GitLab

补齐 gpt-5.4 fallback 的长上下文计费元信息，\n确保超过 272000 输入 token 时对整次会话应用\n2x 输入与 1.5x 输出计费规则。\n\n同时将官方快照 gpt-5.4-2026-03-05 加入前端\n白名单候选与回归测试，避免 whitelist 模式误拦截。\n\nCo-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

(cherry picked from commit d95497af87f608c6dadcbe7d6e851de9413ae147)

- 接入 gpt-5.4 模型识别与规范化，补充默认模型列表
- 增加 gpt-5.4 输入/缓存命中/输出价格与计费兜底逻辑
- 同步前端模型白名单与 OpenCode 上下文窗口（1050000/128000）
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
(cherry picked from commit 924476dcac6181cd0f3ee731ec7b73672ff03793)

LoadFactor should only affect scheduling weight, not load rate reporting.

- Fix bulk edit: send 0 instead of null/NaN to clear load_factor
- Fix edit modal: explicit NaN check instead of implicit falsy
- Fix create modal: use ?? instead of || for load_factor
- Add load_factor upper limit validation (max 10000)
- Add //go:build unit tag and self-contained intPtrHelper in test
- Add design intent comments on WaitPlan.MaxConcurrency

The test could flake depending on exact execution time near midnight
boundaries. Use a range check (29 or 30) instead of exact equality.

- Extract postUsageBilling() to consolidate billing logic across
  GatewayService.RecordUsage, RecordUsageWithLongContext, and
  OpenAIGatewayService.RecordUsage, eliminating ~120 lines of
  duplicated code
- Fix account quota to use TotalCost × accountRateMultiplier
  (was using raw TotalCost, inconsistent with account cost stats)
- Fix RecordUsageWithLongContext API Key quota only updating in
  balance mode (now updates regardless of billing type)
- Fix WebSocket client disconnect detection on Windows by adding
  "an established connection was aborted" to known disconnect errors

Antigravity APIKey accounts were incorrectly routed to
testAntigravityAccountConnection which calls AntigravityTokenProvider,
but the token provider only handles OAuth and Upstream types, causing
"not an antigravity oauth account" error.

Extract routeAntigravityTest to route APIKey accounts to native
Claude/Gemini test paths based on model prefix, matching the
gateway_handler routing logic for normal requests.

- Add configurable spending limit (quota_limit) for apikey-type accounts
- Atomic quota accumulation via PostgreSQL JSONB operations on TotalCost
- Scheduler filters out over-quota accounts with outbox-triggered snapshot refresh
- Display quota usage ($used / $limit) in account capacity column
- Add "Reset Quota" action in account menu to reset usage to zero
- Editing account settings preserves quota_used (no accidental reset)
- Covers all 3 billing paths: Anthropic, Gemini, OpenAI RecordUsage

chore: bump version to 0.1.90.4

When FetchUsageWithOptions receives a non-200 response from the
Anthropic API (e.g. 429 Rate Limited, 401 Unauthorized), the error
was wrapped with fmt.Errorf which infraerrors.FromError cannot
recognize, causing a generic "internal error" message with no details.

Replace fmt.Errorf with infraerrors.New(500, "UPSTREAM_ERROR", msg)
so the upstream error details (status code + body) are included in
the 500 response message. The HTTP status remains 500 to avoid
interfering with frontend auth routing (e.g. 401 would trigger
JWT expiry redirect).

- 移除 ScheduledTestOutcome 中间类型，RunTestBackground 直接返回 *ScheduledTestResult
- 简化 SaveResult 直接接受 *ScheduledTestResult
- 移除 handler 中不必要的 nil 检查
- 移除前端 ScheduledTestsPanel 中多余的 String() 转换
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

将 redis leader lock 逻辑从 service 层抽取为 LeaderLocker 接口，
实现移至 repository 层，消除 service 层对 redis 的直接依赖。
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

