github/sub2api
4
0
Fork 0
mirror of https://github.com/Wei-Shaw/sub2api.git synced 2026-10-27 16:07:02 +00:00
Code Issues Packages Projects Releases Wiki Activity
5,220 Commits 49 Branches 64 Tags
e6eb23eaac8f76f60c7fbe2d707416eb4ea4e343
Commit Graph
2783 Commits
Author SHA1 Message Date
song e6eb23eaac feat(openai): add Live gateway support 2026-07-25 12:50:46 +08:00
shaw 6c9b84cc7a feat: 适配 Anthropic 新模型 claude-opus-5 
模型登记：/v1/models 清单、Bedrock 默认映射（us.anthropic.claude-opus-5-v1）、
定价条目（$5/$25 per MTok、1M 上下文、128K 输出）、前端模型清单与
Anthropic/Bedrock 预设映射、限流 scope 简称。

同时修复两个会静默出错的问题：

- 定价家族兜底 3 倍超收：定价数据缺 claude-opus-5 时，matchByModelFamily
  的 Phase 2 关键字兜底会落到 opus-4 系列、getFallbackPricing 会落到
  claude-3-opus，两条路都按 $15/$75 计费（官方 $5/$25），输入输出双双
  3 倍超收且无任何报错。两处补 opus-5 家族并回退到同价的 4.8；判断用
  opus-5/opus5 子串而非裸 "5"，避免误伤 claude-opus-4-5。顺带补齐兜底表
  缺失的 claude-opus-4.8（此前同样会掉到 claude-3-opus）。

- Bedrock 版本闸门降级：claudeVersionRe 强制要求 major-minor 两段版本号，
  只有主版本号的 claude-opus-5 / claude-sonnet-5 完全不匹配，被当成旧模型：
  isBedrockOpus47OrNewer 假导致 thinking.enabled 不转 adaptive（Opus 5 上游
  已移除 budget_tokens，透传直接 400）、isBedrockClaude45OrNewer 假导致
  cache_control.ttl 被剥离、bedrockModelSupportsToolSearch 假导致 tool search
  被过滤。改为 minor 可选（缺省 minor=0），claude-sonnet-5 的同一问题一并修复。

Vertex 无需改动：normalizeVertexAnthropicModelID 只处理 -YYYYMMDD→@YYYYMMDD，
无日期后缀的裸 ID 原样透传即正确。context-1m-2025-08-07 白名单不动：Opus 系
上游不接受该 beta，且 Opus 5 的 1M 上下文是默认能力。

Antigravity 暂不接入：无上游支持证据，mapAntigravityModel 对未映射模型返回
空字符串即"该账号不支持"，fails closed 安全。

回归测试 internal/service/claude_opus5_test.go 覆盖定价两层兜底、Bedrock
三个闸门、thinking 转换与模型清单；逐个回退上述修复已确认测试会红。
 2026-07-25 11:22:42 +08:00
shaw 1be6f30188 fix(ollama): 测试仓储到期查询深拷贝消除数据竞争 
ollamaUsageTestRepo.ListDueOllamaCloudUsageAccounts 返回浅拷贝共享
Extra/Credentials map，RunDue 过滤循环的无锁读与组写协程在 r.mu 下的
map 删改构成数据竞争（-race 可复现；无 -race 时也可能触发 runtime
concurrent map read/write fatal 导致 CI 偶发崩溃）。两条返回路径改为
r.mu 下 mergeMap 深拷贝，并补全 ListOllamaCloudUsageGroupAccounts
只拷 Extra 不拷 Credentials 的半克隆，与基类 GetByID 惯例对齐。
生产仓储每次查询返回全新行，不受影响。
 2026-07-23 17:14:10 +08:00
shaw 2faa0891e4 fix(ollama): 审计日志不落会话明文并收紧凭证清理守卫 
- PUT /admin/accounts/:id/ollama-cloud-usage/session 加入审计整体不入库
  路由，并把裸键 session 纳入键级脱敏兜底，防止浏览器会话 Cookie 明文
  留存 audit_logs.request_body
- UpdateCredentials 的 Ollama 清理分支加顶层 credentials DISTINCT 守卫，
  凭证未变化的持久化不再误清 openai 探测快照或重写 NULL extra
 2026-07-23 16:43:18 +08:00
alfadb 5ac4a9fac2 feat(ollama): 支持 Cloud 官方用量自动刷新 2026-07-23 15:50:44 +08:00
Wesley Liddick 2c76506e07 Merge pull request #4734 from wjx2951874/feat/alipay-mobile-precreate-deep-link 
feat(payment): add mobile Alipay precreate deep link
 2026-07-23 14:06:18 +08:00
Wesley Liddick 6aeea70ee0 Merge pull request #4749 from heathermhuang/codex/fix-openai-proxy-stream-quarantine 
fix(openai): quarantine proxies after incomplete SSE streams
 2026-07-23 11:20:01 +08:00
Wesley Liddick 6f7bad3f2f Merge pull request #4751 from heathermhuang/codex/fix-grok-402-account-cooldown 
fix(grok): cool down accounts after upstream 402
 2026-07-23 11:19:30 +08:00
Wesley Liddick da5c80e113 Merge pull request #4724 from fengshao1227/fix/passthrough-input-normalize 
fix(openai): OAuth 透传路径补齐 input 规范化，修复 Input must be a list
 2026-07-23 11:19:12 +08:00
Wesley Liddick 09b1309c91 Merge pull request #4755 from wucm667/fix/issue-4754-channel-pricing-model-normalization 
fix(billing): normalize channel pricing model names
 2026-07-23 11:18:10 +08:00
shaw ba88cc239c fix(billing): bill composite alias requests by the concrete forwarded model 
Composite public aliases (e.g. all/claude) reach the Anthropic/Gemini
billing core via OriginalModel/ChannelMappedModel source overrides.
Unknown aliases resolved to no pricing and silently recorded $0 cost,
while family-word aliases were mispriced by the fallback family match
(Opus traffic billed at the Sonnet fallback rate). The OpenAI path
already guards this via usageBillingModelCandidates; the shared
recordUsageCore had neither the guard nor a fallback.

