GitLab

Click on a group name, model name, or endpoint name in the distribution
tables to expand and show per-user usage breakdown (requests, tokens,
actual cost, standard cost).

Backend: new GET /admin/dashboard/user-breakdown API with group_id,
model, endpoint, endpoint_type filters.
Frontend: clickable rows with expand/collapse sub-table in all three
distribution charts.

Antigravity streaming handlers were missing the keepalive mechanism
that exists in the standard gateway, causing proxy/CDN idle timeouts
to break connections during long thinking phases (e.g. claude-opus-4-6).
This resulted in truncated responses with missing tool calls.

Add StreamKeepaliveInterval support to all three Antigravity streaming
paths: Claude SSE, Gemini SSE, and upstream passthrough.

Removes hasMeaningfulWindowStats guard so the /usage endpoint consistently
returns WindowStats for both time windows. The frontend now controls
zero-value display filtering at the component level.

UpdateAccount previously required len(input.Extra) > 0, causing explicit
empty payloads (extra:{}) to be silently skipped. Change condition to
input.Extra != nil so clearing quota keys actually persists.

The updateRateLimitUsageScript Lua script previously performed
unconditional HINCRBYFLOAT on all usage counters without checking
whether the rate limit window had expired. This caused usage to
accumulate across window boundaries in Redis while the DB correctly
reset on expiration, leading to incorrect 429 rate limiting that
could persist for up to 24 hours.

The Lua script now checks each window timestamp before incrementing:
- If the window has expired, usage is reset to the current cost and
  the window timestamp is updated (matching DB-side semantics)
- If the window is still valid, usage is accumulated normally

This also resolves the async race condition where stale HINCRBYFLOAT
tasks from the worker queue could pollute a freshly rebuilt cache
after invalidation, since the script now self-corrects expired windows.

Closes #1049

正常终端事件（response.completed 等）退出后连接归还复用，
仅异常路径（读写错误、error 事件、客户端断连）MarkBroken 销毁。

Generate 模式:
- 引入 cleanExit 标记，仅在 isTerminalEvent break 时设置 true
- defer 中根据 cleanExit 决定是否 MarkBroken
- 所有异常路径已在各自分支中提前调用 MarkBroken

Ingress 模式:
- 引入 lastTurnClean 标记，sendAndRelay 正常完成时设为 true
- releaseSessionLease 根据 lastTurnClean 决定是否 MarkBroken
- 错误路径重置 lastTurnClean = false
- 客户端断连后 drain 仍保守 MarkBroken（L2916）

根因：多个用户共享同一 OAuth 账号时，conversation_id/session_id 头
未做用户隔离，导致上游 chatgpt.com 将不同用户的请求关联到同一会话。

HTTP SSE 修复:
- 新增 isolateOpenAISessionID(apiKeyID, raw)，将 API Key ID 混入
  session 标识符（xxhash），确保不同 Key 的用户产生不同上游会话
- buildUpstreamRequest: OAuth 分支先 Del 客户端透传的 session 头，
  再用隔离值覆盖
- buildUpstreamRequestOpenAIPassthrough: 透传路径同样隔离
- ForwardAsAnthropic: Anthropic Messages 兼容路径同步修复
- buildOpenAIWSHeaders: WS 路径的 OAuth session 头同步隔离

- Fix gofmt alignment in admin_service.go and trailing newline in
  antigravity_credits_overages.go
- Suppress errcheck for fmt.Sscanf in client.go GetMinimumAmount

Remove SimulateClaudeMaxEnabled field and related logic from
admin_service.go, and remove applyClaudeMaxCacheBillingPolicyToUsage,
applyClaudeMaxNonStreamingRewrite, setupClaudeMaxStreamingHook calls
from antigravity_gateway_service.go. These symbols are not yet
available in upstream/main.

Replace process-memory sync.Map + per-model runtime state with a single
"AICredits" key in model_rate_limits, making credits exhaustion fully
isomorphic with model-level rate limiting.

Scheduler: rate-limited accounts with overages enabled + credits available
are now scheduled instead of excluded.

Forwarding: when model is rate-limited + credits available, inject credits
proactively without waiting for a 429 round trip.

Storage: credits exhaustion stored as model_rate_limits["AICredits"] with
5h duration, reusing SetModelRateLimit/isRateLimitActiveForKey.

Frontend: show credits_active (yellow ⚡) when model rate-limited but
credits available, credits_exhausted (red) when AICredits key active.

Tests: add unit tests for shouldMarkCreditsExhausted, injectEnabledCreditTypes,
clearCreditsExhausted, and update existing overages tests.

feat: implement resolveCreditsOveragesModelKey function to stabilize model key resolution for credit overages

Introduce OAuthRefreshAPI as the single entry point for all OAuth token
refresh operations, eliminating the race condition where background
refresh and inline refresh could simultaneously use the same
refresh_token (fixes #1035).

Key changes:
- Add OAuthRefreshExecutor interface extending TokenRefresher with CacheKey
- Add OAuthRefreshAPI.RefreshIfNeeded with lock → DB re-read → double-check flow
- Add ProviderRefreshPolicy / BackgroundRefreshPolicy strategy types
- Simplify all 4 TokenProviders to delegate to OAuthRefreshAPI
- Rewrite TokenRefreshService.refreshWithRetry to use unified API path
- Add MergeCredentials and BuildClaudeAccountCredentials helpers
- Add 40 unit tests covering all new and modified code paths

- Apply InboundEndpointMiddleware to all gateway route groups
- Replace normalizedOpenAIInboundEndpoint/normalizedOpenAIUpstreamEndpoint and normalizedGatewayInboundEndpoint/normalizedGatewayUpstreamEndpoint with GetInboundEndpoint/GetUpstreamEndpoint
- Remove 4 old constants and 4 old normalization functions (-70 lines)
- Migrate existing endpoint normalization test to new API

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode

)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>

Introduce endpoint.go with shared constants, NormalizeInboundEndpoint, DeriveUpstreamEndpoint, InboundEndpointMiddleware, and context helpers. This replaces the two separate normalization implementations (OpenAI and Gateway) with a single source of truth. Includes comprehensive test coverage.