每个测试计划绑定一个账号和一个模型，按 cron 表达式定期执行测试，
保存历史结果并在前端账号管理页面中提供完整的增删改查和结果查看功能。

主要变更：
- 新增 scheduled_test_plans / scheduled_test_results 两张表及迁移
- 后端 service 层：CRUD 服务 + 后台 cron runner（每分钟扫描到期计划并发执行）
- RunTestBackground 方法通过 httptest 在内存中执行账号测试并解析 SSE 输出
- Redis leader lock + pg_try_advisory_lock 双重保障多实例部署只执行一次
- REST API：5 个管理端点（计划 CRUD + 结果查询）
- 前端 ScheduledTestsPanel 组件：计划管理、启用开关、内联编辑、结果展开查看
- 中英文 i18n 支持
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

新增 OpenAI WebSocket v2 passthrough relay 数据面与服务适配层，
支持按账号 ws mode 在 ctx_pool 与 passthrough 间路由。

同步调整前端 OpenAI ws mode 选项为 off/ctx_pool/passthrough，
并补充 i18n 文案与对应单测。

新增 Caddyfile.dmit 与 docker-compose-aicodex.yml 部署配置，
用于宿主机场景下的反向代理与服务编排。
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The existing regex only matched the old format where account_uuid is
empty (account__session_). Real Claude Code clients and newer sub2api
generated user_ids use account_{uuid}_session_ which was silently
skipped, causing the original metadata.user_id to leak to upstream
when User-Agent is rewritten by an intermediate gateway.

Closes #766
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

账号首次 401 仅临时不可调度，给予 token 刷新窗口；若恢复后再次 401
说明凭证确实失效，直接升级为错误状态以避免反复无效调度。

- 缓存中 reason 为空时从 DB 回退读取，防止升级判断失效
- ClearError 同时清除临时不可调度状态，管理员恢复后重新给予一次机会
- 管理后台账号列表添加"临时不可调度"状态筛选
- 补充 DB 回退场景单元测试

- 删除 account_today_stats_cache.go 中重复的 normalizeAccountIDList，统一使用 id_list_utils.go 的 normalizeInt64IDList
- 新增 snapshot_cache_test.go：覆盖 snapshotCache、buildETagFromAny、parseBoolQueryWithDefault
- 新增 id_list_utils_test.go：覆盖 normalizeInt64IDList、buildAccountTodayStatsBatchCacheKey
- 新增 ops_query_mode_test.go：覆盖 shouldFallbackOpsPreagg、cloneOpsFilterWithMode

- quota_limited 模式：返回 Key 级别的总额度、速率限制窗口用量和过期时间
- unrestricted 模式：返回订阅限额或钱包余额信息（向后兼容）
- 新增 model_stats 字段，支持 start_date/end_date 参数查询按模型用量统计
- 提取 buildUsageData/parseUsageDateRange 等辅助方法，减少主函数复杂度
- 新增 APIKeyService.GetRateLimitData 和 UsageService.GetAPIKeyModelStats

将中间件职责拆分为鉴权（Authentication）和计费执行（Billing Enforcement）两层：
- 鉴权层（disabled/IP/用户状态）始终执行
- 计费层（过期/配额/订阅/余额）用单一 skipBilling 守卫整块控制

/v1/usage 端点只需鉴权不需计费，skipBilling 仅出现 2 处（订阅加载错误处理 + 计费块守卫），
取代了之前 isUsageQuery 散布在 7 个 if 分支中的控制流。

新增系统设置 allow_ungrouped_key_scheduling（默认关闭），
未分组的 API Key 在网关请求时直接返回 403，
由 RequireGroupAssignment 中间件统一拦截，
支持 Anthropic / Google 两种错误格式响应。

全栈实现：常量 → 结构体 → 解析/更新/初始化 → DTO → 管理接口 →
中间件 → 路由注册 → 前端设置界面 + i18n。

Remove extra space before inline comment to pass golangci-lint gofmt check.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

将 IgnoreCountTokensErrors 默认值从 false 改为 true。

count_tokens 返回 404 是预期业务行为（上游不支持 endpoint，
客户端应 fallback 到本地 tokenizer 估算），不应被视为错误。
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>