- composite groups: unless the admin explicitly configured channel
  pricing for the alias (OpenRouter-style custom pricing), bill by the
  concrete forwarded model
- general safety net: when the selected billing model has no resolvable
  pricing at all, fall back to the concrete forwarded model instead of
  silently recording $0
- grok media usage records now attribute OriginalModel to the client
  requested public alias, consistent with every other endpoint
  (billing unaffected: empty BillingModelSource never triggers source
  overrides)

Priced traffic and non-composite groups are unaffected.
 2026-07-23 10:26:58 +08:00
Heatherm Huang ee332cee64 Fix composite model defaults for linked platforms 2026-07-23 09:20:52 +08:00
Heatherm Huang 2e774a48b3 Align Grok composite example with live model 2026-07-23 09:20:52 +08:00
Heatherm Huang ce3272c41b Build composite subscription bucket two 2026-07-23 09:20:52 +08:00
Heatherm Huang ff666be530 Fix composite route lint issues 2026-07-23 09:20:18 +08:00
Heatherm Huang a008b63c16 Add composite group route registry 2026-07-23 09:20:18 +08:00
Heatherm Huang c8d1e2e16f Harden composite group product surfaces 2026-07-23 09:19:25 +08:00
Heatherm Huang ebc1028771 Add composite group routing 2026-07-23 09:19:24 +08:00
wucm667 44093579e9 fix(billing): normalize channel pricing model names 2026-07-22 20:25:27 +00:00
Heatherm Huang 47ad29db3e fix(openai): quarantine proxies after stream disconnects 2026-07-23 00:12:28 +08:00
Heatherm Huang ca0d3314cf fix(grok): cool down accounts after 402 2026-07-23 00:06:04 +08:00
wjx2951874 7914433011 feat(payment): add mobile Alipay precreate deep link 2026-07-22 19:18:04 +08:00
li 3e26dfa5bb style: 统一 input 包装类型为 []any，与正常 OAuth 路径一致 2026-07-22 16:49:31 +08:00
li 851436c552 fix(openai): OAuth 透传路径补齐 input 规范化，修复 Input must be a list 
openai_passthrough=true 的 OAuth 账号透传路径
(normalizeOpenAIPassthroughOAuthBody) 未对 input 字段做
string→array 转换，导致上游拒绝 "Input must be a list"。