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode

)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>

Extend RecordUsageInput and RecordUsageLongContextInput structs with InboundEndpoint and UpstreamEndpoint so that Claude, Gemini, and Sora handlers can record endpoint info alongside OpenAI handlers.

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode

)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>

Replace t.Add(24*time.Hour - time.Nanosecond) with t.AddDate(0, 0, 1) and use SQL < instead of <= for end-of-day boundaries. This avoids edge-case misses around DST transitions.

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode

)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>

The upstream Gemini API returns "Insufficient account balance" which
doesn't contain the substring "insufficient balance". Add explicit
match for the full phrase to ensure the filter works correctly.

修复在填写限额时，如果不填写完整的三个限额额度（日限额、周限额、月限额）就会报错的问题。

变更内容：
- 后端：添加 optionalLimitField 类型处理空值和空字符串，兼容部分限额字段为空的情况
- 前端：添加 normalizeOptionalLimit 函数规范化限额输入，将空值、空字符串和无效数字统一处理为 null

- Add 5th error filter switch IgnoreInsufficientBalanceErrors to suppress
  upstream insufficient balance / insufficient_quota errors from ops log
- Extract hardcoded error strings into package-level constants for
  shouldSkipOpsErrorLog, normalizeOpsErrorType, classifyOpsPhase, and
  classifyOpsIsBusinessLimited
- Define ErrNoAvailableAccounts sentinel error and replace all
  errors.New("no available accounts") call sites
- Update tests to use require.ErrorIs with the sentinel error

Previously, v-model.number produced "" when input was cleared, causing
JSON decode errors on the backend. Also, normalizeLimit treated 0 as
"unlimited" which prevented setting a zero quota. Now "" is converted
to null (unlimited) in frontend, and 0 is preserved as a valid limit.

Closes Wei-Shaw/sub2api#1021

When Redis cache is populated from DB with a NULL window_1d_start, the
Lua increment script only updates usage counters without setting window
timestamps. IsWindowExpired(nil) previously returned false, so the
accumulated usage was never reset across time windows, effectively
turning usage_1d into a lifetime counter. Once this exceeded
rate_limit_1d the key was incorrectly blocked with "日限额已用完".

Fixes Wei-Shaw/sub2api#1022

Claude's output_config.effort parameter (low/medium/high/max) was not
being extracted from requests or logged in the reasoning_effort column
of usage logs. Only the OpenAI path populated this field.

Changes:
- Extract output_config.effort in ParseGatewayRequest
- Add ReasoningEffort field to ForwardResult
- Populate reasoning_effort in both RecordUsage and RecordUsageWithLongContext
- Guard against overwriting service-set effort values in handler
- Update stale comments that described reasoning_effort as OpenAI-only
- Add unit tests for extraction, normalization, and persistence

将入站、上游与路径三类端点分布统一到使用记录页的一致化卡片交互中，并补齐端点元数据与统计链路，提升排障与流量分析效率。

当上游在 SSE 流中途返回 event:error 时，handleStreamingResponse 已将
部分 SSE 事件写入客户端，但原先的 failover 逻辑仍会切换到下一个账号
并写入完整流，导致客户端收到两个 message_start 进而产生 400 错误。

修复方案：在每次 Forward 调用前记录 c.Writer.Size()，若 Forward 返回
UpstreamFailoverError 后 writer 字节数增加，说明 SSE 内容已不可撤销地
发送给客户端，此时直接调用 handleFailoverExhausted 发送 SSE error 事件
终止流，而非继续 failover。

Ping-only 场景不受影响：slot 等待期的 ping 字节在 Forward 前后相等，
正常 failover 流程照常进行。
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

applyUsageBillingEffects() 中配额更新条件仅检查了 AccountTypeAPIKey，
遗漏了 AccountTypeBedrock，导致 Bedrock 账户的配额计数器永远不递增。
扩展条件以同时支持 APIKey 和 Bedrock 类型。

同时在前端账户筛选下拉框中添加 AWS Bedrock 选项。

Increase MAX(bucket_start) query timeout from 3s to 5s to reduce
timeout-induced fallbacks. Shrink backfill window from 30 days to
1 hour so that fallback recomputation stays lightweight instead of
scanning the entire retention range.

1. S3 凭证加密存储：使用 SecretEncryptor (AES-256-GCM) 加密 SecretAccessKey，
   防止备份文件中泄露 S3 凭证，兼容旧的未加密数据
2. 修复 saveRecord 竞态条件：添加 recordsMu 互斥锁保护 records 的 load/save
3. 恢复操作增加服务端验证：handler 层要求重新输入管理员密码，通过 bcrypt
   校验，前端弹出密码输入框
4. pg_dump/psql/S3 操作抽象为接口：定义 DBDumper 和 BackupObjectStore 接口，
   实现放入 repository 层，遵循项目依赖注入架构规范
5. 改为流式处理避免大数据库 OOM：备份时 pg_dump stdout -> gzip -> io.Pipe ->
   S3 upload；恢复时 S3 download -> gzip reader -> psql stdin，不再全量加载
6. loadRecords 区分"无数据"和"数据损坏"场景：JSON 解析失败返回明确错误
7. 添加 18 个核心逻辑单元测试：覆盖加密、并发、流式备份/恢复、错误处理等
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>