.AllowRemoteMigration) + s.schedulerMode.Store(settings.SchedulerMode) if settings.ModelMapping != "" { s.modelMapping.Store(settings.ModelMapping) } @@ -1882,7 +1900,7 @@ func NewStore(db *database.DB, tc cache.TokenCache, settings *database.SystemSet fastEnabled := fastSchedulerEnabledFromEnv() || settings.FastSchedulerEnabled s.fastSchedulerEnabled.Store(fastEnabled) if fastEnabled { - s.fastScheduler.Store(NewFastScheduler(int64(settings.MaxConcurrency))) + s.fastScheduler.Store(NewFastScheduler(int64(settings.MaxConcurrency), s.GetSchedulerMode())) log.Printf("快速调度器已启用（请求热路径将优先走本地内存调度器）") } @@ -2281,6 +2299,8 @@ func (s *Store) loadFromDB(ctx context.Context) error { if !row.Enabled { atomic.StoreInt32(&account.DispatchPaused, 1) } + account.CreditEnabled = row.CreditEnabled + account.CreditSkipUsageWindow = row.CreditSkipUsageWindow if row.Status == "error" { account.Status = StatusError account.ErrorMsg = row.ErrorMessage @@ -2844,6 +2864,10 @@ func (s *Store) Release(acc *Account) { // SetMaxConcurrency 动态更新每账号并发上限 func (s *Store) SetMaxConcurrency(n int) { atomic.StoreInt64(&s.maxConcurrency, int64(n)) + // Update existing scheduler's base limit in-place before full rebuild. + if scheduler := s.getFastScheduler(); scheduler != nil { + scheduler.SetBaseLimit(int64(n)) + } s.recomputeAllAccountSchedulerState() s.rebuildFastScheduler() } @@ -2938,6 +2962,28 @@ func (s *Store) GetModelMapping() string { return "{}" } +// GetSchedulerMode 获取当前调度模式 +func (s *Store) GetSchedulerMode() string { + if v, ok := s.schedulerMode.Load().(string); ok { + return v + } + return "round_robin" +} + +// SetSchedulerMode 设置调度模式并传播到 FastScheduler +func (s *Store) SetSchedulerMode(mode string) { + switch mode { + case "round_robin", "remaining_quota": + // ok + default: + mode = "round_robin" + } + s.schedulerMode.Store(mode) + if scheduler := s.getFastScheduler(); scheduler != nil { + scheduler.SetSchedulerMode(mode) + } +} + func promptFilterConfigFromSettings(settings *database.SystemSettings) promptfilter.Config { cfg := promptfilter.DefaultConfig() if settings == nil { @@ -3070,6 +3116,31 @@ func (s *Store) ApplyAccountGroups(dbID int64, groupIDs []int64) bool { return true } +// UpdateAccountCredit 更新账号信用设置 +// 传入 nil 表示不修改该字段。 +func (s *Store) UpdateAccountCredit(dbID int64, creditEnabled, creditSkipUsageWindow *bool) error { + acc := s.FindByID(dbID) + if acc == nil { + return fmt.Errorf("账号 %d 不存在", dbID) + } + ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second) + defer cancel() + if err := s.db.UpdateAccountCredit(ctx, dbID, creditEnabled, creditSkipUsageWindow); err != nil { + return err + } + acc.mu.Lock() + if creditEnabled != nil { + acc.CreditEnabled = *creditEnabled + } + if creditSkipUsageWindow != nil { + acc.CreditSkipUsageWindow = *creditSkipUsageWindow + } + acc.recomputeSchedulerLocked(atomic.LoadInt64(&s.maxConcurrency)) + acc.mu.Unlock() + s.fastSchedulerUpdate(acc) + return nil +} + func (s *Store) ApplyAccountGroupMemberships(memberships map[int64][]int64) { for _, acc := range s.Accounts() { acc.mu.Lock() diff --git a/database/billing.go b/database/billing.go index d2a9c557..826a9329 100644 --- a/database/billing.go +++ b/database/billing.go @@ -2,14 +2,22 @@ package database import "strings" -// ModelPricing 模型价格配置（每百万 token 的价格，单位：美元） +const longContextThreshold = 272000 + type ModelPricing struct { - InputPricePerMToken float64 // 输入价格（美元/百万token） - InputPricePerMTokenPriority float64 // priority service tier 输入价格 - OutputPricePerMToken float64 // 输出价格（美元/百万token） - OutputPricePerMTokenPriority float64 // priority service tier 输出价格 - CacheReadPricePerMToken float64 // 缓存命中输入价格 - CacheReadPricePerMTokenPriority float64 // priority service tier 缓存命中输入价格 + InputPricePerMToken float64 + InputPricePerMTokenPriority float64 + OutputPricePerMToken float64 + OutputPricePerMTokenPriority float64 + CacheReadPricePerMToken float64 + CacheReadPricePerMTokenPriority float64 + + LongInputPricePerMToken float64 + LongInputPricePerMTokenPriority float64 + LongOutputPricePerMToken float64 + LongOutputPricePerMTokenPriority float64 + LongCacheReadPricePerMToken float64 + LongCacheReadPricePerMTokenPriority float64 } type modelPricingRule struct { @@ -17,29 +25,44 @@ type modelPricingRule struct { pricing ModelPricing } -type costBreakdown struct { - InputCost float64 - OutputCost float64 - CacheReadCost float64 - TotalCost float64 - InputPricePerMToken float64 - OutputPricePerMToken float64 - CacheReadPricePerMToken float64 - ServiceTierCostMultiplier float64 +type CostBreakdown struct { + InputCost float64 `json:"input_cost"` + OutputCost float64 `json:"output_cost"` + CacheReadCost float64 `json:"cache_read_cost"` + TotalCost float64 `json:"total_cost"` + InputPricePerMToken float64 `json:"input_price_per_mtoken"` + OutputPricePerMToken float64 `json:"output_price_per_mtoken"` + CacheReadPricePerMToken float64 `json:"cache_read_price_per_mtoken"` + ServiceTierCostMultiplier float64 `json:"service_tier_cost_multiplier"` } var ( defaultModelPricing = &ModelPricing{InputPricePerMToken: 1.0, OutputPricePerMToken: 2.0} modelPricingRules = []modelPricingRule{ - // Codex/GPT-5 系列，参考 sub2api 的动态定价优先、fallback 兜底策略。 {model: "gpt-5.5", pricing: ModelPricing{ - InputPricePerMToken: 2.5, - InputPricePerMTokenPriority: 5.0, - OutputPricePerMToken: 15.0, - OutputPricePerMTokenPriority: 30.0, - CacheReadPricePerMToken: 0.25, - CacheReadPricePerMTokenPriority: 0.5, + InputPricePerMToken: 5.0, + InputPricePerMTokenPriority: 12.5, + OutputPricePerMToken: 30.0, + OutputPricePerMTokenPriority: 75.0, + CacheReadPricePerMToken: 0.5, + CacheReadPricePerMTokenPriority: 1.25, + LongInputPricePerMToken: 10.0, + LongInputPricePerMTokenPriority: 25.0, + LongOutputPricePerMToken: 45.0, + LongOutputPricePerMTokenPriority: 112.5, + LongCacheReadPricePerMToken: 1.0, + LongCacheReadPricePerMTokenPriority: 2.5, + }}, + {model: "gpt-5.5-pro", pricing: ModelPricing{ + InputPricePerMToken: 30.0, + InputPricePerMTokenPriority: 75.0, + OutputPricePerMToken: 180.0, + OutputPricePerMTokenPriority: 450.0, + LongInputPricePerMToken: 60.0, + LongInputPricePerMTokenPriority: 150.0, + LongOutputPricePerMToken: 270.0, + LongOutputPricePerMTokenPriority: 675.0, }}, {model: "gpt-5.4-mini", pricing: ModelPricing{InputPricePerMToken: 0.75, OutputPricePerMToken: 4.5, CacheReadPricePerMToken: 0.075}}, {model: "gpt-5.4-nano", pricing: ModelPricing{InputPricePerMToken: 0.2, OutputPricePerMToken: 1.25, CacheReadPricePerMToken: 0.02}}, @@ -50,6 +73,22 @@ var ( OutputPricePerMTokenPriority: 30.0, CacheReadPricePerMToken: 0.25, CacheReadPricePerMTokenPriority: 0.5, + LongInputPricePerMToken: 5.0, + LongInputPricePerMTokenPriority: 10.0, + LongOutputPricePerMToken: 22.5, + LongOutputPricePerMTokenPriority: 45.0, + LongCacheReadPricePerMToken: 0.5, + LongCacheReadPricePerMTokenPriority: 1.0, + }}, + {model: "gpt-5.4-pro", pricing: ModelPricing{ + InputPricePerMToken: 30.0, + InputPricePerMTokenPriority: 75.0, + OutputPricePerMToken: 180.0, + OutputPricePerMTokenPriority: 450.0, + LongInputPricePerMToken: 60.0, + LongInputPricePerMTokenPriority: 150.0, + LongOutputPricePerMToken: 270.0, + LongOutputPricePerMTokenPriority: 675.0, }}, {model: "gpt-5.3-codex-spark", pricing: ModelPricing{ InputPricePerMToken: 1.25, @@ -75,8 +114,6 @@ var ( CacheReadPricePerMToken: 0.175, CacheReadPricePerMTokenPriority: 0.35, }}, - - // GPT-4 系列。保持最具体模型优先，避免 gpt-4o-mini 被 gpt-4o/gpt-4 抢先匹配。 {model: "gpt-4o-mini", pricing: ModelPricing{InputPricePerMToken: 0.15, OutputPricePerMToken: 0.6}}, {model: "gpt-4o", pricing: ModelPricing{InputPricePerMToken: 2.5, OutputPricePerMToken: 10.0}}, {model: "gpt-4-turbo", pricing: ModelPricing{InputPricePerMToken: 10.0, OutputPricePerMToken: 30.0}}, @@ -85,9 +122,7 @@ var ( } ) -// getModelPricing 获取模型价格配置 -// 优先使用确定性的模型族匹配，避免 Go map 迭代顺序导致重叠前缀随机命中。 -func getModelPricing(model string) *ModelPricing { +func GetModelPricing(model string) *ModelPricing { normalized := normalizeBillingModelName(model) if pricing := claudeFamilyPricing(normalized); pricing != nil { return pricing @@ -104,31 +139,42 @@ func getModelPricing(model string) *ModelPricing { return defaultModelPricing } -// calculateCost 计算使用费用 -// inputTokens: 输入 token 数量 -// outputTokens: 输出 token 数量 -// model: 模型名称 -// 返回：账号计费金额（美元） -func calculateCost(inputTokens, outputTokens, cachedTokens int, model string, serviceTier string) float64 { - return calculateCostBreakdown(inputTokens, outputTokens, cachedTokens, model, serviceTier).TotalCost +func CalculateCost(inputTokens, outputTokens, cachedTokens int, model string, serviceTier string) float64 { + return CalculateCostBreakdown(inputTokens, outputTokens, cachedTokens, model, serviceTier).TotalCost } -func calculateCostBreakdown(inputTokens, outputTokens, cachedTokens int, model string, serviceTier string) costBreakdown { - pricing := getModelPricing(model) +func CalculateCostBreakdown(inputTokens, outputTokens, cachedTokens int, model string, serviceTier string) CostBreakdown { + pricing := GetModelPricing(model) + isLong := inputTokens > longContextThreshold + inputPrice := pricing.InputPricePerMToken outputPrice := pricing.OutputPricePerMToken cacheReadPrice := pricing.CacheReadPricePerMToken + if isLong && pricing.LongInputPricePerMToken > 0 { + inputPrice = pricing.LongInputPricePerMToken + outputPrice = pricing.LongOutputPricePerMToken + if pricing.LongCacheReadPricePerMToken > 0 { + cacheReadPrice = pricing.LongCacheReadPricePerMToken + } + } + tierMultiplier := serviceTierCostMultiplier(serviceTier) if usePriorityPricing(serviceTier, pricing) { tierMultiplier = 1 - if pricing.InputPricePerMTokenPriority > 0 { + if isLong && pricing.LongInputPricePerMTokenPriority > 0 { + inputPrice = pricing.LongInputPricePerMTokenPriority + } else if pricing.InputPricePerMTokenPriority > 0 { inputPrice = pricing.InputPricePerMTokenPriority } - if pricing.OutputPricePerMTokenPriority > 0 { + if isLong && pricing.LongOutputPricePerMTokenPriority > 0 { + outputPrice = pricing.LongOutputPricePerMTokenPriority + } else if pricing.OutputPricePerMTokenPriority > 0 { outputPrice = pricing.OutputPricePerMTokenPriority } - if pricing.CacheReadPricePerMTokenPriority > 0 { + if isLong && pricing.LongCacheReadPricePerMTokenPriority > 0 { + cacheReadPrice = pricing.LongCacheReadPricePerMTokenPriority + } else if pricing.CacheReadPricePerMTokenPriority > 0 { cacheReadPrice = pricing.CacheReadPricePerMTokenPriority } } @@ -149,7 +195,7 @@ func calculateCostBreakdown(inputTokens, outputTokens, cachedTokens int, model s cacheReadCost := float64(cachedTokens) / 1000000.0 * cacheReadPrice outputCost := float64(outputTokens) / 1000000.0 * outputPrice - return costBreakdown{ + return CostBreakdown{ InputCost: inputCost * tierMultiplier, OutputCost: outputCost * tierMultiplier, CacheReadCost: cacheReadCost * tierMultiplier, @@ -178,14 +224,18 @@ func normalizeBillingModelName(model string) string { } func normalizeCodexBillingModel(model string) (string, bool) { - compact := strings.NewReplacer(" ", "-", "_", "-").Replace(model) + compact := strings.NewReplacer(" ", "-", "_", "-").Replace(strings.ToLower(model)) switch { + case strings.Contains(compact, "gpt-5.5-pro") || strings.Contains(compact, "gpt5-5-pro") || strings.Contains(compact, "gpt5.5-pro"): + return "gpt-5.5-pro", true case strings.Contains(compact, "gpt-5.5") || strings.Contains(compact, "gpt5-5") || strings.Contains(compact, "gpt5.5"): return "gpt-5.5", true case strings.Contains(compact, "gpt-5.4-mini") || strings.Contains(compact, "gpt5-4-mini") || strings.Contains(compact, "gpt5.4-mini"): return "gpt-5.4-mini", true case strings.Contains(compact, "gpt-5.4-nano") || strings.Contains(compact, "gpt5-4-nano") || strings.Contains(compact, "gpt5.4-nano"): return "gpt-5.4-nano", true + case strings.Contains(compact, "gpt-5.4-pro") || strings.Contains(compact, "gpt5-4-pro") || strings.Contains(compact, "gpt5.4-pro"): + return "gpt-5.4-pro", true case strings.Contains(compact, "gpt-5.4") || strings.Contains(compact, "gpt5-4") || strings.Contains(compact, "gpt5.4"): return "gpt-5.4", true case strings.Contains(compact, "gpt-5.2") || strings.Contains(compact, "gpt5-2") || strings.Contains(compact, "gpt5.2"): @@ -196,6 +246,12 @@ func normalizeCodexBillingModel(model string) (string, bool) { return "gpt-5.3-codex", true case strings.Contains(compact, "gpt-5.3") || strings.Contains(compact, "gpt5-3") || strings.Contains(compact, "gpt5.3"): return "gpt-5.3-codex", true + case strings.Contains(compact, "codex-auto-review"): + // Codex internal auto-review model. ChatGPT backend API only + // (chatgpt.com/backend-api/codex). Not available via public API. + // Official catalog: Plus/Pro/Team/Business only, excludes free. + // Specs match gpt-5.4 (272K context, 4 thinking levels). + return "gpt-5.4", true case strings.Contains(compact, "codex"): return "gpt-5.3-codex", true case strings.Contains(compact, "gpt-5") || strings.Contains(compact, "gpt5"): @@ -292,3 +348,12 @@ func serviceTierCostMultiplier(serviceTier string) float64 { func normalizeServiceTier(serviceTier string) string { return strings.ToLower(strings.TrimSpace(serviceTier)) } + +// lowercase aliases for internal callers +func calculateCost(inputTokens, outputTokens, cachedTokens int, model string, serviceTier string) float64 { + return CalculateCost(inputTokens, outputTokens, cachedTokens, model, serviceTier) +} + +func calculateCostBreakdown(inputTokens, outputTokens, cachedTokens int, model string, serviceTier string) CostBreakdown { + return CalculateCostBreakdown(inputTokens, outputTokens, cachedTokens, model, serviceTier) +} diff --git a/database/billing_test.go b/database/billing_test.go index acc8a215..e7d65396 100644 --- a/database/billing_test.go +++ b/database/billing_test.go @@ -18,7 +18,7 @@ func TestGetModelPricingUsesMostSpecificOpenAIPrefix(t *testing.T) { for _, tt := range tests { t.Run(tt.model, func(t *testing.T) { - got := getModelPricing(tt.model) + got := GetModelPricing(tt.model) assertPricing(t, got, tt.wantInput, tt.wantOutput) }) } @@ -33,12 +33,12 @@ func TestGetModelPricingUsesSub2APICodexFallbacks(t *testing.T) { {model: "gpt-5.4-mini-20260401", wantInput: 0.75, wantOutput: 4.5}, {model: "gpt-5.3-codex-spark", wantInput: 1.25, wantOutput: 10.0}, {model: "gpt-5.3-codex", wantInput: 1.75, wantOutput: 14.0}, - {model: "gpt-5.5", wantInput: 2.5, wantOutput: 15.0}, + {model: "gpt-5.5", wantInput: 5.0, wantOutput: 30.0}, } for _, tt := range tests { t.Run(tt.model, func(t *testing.T) { - got := getModelPricing(tt.model) + got := GetModelPricing(tt.model) assertPricing(t, got, tt.wantInput, tt.wantOutput) }) } @@ -59,7 +59,7 @@ func TestGetModelPricingUsesSub2APIClaudeFamilies(t *testing.T) { for _, tt := range tests { t.Run(tt.model, func(t *testing.T) { - got := getModelPricing(tt.model) + got := GetModelPricing(tt.model) assertPricing(t, got, tt.wantInput, tt.wantOutput) }) } @@ -84,7 +84,7 @@ func TestCalculateCostHandlesCachedTokensAndServiceTier(t *testing.T) { want: 0.00955, }, { - name: "keeps legacy input price when cache pricing is unavailable", + name: "without cache pricing (gpt-4o)", model: "gpt-4o", inputTokens: 1000, outputTokens: 500, @@ -134,20 +134,22 @@ func TestCalculateCostBreakdownExposesDisplayFields(t *testing.T) { assertFloatEqual(t, got.ServiceTierCostMultiplier, 0.5) } -func TestGPT55PricingMatchesGPT54Fallback(t *testing.T) { - gpt54 := getModelPricing("gpt-5.4") - gpt55 := getModelPricing("gpt-5.5") +func