正常 OAuth 路径 (openai_codex_transform.go:290) 已有此逻辑，
透传路径遗漏。补齐三种非数组 input 的规范化：
- string → [{type:"message",role:"user",content:<text>}]
- 空白 string → []
- 单 object → [<object>]
 2026-07-22 16:46:29 +08:00
Wesley Liddick 5e9d75b863 Merge pull request #4719 from superman2003/fix/grok-403-model-sync-4713-4715 
fix(grok): sync OAuth models and isolate policy 403s
 2026-07-22 14:26:07 +08:00
superman2003 ef8275aede fix(grok): sync OAuth models and isolate policy 403s 2026-07-22 13:32:35 +08:00
superman2003 f8445a241b fix(grok): round-trip Codex client tools on Responses 2026-07-22 12:15:48 +08:00
Wesley Liddick ebfaf2496b Merge pull request #4674 from AdrianZhaoDev/main 
feat(groups): add OpenAI reasoning policy
 2026-07-22 09:37:06 +08:00
Wesley Liddick a185705d19 Merge pull request #4684 from superman2003/fix/opencode-codebuddy-cache 
fix(grok): preserve OpenCode and CodeBuddy cache sessions
 2026-07-22 09:36:41 +08:00
Wesley Liddick 003deac540 Merge pull request #4696 from tylzh97/fix/issue-4694-codex-relay 
fix(codex): 保留级联中继的客户端指纹
 2026-07-22 09:36:28 +08:00
Chenhao 8ffa2a31ff fix(codex): 保留级联中继的客户端指纹 2026-07-21 05:02:30 -07:00
superman2003 e31783e443 fix(grok): preserve OpenCode and CodeBuddy cache sessions 2026-07-21 15:22:12 +08:00
superman2003 bd62c6f66c fix(monitor): stop scheduling after API key decryption failure 2026-07-21 11:55:38 +08:00
zhaozewu 6c93f01c97 fix(groups): tighten reasoning policy 
Remove unsupported none handling, align the auth cache version with the current baseline, consolidate the unshipped migrations, and update the public API contract.
 2026-07-21 11:16:04 +08:00
zhaozewu 6af622c340 feat(groups): add OpenAI reasoning policy 
Persist reasoning ceilings and exact mappings for OpenAI groups, enforce them across HTTP and WebSocket forwarding, and invalidate cached auth snapshots.
 2026-07-21 11:02:43 +08:00
Wesley Liddick 04a61ee6ad Merge pull request #4667 from jianjianai/new/perf-scheduler-snapshot-id-members-upstream 
优化快照发布临时分配
 2026-07-21 10:43:37 +08:00
Wesley Liddick 1f9eac4fbd Merge pull request #4660 from superman2003/fix/openai-scheduler-exclusion-observability-20260720 
fix(scheduler): 高级调度器无可用账号错误补充排除原因统计，可自诊断静默剔除 (#4599)
 2026-07-21 10:43:21 +08:00
Wesley Liddick 0b9d44545b Merge pull request #4641 from abbzbb/pr/grok-responses-compact 
feat(grok): 基于 Grok Responses 实现 /responses/compact 适配 (#4554)
 2026-07-21 10:43:04 +08:00
Wesley Liddick 927fcbc1e3 Merge pull request #4645 from fengshao1227/fix/responses-hosted-image-gen-billing 
fix(billing): 合并 hosted image_generation 工具的图片 token 到 /responses 计费
 2026-07-21 10:42:46 +08:00
Wesley Liddick 14701ff070 Merge pull request #4664 from wucm667/fix/issue-4656-sync-cache-billing 
fix: align sync cache billing after failover
 2026-07-21 10:42:07 +08:00
Wesley Liddick e85878624a Merge pull request #4657 from suuuuuu-1/fix/sse-field-parsing-compatibility 
fix(gateway): support compact SSE format in Anthropic /responses bridge
 2026-07-21 10:41:50 +08:00
Wesley Liddick 8b4c8823d9 Merge pull request #4658 from superman2003/fix/cc-bridge-compact-sse 
fix(gateway): Chat Completions 桥接受 Anthropic 上游的 SSE 紧凑格式
 2026-07-21 10:41:33 +08:00
jjaw af3b6bbd68 优化快照发布临时分配 2026-07-21 03:07:01 +08:00
wucm667 addd5ef1dc [verified] fix: align sync cache billing after failover 2026-07-20 22:43:11 +08:00
superman2003andCursor a319333161 fix(lint): summary 检查 strings.Builder WriteString 返回值 (errcheck) 
Co-authored-by: Cursor <cursoragent@cursor.com>
 2026-07-20 21:52:46 +08:00
superman2003andCursor 29bea0a758 fix(scheduler): 高级调度器无可用账号错误补充排除原因统计 
问题背景（Refs #4599）：OpenAI OAuth 账号请求 /v1/responses 全模型 503，
日志仅有 no available OpenAI accounts supporting model 且
excluded_account_count=0。selectByLoadBalance 初筛循环的每个过滤分支
（配额自动暂停、渠道上游模型限制、运行时熔断等）都是静默 continue，
最多只有 Debug 级日志，用户与维护者都无法定位候选账号被哪个隐式
过滤点剔除。