TestGPT55PricingDoesNotMatchGPT54(t *testing.T) { + gpt54 := GetModelPricing("gpt-5.4") + gpt55 := GetModelPricing("gpt-5.5") - assertFloatEqual(t, gpt55.InputPricePerMToken, gpt54.InputPricePerMToken) - assertFloatEqual(t, gpt55.OutputPricePerMToken, gpt54.OutputPricePerMToken) - assertFloatEqual(t, gpt55.CacheReadPricePerMToken, gpt54.CacheReadPricePerMToken) - assertFloatEqual(t, gpt55.InputPricePerMTokenPriority, gpt54.InputPricePerMTokenPriority) - assertFloatEqual(t, gpt55.OutputPricePerMTokenPriority, gpt54.OutputPricePerMTokenPriority) - assertFloatEqual(t, gpt55.CacheReadPricePerMTokenPriority, gpt54.CacheReadPricePerMTokenPriority) + // gpt-5.5 is 2x gpt-5.4: $5/$30 vs $2.5/$15 + assertFloatEqual(t, gpt55.InputPricePerMToken, 5.0) + assertFloatEqual(t, gpt55.OutputPricePerMToken, 30.0) + assertFloatEqual(t, gpt55.CacheReadPricePerMToken, 0.5) + assertFloatEqual(t, gpt55.InputPricePerMTokenPriority, 12.5) + assertFloatEqual(t, gpt55.OutputPricePerMTokenPriority, 75.0) + assertFloatEqual(t, gpt55.CacheReadPricePerMTokenPriority, 1.25) + _ = gpt54 } func TestSparkPricingUsesGpt51CodexFallback(t *testing.T) { - spark := getModelPricing("gpt-5.3-codex-spark-high") + spark := GetModelPricing("gpt-5.3-codex-spark-high") assertFloatEqual(t, spark.InputPricePerMToken, 1.25) assertFloatEqual(t, spark.OutputPricePerMToken, 10.0) @@ -158,8 +160,8 @@ func TestSparkPricingUsesGpt51CodexFallback(t *testing.T) { } func TestGPT53CodexPricingUsesGPT52CodexFallback(t *testing.T) { - codex := getModelPricing("gpt-5.3-codex-xhigh") - gpt52 := getModelPricing("gpt-5.2") + codex := GetModelPricing("gpt-5.3-codex-xhigh") + gpt52 := GetModelPricing("gpt-5.2") assertFloatEqual(t, codex.InputPricePerMToken, gpt52.InputPricePerMToken) assertFloatEqual(t, codex.OutputPricePerMToken, gpt52.OutputPricePerMToken) @@ -171,7 +173,7 @@ func TestGPT53CodexPricingUsesGPT52CodexFallback(t *testing.T) { func TestUsageLogBreakdownScalesToStoredBilledTotal(t *testing.T) { log := &UsageLog{ - Model: "gpt-5.5", + Model: "gpt-5.4", InputTokens: 1000, StatusCode: 200, AccountBilled: 0.0025, @@ -188,7 +190,7 @@ func TestUsageLogBreakdownScalesToStoredBilledTotal(t *testing.T) { func assertPricing(t *testing.T, got *ModelPricing, wantInput, wantOutput float64) { t.Helper() if got == nil { - t.Fatal("getModelPricing returned nil") + t.Fatal("GetModelPricing returned nil") } if math.Abs(got.InputPricePerMToken-wantInput) > 1e-12 || math.Abs(got.OutputPricePerMToken-wantOutput) > 1e-12 { t.Fatalf("pricing = input %.12f output %.12f, want input %.12f output %.12f", @@ -202,3 +204,122 @@ func assertFloatEqual(t *testing.T, got, want float64) { t.Fatalf("got %.12f, want %.12f", got, want) } } + +func TestProModelsHaveCorrectPricing(t *testing.T) { + p55 := GetModelPricing("gpt-5.5-pro") + if p55.InputPricePerMToken != 30.0 || p55.OutputPricePerMToken != 180.0 { + t.Fatalf("gpt-5.5-pro pricing = input %.1f output %.1f, want input 30.0 output 180.0", + p55.InputPricePerMToken, p55.OutputPricePerMToken) + } + + p54 := GetModelPricing("gpt-5.4-pro") + if p54.InputPricePerMToken != 30.0 || p54.OutputPricePerMToken != 180.0 { + t.Fatalf("gpt-5.4-pro pricing = input %.1f output %.1f, want input 30.0 output 180.0", + p54.InputPricePerMToken, p54.OutputPricePerMToken) + } + + // Verify pro is NOT same as standard + std55 := GetModelPricing("gpt-5.5") + if std55.InputPricePerMToken == p55.InputPricePerMToken { + t.Fatal("gpt-5.5-pro should have different pricing from gpt-5.5") + } +} + +func TestLongContextPricingTriggersAbove272KTokens(t *testing.T) { + // Below threshold: standard pricing. + std := CalculateCostBreakdown(272000, 1000, 0, "gpt-5.4", "") + assertFloatEqual(t, std.InputPricePerMToken, 2.5) + assertFloatEqual(t, std.OutputPricePerMToken, 15.0) + + // Above threshold: long context premium pricing. + long := CalculateCostBreakdown(272001, 1000, 0, "gpt-5.4", "") + assertFloatEqual(t, long.InputPricePerMToken, 5.0) + assertFloatEqual(t, long.OutputPricePerMToken, 22.5) + + // Verify total cost is higher for long context. + if long.TotalCost <= std.TotalCost { + t.Fatalf("long context cost %.12f should be > standard cost %.12f", long.TotalCost, std.TotalCost) + } +} + +func TestLongContextPricingWithPriorityTier(t *testing.T) { + long := CalculateCostBreakdown(300000, 1000, 200, "gpt-5.4", "priority") + assertFloatEqual(t, long.InputPricePerMToken, 10.0) + assertFloatEqual(t, long.OutputPricePerMToken, 45.0) + assertFloatEqual(t, long.CacheReadPricePerMToken, 1.0) + assertFloatEqual(t, long.ServiceTierCostMultiplier, 1.0) +} + +func TestLongContextPricingDoesNotApplyWhenNoLongPricingDefined(t *testing.T) { + // gpt-4o has no long context pricing fields defined. + std := CalculateCostBreakdown(1000, 500, 0, "gpt-4o", "") + long := CalculateCostBreakdown(300000, 500, 0, "gpt-4o", "") + // Input price should be the same since no long variant exists. + assertFloatEqual(t, std.InputPricePerMToken, long.InputPricePerMToken) + assertFloatEqual(t, std.OutputPricePerMToken, long.OutputPricePerMToken) +} + +func TestCodexAutoReviewModelNormalizesToGPT54(t *testing.T) { + pricing := GetModelPricing("codex-auto-review") + if pricing == nil { + t.Fatal("GetModelPricing(codex-auto-review) returned nil") + } + // Should normalize to gpt-5.4 pricing. + gpt54 := GetModelPricing("gpt-5.4") + assertFloatEqual(t, pricing.InputPricePerMToken, gpt54.InputPricePerMToken) + assertFloatEqual(t, pricing.OutputPricePerMToken, gpt54.OutputPricePerMToken) + assertFloatEqual(t, pricing.CacheReadPricePerMToken, gpt54.CacheReadPricePerMToken) + assertFloatEqual(t, pricing.InputPricePerMTokenPriority, gpt54.InputPricePerMTokenPriority) + assertFloatEqual(t, pricing.OutputPricePerMTokenPriority, gpt54.OutputPricePerMTokenPriority) + assertFloatEqual(t, pricing.CacheReadPricePerMTokenPriority, gpt54.CacheReadPricePerMTokenPriority) +} + +func TestCodexAutoReviewLongContextPricing(t *testing.T) { + // codex-auto-review maps to gpt-5.4 which has long context pricing. + long := CalculateCostBreakdown(300000, 500, 100, "codex-auto-review", "") + assertFloatEqual(t, long.InputPricePerMToken, 5.0) // long input price + assertFloatEqual(t, long.OutputPricePerMToken, 22.5) // long output price + assertFloatEqual(t, long.CacheReadPricePerMToken, 0.5) // long cache read price +} + +func TestCodexAutoReviewPriorityPricing(t *testing.T) { + bd := CalculateCostBreakdown(1000, 500, 0, "codex-auto-review", "priority") + assertFloatEqual(t, bd.InputPricePerMToken, 5.0) + assertFloatEqual(t, bd.OutputPricePerMToken, 30.0) + assertFloatEqual(t, bd.ServiceTierCostMultiplier, 1.0) +} + +func TestNormalizeCodexBillingModelCodexAutoReview(t *testing.T) { + tests := []struct { + model string + want string + }{ + {"codex-auto-review", "gpt-5.4"}, + {"codex-auto-review-v2", "gpt-5.4"}, // variant suffix should match + {"CODEX-AUTO-REVIEW", "gpt-5.4"}, // case-insensitive + {"codex_auto_review", "gpt-5.4"}, // underscores normalized + } + for _, tt := range tests { + t.Run(tt.model, func(t *testing.T) { + got, ok := normalizeCodexBillingModel(tt.model) + if !ok { + t.Fatalf("normalizeCodexBillingModel(%q) ok=false, want true", tt.model) + } + if got != tt.want { + t.Fatalf("normalizeCodexBillingModel(%q) = %q, want %q", tt.model, got, tt.want) + } + }) + } +} + +func TestGPT55LongContextPricing(t *testing.T) { + // gpt-5.5 has long pricing: $10/$45 standard, $25/$112.5 priority. + long := CalculateCostBreakdown(300000, 500, 0, "gpt-5.5", "") + assertFloatEqual(t, long.InputPricePerMToken, 10.0) + assertFloatEqual(t, long.OutputPricePerMToken, 45.0) + + longPri := CalculateCostBreakdown(300000, 500, 0, "gpt-5.5", "priority") + assertFloatEqual(t, longPri.InputPricePerMToken, 25.0) + assertFloatEqual(t, longPri.OutputPricePerMToken, 112.5) + assertFloatEqual(t, longPri.ServiceTierCostMultiplier, 1.0) +} diff --git a/database/postgres.go b/database/postgres.go index f34811c4..dc56414f 100644 --- a/database/postgres.go +++ b/database/postgres.go @@ -30,6 +30,8 @@ type AccountRow struct { ErrorMessage string Enabled bool Locked bool + CreditEnabled bool + CreditSkipUsageWindow bool ScoreBiasOverride sql.NullInt64 BaseConcurrencyOverride sql.NullInt64 Tags []string @@ -481,6 +483,8 @@ func (db *DB) migrate(ctx context.Context) error { ALTER TABLE accounts ADD COLUMN IF NOT EXISTS today_used_count INT DEFAULT 0; ALTER TABLE accounts ADD COLUMN IF NOT EXISTS image_quota_reset_at TIMESTAMPTZ NULL; ALTER TABLE accounts ADD COLUMN IF NOT EXISTS tags JSONB DEFAULT '[]'::jsonb; + ALTER TABLE accounts ADD COLUMN IF NOT EXISTS credit_enabled BOOLEAN DEFAULT FALSE; + ALTER TABLE accounts ADD COLUMN IF NOT EXISTS credit_skip_usage_window BOOLEAN DEFAULT FALSE; CREATE TABLE IF NOT EXISTS account_groups ( id SERIAL PRIMARY KEY, @@ -597,7 +601,8 @@ func (db *DB) migrate(ctx context.Context) error { auto_clean_rate_limited BOOLEAN DEFAULT FALSE, background_refresh_interval_minutes INT DEFAULT 2, usage_probe_max_age_minutes INT DEFAULT 10, - recovery_probe_interval_minutes INT DEFAULT 30 + recovery_probe_interval_minutes INT DEFAULT 30, + scheduler_mode VARCHAR(20) DEFAULT 'round_robin' ); CREATE TABLE IF NOT EXISTS account_model_cooldowns ( account_id BIGINT NOT NULL, @@ -627,6 +632,7 @@ func (db *DB) migrate(ctx context.Context) error { ALTER TABLE system_settings ADD COLUMN IF NOT EXISTS background_refresh_interval_minutes INT DEFAULT 2; ALTER TABLE system_settings ADD COLUMN IF NOT EXISTS usage_probe_max_age_minutes INT DEFAULT 10; ALTER TABLE system_settings ADD COLUMN IF NOT EXISTS recovery_probe_interval_minutes INT DEFAULT 30; + ALTER TABLE system_settings ADD COLUMN IF NOT EXISTS scheduler_mode VARCHAR(20) DEFAULT 'round_robin'; ALTER TABLE system_settings ADD COLUMN IF NOT EXISTS resin_url TEXT DEFAULT ''; ALTER TABLE system_settings ADD COLUMN IF NOT EXISTS resin_platform_name TEXT DEFAULT ''; ALTER TABLE system_settings ADD COLUMN IF NOT EXISTS prompt_filter_enabled BOOLEAN DEFAULT FALSE; @@ -1093,6 +1099,7 @@ type SystemSettings struct { BackgroundRefreshIntervalMinutes int UsageProbeMaxAgeMinutes int RecoveryProbeIntervalMinutes int + SchedulerMode string ResinURL string // Resin 代理池地址（含 Token），例如 http://127.0.0.1:2260/my-token ResinPlatformName string // Resin 平台标识，例如 codex2api PromptFilterEnabled bool @@ -1132,6 +1139,7 @@ func (db *DB) GetSystemSettings(ctx context.Context) (*SystemSettings, error) { COALESCE(background_refresh_interval_minutes, 2), COALESCE(usage_probe_max_age_minutes, 10), COALESCE(recovery_probe_interval_minutes, 30), + COALESCE(scheduler_mode, 'round_robin'), COALESCE(resin_url, ''), COALESCE(resin_platform_name, ''), COALESCE(prompt_filter_enabled, false), @@ -1159,6 +1167,7 @@ func (db *DB) GetSystemSettings(ctx context.Context) (*SystemSettings, error) { &s.ProxyPoolEnabled, &s.FastSchedulerEnabled, &s.MaxRetries, &s.MaxRateLimitRetries, &s.AllowRemoteMigration, &s.AutoCleanError, &s.AutoCleanExpired, &s.ModelMapping, &s.BackgroundRefreshIntervalMinutes, &s.UsageProbeMaxAgeMinutes, &s.RecoveryProbeIntervalMinutes, + &s.SchedulerMode, &s.ResinURL, &s.ResinPlatformName, &s.PromptFilterEnabled, &s.PromptFilterMode, &s.PromptFilterThreshold, &s.PromptFilterStrictThreshold, &s.PromptFilterLogMatches, &s.PromptFilterMaxTextLength, &s.PromptFilterSensitiveWords, @@ -1188,9 +1197,10 @@ func (db *DB) UpdateSystemSettings(ctx context.Context, s *SystemSettings) error prompt_filter_sensitive_words, prompt_filter_custom_patterns, prompt_filter_disabled_patterns, client_compat_mode, codex_min_cli_version, usage_log_mode, usage_log_batch_size, usage_log_flush_interval_seconds, stream_flush_policy, stream_flush_interval_ms, - image_storage_config + image_storage_config, + scheduler_mode ) - VALUES (1, $1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31, $32, $33, $34, $35, $36, $37, $38, $39, $40, $41, $42, $43) + VALUES (1, $1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31, $32, $33, $34, $35, $36, $37, $38, $39, $40, $41, $42, $43, $44) ON CONFLICT (id) DO UPDATE SET site_name = EXCLUDED.site_name, site_logo = EXCLUDED.site_logo, @@ -1234,7 +1244,8 @@ func (db *DB) UpdateSystemSettings(ctx context.Context, s *SystemSettings) error usage_log_flush_interval_seconds = EXCLUDED.usage_log_flush_interval_seconds, stream_flush_policy = EXCLUDED.stream_flush_policy, stream_flush_interval_ms = EXCLUDED.stream_flush_interval_ms, - image_storage_config = EXCLUDED.image_storage_config + image_storage_config = EXCLUDED.image_storage_config, + scheduler_mode = EXCLUDED.scheduler_mode `, NormalizeSiteName(s.SiteName), strings.TrimSpace(s.SiteLogo), s.MaxConcurrency, s.GlobalRPM, s.TestModel, s.TestConcurrency, s.ProxyURL, s.PgMaxConns, s.RedisPoolSize, s.AutoCleanUnauthorized, s.AutoCleanRateLimited, s.AdminSecret, s.AutoCleanFullUsage, s.ProxyPoolEnabled, @@ -1245,7 +1256,7 @@ func (db *DB) UpdateSystemSettings(ctx context.Context, s *SystemSettings) error s.PromptFilterSensitiveWords, s.PromptFilterCustomPatterns, s.PromptFilterDisabledPatterns, s.ClientCompatMode, s.CodexMinCLIVersion, s.UsageLogMode, s.UsageLogBatchSize, s.UsageLogFlushIntervalSeconds, s.StreamFlushPolicy, s.StreamFlushIntervalMS, - s.ImageStorageConfig) + s.ImageStorageConfig, s.SchedulerMode) return err } @@ -2928,12 +2939,21 @@ func (db *DB) GetAccountTimeRangeUsage(ctx context.Context, since time.Time) (ma return result, rows.Err() } +// GetAccountBilledSince 返回指定时间戳以来 account_billed 的总和 +func (db *DB) GetAccountBilledSince(ctx context.Context, accountID int64, since time.Time) (float64, error) { + var billed float64 + err := db.conn.QueryRowContext(ctx, + `SELECT COALESCE(SUM(account_billed), 0) FROM usage_logs WHERE account_id = $1 AND created_at >= $2 AND status_code <> 499`, + accountID, db.timeArg(since)).Scan(&billed) + return billed, err +} + // ==================== Accounts ==================== // ListActive 获取所有未删除账号。 func (db *DB) ListActive(ctx context.Context) ([]*AccountRow, error) { query := ` - SELECT id, name, platform, type, credentials, proxy_url, status, cooldown_reason, cooldown_until, error_message, COALESCE(enabled, true), COALESCE(locked, false), score_bias_override, base_concurrency_override, COALESCE(tags, '[]'), created_at, updated_at + SELECT id, name, platform, type, credentials, proxy_url, status, cooldown_reason, cooldown_until, error_message, COALESCE(enabled, true), COALESCE(locked, false), COALESCE(credit_enabled, false), COALESCE(credit_skip_usage_window, false), score_bias_override, base_concurrency_override, COALESCE(tags, '[]'), created_at, updated_at FROM accounts WHERE status <> 'deleted' AND COALESCE(error_message, '') <> 'deleted' ORDER BY id @@ -2965,6 +2985,8 @@ func (db *DB) ListActive(ctx context.Context) ([]*AccountRow, error) { &a.ErrorMessage, &a.Enabled, &a.Locked, + &a.CreditEnabled, + &a.CreditSkipUsageWindow, &a.ScoreBiasOverride, &a.BaseConcurrencyOverride, &tagsRaw, @@ -3074,7 +3096,7 @@ func (db *DB) ClearExpiredModelCooldowns(ctx context.Context) error { // GetAccountByID 获取未删除账号的完整数据库行。 func (db *DB) GetAccountByID(ctx context.Context, id int64) (*AccountRow, error) { query := ` - SELECT id, name, platform, type, credentials, proxy_url, status, cooldown_reason, cooldown_until, error_message, COALESCE(enabled, true), COALESCE(locked, false), score_bias_override, base_concurrency_override, COALESCE(tags, '[]'), created_at, updated_at + SELECT id, name, platform, type, credentials, proxy_url, status, cooldown_reason, cooldown_until, error_message, COALESCE(enabled, true), COALESCE(locked, false), COALESCE(credit_enabled, false), COALESCE(credit_skip_usage_window, false), score_bias_override, base_concurrency_override, COALESCE(tags, '[]'), created_at, updated_at FROM accounts WHERE id = $1 AND status <> 'deleted' AND COALESCE(error_message, '') <> 'deleted' LIMIT 1 @@ -3098,6 +3120,8 @@ func (db *DB) GetAccountByID(ctx context.Context, id int64) (*AccountRow, error) &a.ErrorMessage, &a.Enabled, &a.Locked, + &a.CreditEnabled, + &a.CreditSkipUsageWindow, &a.ScoreBiasOverride, &a.BaseConcurrencyOverride, &tagsRaw, @@ -3340,6 +3364,46 @@ func (db *DB) SetAccountLocked(ctx context.Context, id int64, locked bool) error return err } +// UpdateAccountCredit 更新账号的信用设置（credit_enabled / credit_skip_usage_window） +// 传入 nil 表示不修改该字段，仅 SET 非 nil 的列。 +func (db *DB) UpdateAccountCredit(ctx context.Context, id int64, creditEnabled, creditSkipUsageWindow *bool) error { + var setClauses []string + var args []interface{} + argIdx := 1 + + if creditEnabled != nil { + setClauses = append(setClauses, fmt.Sprintf("credit_enabled = $%d", argIdx)) + args = append(args, *creditEnabled) + argIdx++ + } + if creditSkipUsageWindow != nil { + setClauses = append(setClauses, fmt.Sprintf("credit_skip_usage_window = $%d", argIdx)) + args = append(args, *creditSkipUsageWindow) + argIdx++ + } + + if len(setClauses) == 0 { + return nil // 没有要更新的字段 + } + + setClauses = append(setClauses, "updated_at = CURRENT_TIMESTAMP") + query := "UPDATE accounts SET " + strings.Join(setClauses, ", ") + fmt.Sprintf(" WHERE id = $%d", argIdx) + args = append(args, id) + + res, err := db.conn.ExecContext(ctx, query, args...) + if err != nil { + return err + } + affected, err := res.RowsAffected() + if err != nil { + return err + } + if affected == 0 { + return sql.ErrNoRows + } + return nil +} + // UpdateCredentials 原子合并更新账号的 credentials（JSONB || 运算符，不覆盖已有字段） // 解决并发刷新时一个进程覆盖另一个进程写入的字段的问题 func (db *DB) UpdateCredentials(ctx context.Context, id int64, credentials map[string]interface{}) error { diff --git a/database/sqlite.go b/database/sqlite.go index 5609e81f..a10aad78 100644 --- a/database/sqlite.go +++ b/database/sqlite.go @@ -138,7 +138,8 @@ func (db *DB) migrateSQLite(ctx context.Context) error { usage_log_flush_interval_seconds INTEGER DEFAULT 5, stream_flush_policy TEXT DEFAULT 'immediate', stream_flush_interval_ms INTEGER DEFAULT 20, - image_storage_config TEXT DEFAULT '{}' + image_storage_config TEXT DEFAULT '{}', + scheduler_mode TEXT DEFAULT 'round_robin' );`, `CREATE TABLE IF NOT EXISTS model_registry ( id TEXT PRIMARY KEY, @@ -330,8 +331,11 @@ func (db *DB) migrateSQLite(ctx context.Context) error { {"system_settings", "stream_flush_policy", "TEXT DEFAULT 'immediate'"}, {"system_settings", "stream_flush_interval_ms", "INTEGER DEFAULT 20"}, {"system_settings", "image_storage_config", "TEXT DEFAULT '{}'"}, + {"system_settings", "scheduler_mode", "TEXT DEFAULT 'round_robin'"}, {"accounts", "enabled", "INTEGER DEFAULT 1"}, {"accounts", "locked", "INTEGER DEFAULT 0"}, + {"accounts", "credit_enabled", "INTEGER DEFAULT 0"}, + {"accounts", "credit_skip_usage_window", "INTEGER DEFAULT 0"}, {"accounts", "image_quota_remaining", "INTEGER NULL"}, {"accounts", "image_quota_total", "INTEGER NULL"}, {"accounts", "today_used_count", "INTEGER DEFAULT 0"}, diff --git a/docker-compose.sqlite.local.yml b/docker-compose.sqlite.local.yml index cac25167..87d74567 100644 --- a/docker-compose.sqlite.local.yml +++ b/docker-compose.sqlite.local.yml @@ -5,7 +5,9 @@ services: build: . container_name: codex2api-sqlite-local ports: - - "${BIND_HOST:-0.0.0.0}:${CODEX_PORT:-8080}:${CODEX_PORT:-8080}" + # Bind to localhost only by default for security. + # To expose on all interfaces, set BIND_HOST=0.0.0.0 in .env or override here. + - "${BIND_HOST:-127.0.0.1}:${CODEX_PORT:-8080}:${CODEX_PORT:-8080}" env_file: - .env volumes: diff --git a/docker-compose.sqlite.yml b/docker-compose.sqlite.yml index b20a791d..2833de11 100644 --- a/docker-compose.sqlite.yml +++ b/docker-compose.sqlite.yml @@ -5,7 +5,9 @@ services: image: ghcr.io/james-6-23/codex2api:latest container_name: codex2api-sqlite ports: - - "${BIND_HOST:-0.0.0.0}:${CODEX_PORT:-8080}:${CODEX_PORT:-8080}" + # Bind to localhost only by default for security. + # To expose on all interfaces, set BIND_HOST=0.0.0.0 in .env or override here. + - "${BIND_HOST:-127.0.0.1}:${CODEX_PORT:-8080}:${CODEX_PORT:-8080}" env_file: - .env volumes: diff --git a/docs/API.md b/docs/API.md index 1c93ebae..6273cea3 100644 --- a/docs/API.md +++ b/docs/API.md @@ -21,6 +21,7 @@ - [代理池管理](#代理池管理) - [运维监控](#运维监控) - [模型管理](#模型管理) + - [生图工作台](#生图工作台) — 文生图、图生图任务与图库管理 - [OAuth 授权](#oauth-授权) — PKCE 流程获取 Token - [支持模型](#支持模型) - [错误码](#错误码) @@ -287,7 +288,7 @@ data: [DONE] "object": "list", "data": [ { "id": "gpt-5.5", "object": "model", "owned_by": "openai" }, - { "id": "gpt-5.5", "object": "model", "owned_by": "openai" }, + { "id": "gpt-5.4", "object": "model", "owned_by": "openai" }, { "id": "gpt-5.4-mini", "object": "model", "owned_by": "openai" }, { "id": "gpt-5.3-codex", "object": "model", "owned_by": "openai" }, { "id": "gpt-5.3-codex-spark", "object": "model", "owned_by": "openai" }, @@ -384,6 +385,10 @@ data: [DONE] "last_used_at": "2024-01-01T11:00:00Z", "success_requests": 95, "error_requests": 5, + "credit_enabled": false, + "credit_skip_usage_window": false, + "billed_5h": 0.25, + "billed_7d": 3.50, "usage_percent_7d": 45.2, "usage_percent_5h": 12.5, "reset_5h_at": "2024-01-01T17:00:00Z", @@ -416,6 +421,10 @@ data: [DONE] | base_concurrency_override | integer/null | 手工配置的基础并发覆盖值，`null` 表示跟随全局 `max_concurrency` | | base_concurrency_effective | integer | 当前生效的基础并发值 | | allowed_api_key_ids | integer[] | 允许调用该账号的 API Key ID 列表；空数组表示所有 API Key 均可调用 | +| credit_enabled | bool | 是否为信用计费模式账号 | +| credit_skip_usage_window | bool | 是否跳过 7 天/5 小时用量窗口惩罚 | +| billed_5h | number/null | 过去 5 小时窗口内的累计计费金额（USD） | +| billed_7d | number/null | 过去 7 天窗口内的累计计费金额（USD） | #### PATCH /api/admin/accounts/:id/scheduler @@ -457,6 +466,36 @@ data: [DONE] } ``` +#### PATCH /api/admin/accounts/:id/credit + +更新账号信用设置。 + +**请求:** + +```json +{ + "credit_enabled": true, + "credit_skip_usage_window": true +} +``` + +**参数说明:** + +| 参数 | 类型 | 必填 | 说明 | +| -------------------------- | ----- | ---- | ---------------------------------------- | +| credit_enabled | bool | 否 | 标记账号为信用计费模式，省略时保持原值 | +| credit_skip_usage_window | bool | 否 | 跳过 7 天/5 小时用量窗口惩罚（仅在 `credit_enabled=true` 时生效），省略时保持原值 | + +**响应:** + +```json +{ + "message": "信用设置已更新", + "credit_enabled": true, + "credit_skip_usage_window": true +} +``` + #### POST /api/admin/accounts 添加 Refresh Token 账号（支持批量）。 @@ -1152,6 +1191,8 @@ curl -X DELETE "http://localhost:8080/api/admin/account-groups/1?force=true" \ "proxy_pool_enabled": false, "fast_scheduler_enabled": false, "max_retries": 3, + "max_rate_limit_retries": 2, + "scheduler_mode": "round_robin", "allow_remote_migration": false, "database_driver": "postgres", "database_label": "PostgreSQL", @@ -1177,7 +1218,9 @@ curl -X DELETE "http://localhost:8080/api/admin/account-groups/1?force=true" \ "proxy_url": "http://proxy.example.com:8080", "auto_clean_unauthorized": true, "auto_clean_rate_limited": false, - "fast_scheduler_enabled": true + "fast_scheduler_enabled": true, + "scheduler_mode": "remaining_quota", + "max_rate_limit_retries": 2 } ``` @@ -1413,6 +1456,127 @@ curl -X DELETE "http://localhost:8080/api/admin/account-groups/1?force=true" \ } ``` +### 生图工作台 + +管理后台生图工作台的 API，支持文生图（`/images/jobs`）和图生图（`/images/edit-jobs`）任务，以及图库管理。 + +#### POST /api/admin/images/jobs + +创建文生图任务。 + +**请求:** + +```json +{ + "prompt": "A small orange cat sitting on a cloud", + "model": "gpt-image-2", + "size": "1024x1024", + "quality": "high", + "output_format": "png", + "style": "photorealistic", + "api_key_id": 0 +} +``` + +**参数说明:** + +| 参数 | 类型 | 必填 | 说明 | +| ------------- | ------ | ---- | ----------------------------- | +| prompt | string | 是 | 提示词，最长 8000 字符 | +| model | string | 否 | 模型，默认 `gpt-image-2` | +| size | string | 否 | 输出尺寸 | +| quality | string | 否 | 质量等级 | +| output_format | string | 否 | 输出格式，默认 `png` | +| style | string | 否 | 风格说明 | +| upscale | string | 否 | 超分选项 | +| api_key_id | int | 否 | 指定 API Key ID | + +**响应:** + +```json +{ + "id": 1, + "status": "pending" +} +``` + +#### POST /api/admin/images/edit-jobs + +创建图生图（image-to-image）任务。与文生图参数类似，但需要额外提供参考图片。 + +**请求:** + +```json +{ + "prompt": "Replace the background with a sunset scene", + "model": "gpt-image-2", + "size": "1024x1024", + "output_format": "png", + "input_images": [ + "https://example.com/source.png", + "data:image/png;base64,iVBORw0KGgo..." + ] +} +``` + +**参数说明:** + +| 参数 | 类型 | 必填 | 说明 | +| ------------- | -------- | ---- | ------------------------------------ | +| prompt | string | 是 | 编辑提示词，最长 8000 字符 | +| model | string | 否 | 模型，默认 `gpt-image-2` | +| input_images | string[] | 是 | 参考图片 URL 或 data URI 列表 | +| size | string | 否 | 输出尺寸 | +| quality | string | 否 | 质量等级 | +| output_format | string | 否 | 输出格式，默认 `png` | +| api_key_id | int | 否 | 指定 API Key ID | + +#### GET /api/admin/images/jobs + +获取生图任务列表。支持分页和状态过滤。 + +**响应:** + +```json +{ + "jobs": [ + { + "id": 1, + "prompt": "A small orange cat", + "model": "gpt-image-2", + "status": "completed", + "created_at": "2024-01-01T12:00:00Z" + } + ], + "total": 50 +} +``` + +#### GET /api/admin/images/jobs/:id + +获取单个生图任务详情及结果。 + +#### DELETE /api/admin/images/jobs/:id + +删除一个生图任务及其关联的所有图库资产。 + +```bash +curl -X DELETE http://localhost:8080/api/admin/images/jobs/1 \ + -H "X-Admin-Key: your-secret" +``` + +#### GET /api/admin/images/assets + +获取图库资产列表。 + +#### GET /api/admin/images/assets/:id/file + +获取单个图库资产文件（返回图片二进制或签名 URL）。 + +#### DELETE /api/admin/images/assets/:id + +删除单个图库资产。 + ### OAuth 授权 通过 OAuth PKCE 流程授权获取 Codex 账号的 Refresh Token，适用于无法手动获取 RT 的场景。 @@ -1491,10 +1655,10 @@ curl -X DELETE "http://localhost:8080/api/admin/account-groups/1?force=true" \ ## 支持模型 -| 模型 | 说明 | -| ------------------- | --------------------------------------- | -| gpt-5.5 | 最新旗舰模型 | -| gpt-5.4 | 旗舰模型 | +| 模型 | 说明 | +| ------------------- | ----------------------------------------------------------- | +| gpt-5.5 | 最新旗舰模型。计费：$5.00/M 输入 / $30.00/M 输出（标准），priority 分别为 $12.50/M / $75.00/M | +| gpt-5.4 | 旗舰模型 | | gpt-5.4-mini | 轻量版 | | gpt-5.3-codex | 较新版本 | | gpt-5.3-codex-spark | Codex Spark 模型，仅 Pro 订阅账号可调用 | diff --git a/docs/CHANGELOG.md b/docs/CHANGELOG.md new file mode 100644 index 00000000..c13b8a3a --- /dev/null +++ b/docs/CHANGELOG.md @@ -0,0 +1,39 @@ +# Changelog — iteration/may-2026-v2 + +Dates: 2026-05-13 to 2026-05-20. 17 commits. + +## Features + +- **Credit quota support (#141).** Added `credit_enabled` and `credit_skip_usage_window` flags to the accounts table. Credit-marked accounts skip usage-window penalties in the scheduler. Managed via `PATCH /api/admin/accounts/:id/credit`. + +- **Scheduler mode (#133).** Added `scheduler_mode` system setting with two modes: `round_robin` (default, weighted by dispatch score) and `remaining_quota` (prioritize accounts with lowest usage percent). Configurable from Admin Settings page. + +- **5h/7d windowed USD cost display.** Replaced the single total-cost column with a windowed billing view. Each account now shows `billed_5h` and `billed_7d` fields aligned with the account's usage-reset boundaries. This reflects actual spending, not estimated token costs. + +- **Image-to-image in Image Studio (#135, #136).** The admin Image Studio now supports image-to-image generation via `POST /api/admin/images/edit-jobs`, accepting reference image URLs or data URIs. Added text-to-image and image-to-image tabs in the frontend. + +- **Billing model expansion.** Added pricing for gpt-5.5-pro and gpt-5.4-pro families. Implemented long context (>272K tokens) premium pricing for gpt-5.5, gpt-5.5-pro, gpt-5.4, gpt-5.4-pro with automatic detection. Fixed gpt-4o and gpt-4o-mini cache-read pricing. + +## Fixes + +- **GPT-5.5 pricing corrected.** Updated standard-tier billing from old values to $5.00/M input / $30.00/M output (priority: $12.50/M / $75.00/M), matching current official pricing. + +- **SSE stream isolation.** Prevented SSE response mixing when retrying across accounts, using `c.Writer.Written()` as the retry guard instead of a package-level flag. + +- **Usage logging for image errors.** Added usage-log emits for read-error paths in image generation, ensuring billing records are not silently dropped on stream failures. + +- **Model mapping initialization.** Restored `modelMapping` init that was accidentally removed during the scheduler_mode refactor. + +- **Credit field Scan order.** Fixed PostgreSQL `Scan` argument ordering for credit fields that was causing silent zero-values. + +- **Round 2 review fixes.** Addressed Haiku review findings including api.ts syntax cleanup, billing test corrections, and several CRITICAL/HIGH issues from automated review. + +## Security + +- **SQLite default binds to localhost.** SQLite compose files (`docker-compose.sqlite.yml`, `docker-compose.sqlite.local.yml`) now bind ports to `127.0.0.1` by default. Previously they bound to `0.0.0.0`, exposing the service on all interfaces. Standard (PostgreSQL) compose files retain the `0.0.0.0` default. + +- **BIND_HOST env var.** Added `BIND_HOST` environment variable support to control the HTTP listen address across all deployment modes. Documented in `.env.example`, `.env.sqlite.example`, and `CONFIGURATION.md`. + +## Breaking Changes + +- **SQLite compose port binding.** SQLite deployments upgrading from a previous version that relied on external access via the default compose configuration must now explicitly set `BIND_HOST=0.0.0.0` in `.env` or override the port binding in the compose file. All other behavior remains backwards-compatible. diff --git a/docs/CONFIGURATION.md b/docs/CONFIGURATION.md index b57635ff..af2ad25f 100644 --- a/docs/CONFIGURATION.md +++ b/docs/CONFIGURATION.md @@ -43,7 +43,10 @@ Codex2API 采用三层配置架构： | 变量 | 必填 | 默认值 | 说明 | |------|------|--------|------| | `CODEX_PORT` | 否 | 8080 | HTTP 服务端口 | +| `BIND_HOST` | 否 | `127.0.0.1`（SQLite）/ `0.0.0.0`（PostgreSQL） | Docker 端口发布绑定地址（非进程监听地址，由 `CODEX_BIND` 控制）。SQLite compose 默认 `127.0.0.1` 仅本机访问；标准 compose 默认 `0.0.0.0` 所有网络接口 | | `ADMIN_SECRET` | 否 | - | 管理后台登录密钥 | +| `CODEX_ALLOW_ANONYMOUS` | 否 | `false` | 设为 `true` 时，未配置任何对外 API Key 也允许 `/v1/*` 直接调用（仅限内网测试场景） | +| `FAST_SCHEDULER_ENABLED` | 否 | `false` | 通过环境变量启用快速调度器（也可在管理后台运行时开启） | | `TZ` | 否 | UTC | 时区，如 `Asia/Shanghai` | ### Codex 上游稳定性配置 @@ -51,6 +54,7 @@ Codex2API 采用三层配置架构： | 变量 | 必填 | 默认值 | 说明 | |------|------|--------|------| | `CODEX_UPSTREAM_TRANSPORT` | 否 | `http` | Codex 上游协议：`http` / `auto` / `ws`。HTTP 入站在 `auto` 下仍走 HTTP 上游 | +| `CODEX_PROXY_URL` | 否 | - | 全局代理 URL，适用于需要为所有 Codex 上游请求统一配置代理的场景 | | `USE_WEBSOCKET` | 否 | `false` | 旧版开关；未设置 `CODEX_UPSTREAM_TRANSPORT` 时，`true` 等价于 `CODEX_UPSTREAM_TRANSPORT=ws` | | `CODEX_TRANSPORT_MODE` | 否 | `standard` | Codex HTTP transport：默认标准 Go TLS；`utls_chrome` 可回滚旧 Chrome uTLS 行为 | | `CODEX_WS_SEND_USER_AGENT` | 否 | `false` | WS 握手是否发送 `User-Agent`/`Version`；默认关闭 | @@ -128,7 +132,9 @@ Codex2API 采用三层配置架构： | `MaxConcurrency` | int | 2 | 1-50 | 单账号最大并发请求数 | | `GlobalRPM` | int | 0 | 0-∞ | 全局每分钟请求限制，0 表示不限 | | `MaxRetries` | int | 3 | 0-10 | 请求失败最大重试次数 | +| `MaxRateLimitRetries` | int | 2 | 0-10 | 遇到 429 限流时的最大额外重试次数 | | `FastSchedulerEnabled` | bool | false | - | 启用快速调度器 | +| `SchedulerMode` | string | `round_robin` | - | 调度模式：`round_robin`（轮询，按调度分权重排序）或 `remaining_quota`（优先使用用量少的账号） | ### 测试配置 @@ -144,6 +150,17 @@ Codex2API 采用三层配置架构： | `ProxyURL` | string | "" | 全局代理 URL | | `ProxyPoolEnabled` | bool | false | 启用代理池 | +### 账号级设置（单账号） + +以下字段存储在 `accounts` 表中，可通过管理后台账号详情或 API 按账号单独设置： + +| 字段 | 类型 | 默认值 | 说明 | +|------|------|--------|------| +| `credit_enabled` | bool | false | 标记账号为信用计费模式 | +| `credit_skip_usage_window` | bool | false | 跳过 7 天/5 小时用量窗口惩罚（适用于信用账号） | +| `score_bias_override` | int/null | null | 手工覆盖调度权重分，`null` 跟随套餐默认 | +| `base_concurrency_override` | int/null | null | 手工覆盖基础并发值，`null` 跟随全局默认 | + ### 连接池配置 | 字段 | 类型 | 默认值 | 范围 | 说明 | diff --git a/frontend/src/api.ts b/frontend/src/api.ts index e77229c3..cdba8615 100644 --- a/frontend/src/api.ts +++ b/frontend/src/api.ts @@ -222,6 +222,8 @@ export const api = { request<{ message: string; success: number; failed: number }>('/accounts/batch-reset-status', { method: 'POST', body: JSON.stringify({ ids }) }), getAccountUsage: (id: number) => request(`/accounts/${id}/usage`), + updateAccountCredit: (id: number, data: { credit_enabled: boolean; credit_skip_usage_window: boolean }) => + request(`/accounts/${id}/credit`, { method: 'PATCH', body: JSON.stringify(data) }), getHealth: () => request('/health'), getOpsOverview: () => request('/ops/overview'), getOpsErrorSummary: (params: { @@ -336,6 +338,8 @@ export const api = { request(`/image-prompts/${id}`, { method: 'DELETE' }), createImageJob: (data: CreateImageJobPayload) => request('/images/jobs', { method: 'POST', body: JSON.stringify(data) }), + createImageEditJob: (data: CreateImageJobPayload) => + request('/images/edit-jobs', { method: 'POST', body: JSON.stringify(data) }), getImageJobs: (params: { page?: number; pageSize?: number } = {}) => { const sp = new URLSearchParams() if (params.page) sp.set('page', String(params.page)) diff --git a/frontend/src/components/AccountUsageModal.tsx b/frontend/src/components/AccountUsageModal.tsx index abb3aed4..21819293 100644 --- a/frontend/src/components/AccountUsageModal.tsx +++ b/frontend/src/components/AccountUsageModal.tsx @@ -19,6 +19,11 @@ export default function AccountUsageModal({ account, onClose }: Props) { const [loading, setLoading] = useState(true) const [error, setError] = useState(null) + const [creditEnabled, setCreditEnabled] = useState(account.credit_enabled ?? false) + const [creditSkipWindow, setCreditSkipWindow] = useState(account.credit_skip_usage_window ?? false) + const [savingCredit, setSavingCredit] = useState(false) + const [creditError, setCreditError] = useState(null) + const load = useCallback(async () => { setLoading(true) setError(null) @@ -39,6 +44,25 @@ export default function AccountUsageModal({ account, onClose }: Props) { : (account.email || account.name || `#${account.id}`) const title = t('accounts.usageDetailTitle') + ' — ' + accountLabel + const handleCreditToggle = async (field: 'credit_enabled' | 'credit_skip_usage_window', value: boolean) => { + setCreditError(null) + const newEnabled = field === 'credit_enabled' ? value : creditEnabled + const newSkip = field === 'credit_skip_usage_window' ? value : creditSkipWindow + setSavingCredit(true) + try { + await api.updateAccountCredit(account.id, { + credit_enabled: newEnabled, + credit_skip_usage_window: newSkip, + }) + if (field === 'credit_enabled') setCreditEnabled(value) + if (field === 'credit_skip_usage_window') setCreditSkipWindow(value) + } catch (err) { + setCreditError(getErrorMessage(err)) + } finally { + setSavingCredit(false) + } + } + return ( {loading ? ( @@ -101,6 +125,50 @@ export default function AccountUsageModal({ account, onClose }: Props) { )} + + {/* 信用设置 */} +
+

{t('accounts.creditSettings')}

+ {creditError && ( +
{creditError}
+ )} +
+
+

{t('accounts.creditEnabled')}

+

{t('accounts.creditEnabledHint')}

+
+ +
+ {creditEnabled && ( +
+
+

{t('accounts.creditSkipWindow')}

+

{t('accounts.creditSkipWindowHint')}

+
+ +
+ )} +
) } diff --git a/frontend/src/locales/en.json b/frontend/src/locales/en.json index 3dce35bd..df239619 100644 --- a/frontend/src/locales/en.json +++ b/frontend/src/locales/en.json @@ -550,7 +550,13 @@ "batchMetaDesc": "The following tags and groups will be saved to {{count}} selected accounts. Leaving a field empty clears it.", "batchMetaDone": "Batch metadata save complete: {{success}} succeeded, {{fail}} failed", "batchMetaFailed": "Batch metadata save failed: {{error}}", - "testAutoReset": "Test passed, account status has been automatically restored to normal" + "testAutoReset": "Test passed, account status has been automatically restored to normal", + "creditSettings": "Credit Settings", + "creditEnabled": "Credit Enabled", + "creditEnabledHint": "Mark this account as having extra credits", + "creditSkipWindow": "Skip Usage Window Penalty", + "creditSkipWindowHint": "Bypass 7d/5h usage window penalties during scheduling", + "billed": "Cost" }, "status": { "active": "Available", @@ -938,6 +944,10 @@ "timezoneAuto": "Auto (Browser Timezone)", "fastSchedulerEnabled": "Fast Scheduler", "fastSchedulerEnabledDesc": "Uses an in-memory fast scheduling algorithm, significantly reducing dispatch latency under high concurrency. Recommended for large account pools (100+).", + "schedulerMode": "Scheduler Mode", + "schedulerModeDesc": "Round Robin distributes requests evenly across accounts. Remaining Quota prefers accounts with lower usage percentage.", + "schedulerModeRoundRobin": "Round Robin", + "schedulerModeRemainingQuota": "Remaining Quota", "allowRemoteMigration": "Allow Remote Migration", "allowRemoteMigrationDesc": "When enabled, other codex2api instances can pull healthy accounts from this instance.", "allowRemoteMigrationRequiresSecret": "Set an admin secret first. Remote migration cannot be enabled without admin authentication.", @@ -1251,6 +1261,7 @@ "images": { "title": "Image Studio", "description": "Text-to-image generation, prompt templates, and server gallery", + "mode": "Mode", "views": { "studio": "Studio", "prompts": "Prompts", @@ -1373,6 +1384,14 @@ "deleteAssetTitle": "Delete gallery image", "assetDeleted": "Image deleted", "assetDeletedInHistory": "Image has been deleted", + "imageToImage": "Image to Image", + "textToImage": "Text to Image", + "inputImage": "Reference Image", + "inputImageHint": "Upload one or more reference images; AI will modify based on these images", + "inputImageRequired": "Reference image is required for image-to-image mode", + "maxInputImages": "Maximum {{max}} reference images allowed", + "removeImage": "Remove image", + "upload": "Upload", "loadFailed": "Failed to load image studio", "status": { "queued": "Queued", diff --git a/frontend/src/locales/zh.json b/frontend/src/locales/zh.json index 055c2297..da8dbf28 100644 --- a/frontend/src/locales/zh.json +++ b/frontend/src/locales/zh.json @@ -550,7 +550,13 @@ "batchMetaDesc": "将为 {{count}} 个选中账号统一保存以下标签和分组。留空会清空对应字段。", "batchMetaDone": "批量保存完成：成功 {{success}}，失败 {{fail}}", "batchMetaFailed": "批量保存失败：{{error}}", - "testAutoReset": "测试成功，账号状态已自动恢复为正常" + "testAutoReset": "测试成功，账号状态已自动恢复为正常", + "creditSettings": "信用设置", + "creditEnabled": "启用信用", + "creditEnabledHint": "标记该账号拥有额外信用", + "creditSkipWindow": "跳过用量窗口惩罚", + "creditSkipWindowHint": "调度时不使用 7d/5h 用量窗口降权", + "billed": "成本" }, "status": { "active": "可用", @@ -938,6 +944,10 @@ "timezoneAuto": "自动（浏览器时区）", "fastSchedulerEnabled": "快速调度器", "fastSchedulerEnabledDesc": "启用后，账号选择将使用内存快速调度算法，大幅降低高并发场景下的调度延迟。适合大号池（100+ 账号）场景。", + "schedulerMode": "调度模式", + "schedulerModeDesc": "轮询模式在所有账号间平均分配请求。剩余配额模式优先使用用量较低的账号。", + "schedulerModeRoundRobin": "轮询", + "schedulerModeRemainingQuota": "剩余配额", "allowRemoteMigration": "允许远程迁移", "allowRemoteMigrationDesc": "开启后，其他 codex2api 实例可通过导出接口拉取本实例的健康账号。", "allowRemoteMigrationRequiresSecret": "请先设置管理密钥，未设置密码时不能启用远程迁移。", @@ -1251,6 +1261,7 @@ "images": { "title": "生图工作台", "description": "文生图、提示词模板和服务器图库", + "mode": "生成模式", "views": { "studio": "工作台", "prompts": "提示词管理", @@ -1373,6 +1384,14 @@ "deleteAssetTitle": "删除图库图片", "assetDeleted": "图片已删除", "assetDeletedInHistory": "图片已经被删除", + "imageToImage": "图生图", + "textToImage": "文生图", + "inputImage": "参考图片", + "inputImageHint": "上传一张或多张参考图片，AI 将基于这些图片进行修改", + "inputImageRequired": "图生图模式需要上传参考图片", + "maxInputImages": "最多可上传 {{max}} 张参考图片", + "removeImage": "移除图片", + "upload": "上传", "loadFailed": "加载生图工作台失败", "status": { "queued": "排队中", diff --git a/frontend/src/pages/Accounts.tsx b/frontend/src/pages/Accounts.tsx index ad6f79cd..bb0675ef 100644 --- a/frontend/src/pages/Accounts.tsx +++ b/frontend/src/pages/Accounts.tsx @@ -88,6 +88,7 @@ const ACCOUNT_TABLE_COLUMNS = [ "status", "requests", "usage", + "billed", "importTime", "updatedAt", "actions", @@ -2435,6 +2436,7 @@ export default function Accounts() { status: t("accounts.status"), requests: t("accounts.requests"), usage: t("accounts.usage"), + billed: t("accounts.billed"), importTime: t("accounts.importTime"), updatedAt: t("accounts.updatedAt"), actions: t("accounts.actions"), @@ -2645,6 +2647,11 @@ export default function Accounts() { : ""} )} + {visibleColumns.billed && ( + + {t("accounts.billed")} + + )} {visibleColumns.importTime && ( )} + {visibleColumns.billed && ( + + + + )} {visibleColumns.importTime && ( {formatBeijingTime(account.created_at)} @@ -6099,6 +6111,19 @@ function UsageCell({ account }: { account: AccountRow }) { return -; } +function BilledCell({ account }: { account: AccountRow }) { + const h5 = typeof account.billed_5h === "number" ? account.billed_5h.toFixed(2) : null; + const d7 = typeof account.billed_7d === "number" ? account.billed_7d.toFixed(2) : null; + if (h5 === null && d7 === null) return -; + return ( + + {h5 !== null ? `5h: $${h5}` : "5h: -"} + {" / "} + {d7 !== null ? `7d: $${d7}` : "7d: -"} + + ); +} + function getAccountStatusCountdownUntil( account: AccountRow, ): string | undefined { diff --git a/frontend/src/pages/ImageStudio.tsx b/frontend/src/pages/ImageStudio.tsx index 2480017b..fa46b4e5 100644 --- a/frontend/src/pages/ImageStudio.tsx +++ b/frontend/src/pages/ImageStudio.tsx @@ -23,7 +23,7 @@ import { import { Input } from '@/components/ui/input' import { Select } from '@/components/ui/select' import { Tooltip, TooltipContent, TooltipProvider, TooltipTrigger } from '@/components/ui/tooltip' -import { Copy, Download, Eye, Image as ImageIcon, Loader2, Pencil, Play, Plus, RefreshCcw, Save, Search, Sparkles, Star, Trash2, X } from 'lucide-react' +import { Copy, Download, Eye, Image as ImageIcon, Loader2, Pencil, Play, Plus, RefreshCcw, Save, Search, Sparkles, Star, Trash2, Upload, X } from 'lucide-react' const IMAGE_VIEWS = ['studio', 'prompts', 'gallery', 'history'] as const type ImageView = typeof IMAGE_VIEWS[number] @@ -431,6 +431,44 @@ export default function ImageStudio() { const [apiKeyID, setAPIKeyID] = useState('') const [templateName, setTemplateName] = useState('') const [templateTags, setTemplateTags] = useState('') + const [imageToImageMode, setImageToImageMode] = useState(false) + const [inputImageDataURLs, setInputImageDataURLs] = useState([]) + + const handleImageFileChange = useCallback((e: React.ChangeEvent) => { + const files = e.target.files + if (!files || files.length === 0) return + const MAX_INPUT_IMAGES = 10 + const currentCount = inputImageDataURLs.length + if (currentCount >= MAX_INPUT_IMAGES) { + showToast(t('images.maxInputImages', { max: MAX_INPUT_IMAGES }), 'error') + e.target.value = '' + return + } + const remaining = MAX_INPUT_IMAGES - currentCount + const filesToRead = Array.from(files).slice(0, remaining) + const reads: Promise[] = [] + for (let i = 0; i < filesToRead.length; i++) { + reads.push(new Promise((resolve, reject) => { + const reader = new FileReader() + reader.onload = () => resolve(reader.result as string) + reader.onerror = () => reject(new Error('Failed to read file')) + reader.readAsDataURL(filesToRead[i]) + })) + } + Promise.allSettled(reads).then(results => { + const dataURLs: string[] = [] + for (const r of results) { + if (r.status === 'fulfilled') dataURLs.push(r.value) + } + if (dataURLs.length > 0) { + setInputImageDataURLs(prev => [...prev, ...dataURLs].slice(0, MAX_INPUT_IMAGES)) + } + if (dataURLs.length < results.length) { + showToast(t('images.loadFailed'), 'error') + } + }) + e.target.value = '' + }, [showToast, t, inputImageDataURLs]) useEffect(() => { if (view && !IMAGE_VIEWS.includes(view as ImageView)) { @@ -766,17 +804,27 @@ export default function ImageStudio() { if (style.trim()) payload.style = style.trim() if (apiKeyID) payload.api_key_id = Number(apiKeyID) if (selectedTemplateId) payload.template_id = selectedTemplateId + if (imageToImageMode && inputImageDataURLs.length > 0) payload.input_images = inputImageDataURLs return payload } - const submitJob = async (payload = createJobPayload()) => { + const submitJob = async (payload = createJobPayload(), forceMode?: 'text' | 'edit') => { + const isEditMode = forceMode != null + ? forceMode === 'edit' + : Array.isArray(payload.input_images) && payload.input_images.length > 0 if (!payload.prompt.trim()) { showToast(t('images.promptRequired'), 'error') return } + if (isEditMode && (!payload.input_images || payload.input_images.length === 0)) { + showToast(t('images.inputImageRequired'), 'error') + return + } setSubmitting(true) try { - const res = await api.createImageJob(payload) + const res = isEditMode + ? await api.createImageEditJob(payload) + : await api.createImageJob(payload) setCurrentJob(res.job) await loadJobs() showToast(t('images.jobCreated'), 'success') @@ -791,6 +839,7 @@ export default function ImageStudio() { const params = jobParams(job) const nextModel = params.model || 'gpt-image-2' const nextSize = normalizeImageSizeForModel(nextModel, params.size || 'auto') + const isEditJob = params.input_images && params.input_images.length > 0 setPrompt(job.prompt) setModel(nextModel) setSize(nextSize) @@ -800,6 +849,13 @@ export default function ImageStudio() { setUpscale(normalizeUpscale(params.upscale)) setStyle(params.style || '') setSelectedTemplateId(params.template_id ? Number(params.template_id) : null) + if (isEditJob) { + setImageToImageMode(true) + setInputImageDataURLs(params.input_images!) + } else { + setImageToImageMode(false) + setInputImageDataURLs([]) + } navigate('/images/studio') void submitJob({ prompt: job.prompt, @@ -812,7 +868,8 @@ export default function ImageStudio() { style: params.style, api_key_id: apiKeyID ? Number(apiKeyID) : undefined, template_id: params.template_id ? Number(params.template_id) : undefined, - }) + input_images: isEditJob ? params.input_images : undefined, + }, isEditJob ? 'edit' : 'text') } const rerunFromAsset = (asset: ImageAsset) => { @@ -1000,7 +1057,9 @@ export default function ImageStudio() { outputFormat !== 'png' || background !== 'auto' || upscale || - apiKeyID + apiKeyID || + imageToImageMode || + inputImageDataURLs.length > 0 ) const clearGenerationForm = () => { @@ -1016,6 +1075,8 @@ export default function ImageStudio() { setAPIKeyID('') setTemplateName('') setTemplateTags('') + setImageToImageMode(false) + setInputImageDataURLs([]) } const changeGenerationModel = (value: string) => { @@ -1035,6 +1096,28 @@ export default function ImageStudio() { /> +
+ {t('images.mode')}: +
+ + +
+
+
@@ -1058,6 +1141,54 @@ export default function ImageStudio() { showToast(t('images.stylePresetApplied'), 'success')} /> + {imageToImageMode && ( +
+
+ {t('images.inputImage')} + +
+ {inputImageDataURLs.length > 0 ? ( +
+ {inputImageDataURLs.map((dataURL, index) => ( +
+ {`Input + +
+ ))} +
+ ) : ( + + )} +
+ )} +