修改：
- selectByLoadBalance 初筛循环为每个 continue 分支记录原因计数
  （excluded/not_schedulable/platform_mismatch/runtime_blocked/
  privacy_not_set/transport_incompatible 及请求兼容性细分原因）
- isAccountRequestCompatible 新增 isAccountRequestCompatibleReason
  变体，细分 runtime_blocked/quota_auto_pause_<窗口>/
  shadow_parent_unhealthy/model_not_supported/
  channel_upstream_restricted/capability_mismatch；原 bool 函数改为
  包装变体，其余调用点行为不变
- noAvailableOpenAISelectionError 增加 details 参数，初筛全灭时输出
  如 (pool=3, filtered: excluded=1 model_not_supported=1
  quota_auto_pause_7d=1)，原因按字典序排序保证输出确定；快照池为空
  时输出 (pool=0)；选择序列耗尽路径追加 selection_order_exhausted
  标注；legacy 调用点传空保持原消息不变

兼容性：错误仍通过 Unwrap 匹配 ErrNoAvailableAccounts，分类与 ops
标记逻辑不变；客户端可见错误仍为通用分类消息，统计仅进服务端日志；
happy path 原因 map 惰性分配，无额外开销。

Refs #4599

Co-authored-by: Cursor <cursoragent@cursor.com>
 2026-07-20 21:44:42 +08:00
superman2003 2707f2793d fix(gateway): CC 桥接受 Anthropic 上游的 SSE 紧凑格式 
SSE 规范允许字段冒号后省略空格（event:message_start / data:{...}）。
Kimi 等 Anthropic 兼容上游返回紧凑格式时，Anthropic → Chat Completions
桥按 "event: " / "data: " 严格匹配会丢弃全部事件，且 dataLine[6:] 按带
空格前缀硬切载荷，最终流式/缓冲两条路径都报 "Upstream stream ended
without a response"（与 #4653 同根因；#4657 只修了 /v1/responses 桥，
本补丁补齐对称的 /v1/chat/completions 桥）。

复用 OpenAI 链路既有的 extractOpenAISSEEventLine / extractOpenAISSEDataLine
（正确处理可选空格/tab），流式与缓冲两处统一替换；补充紧凑格式的
缓冲/流式回归测试，原带空格格式用例保持通过。
 2026-07-20 21:00:29 +08:00
suuuuuu-1 e5ad287843 fix(gateway): support compact SSE format in Anthropic /responses bridge 
Fixes #4653

## Problem

The /v1/responses endpoint (Responses → Anthropic bridge) failed when
upstream providers returned SSE events in compact format without spaces
after colons (e.g. 'event:message_start' instead of 'event: message_start').

This affected Kimi API (api.kimi.com) and potentially other upstreams,
causing 'Upstream stream ended without a response' errors.

## Root Cause

The SSE parser in gateway_forward_as_responses.go used strict prefix
matching with hardcoded spaces:

    if !strings.HasPrefix(line, "event: ") { ... }
    if !strings.HasPrefix(dataLine, "data: ") { ... }

According to the SSE spec
(https://html.spec.whatwg.org/multipage/server-sent-events.html),
the space after the colon is optional. Both formats are valid:

    event: message_start  ✓
    event:message_start   ✓

## Solution

Introduced parseAnthropicSSEField() helper that:
- Accepts both 'field: value' and 'field:value' formats
- Uses TrimSpace() to handle optional/multiple spaces
- Returns (value, ok) for consistent error handling

Applied to both streaming handlers:
- handleResponsesBufferedStreamingResponse
- handleResponsesStreamingResponse

## Testing

Added comprehensive unit tests:
- TestParseAnthropicSSEField: validates parser with 8 edge cases
- TestHandleResponsesBufferedStreamingResponse_CompactSSEFormat
- TestHandleResponsesStreamingResponse_CompactSSEFormat

All existing tests pass, confirming backward compatibility.
 2026-07-20 20:40:08 +08:00
superman2003andCursor c401595e01 fix(openai): preserve 路径 call_id 超长时压缩至 64 字符上限 
生产环境真实报错：客户端走 /v1/messages（Anthropic Messages 格式）入站，
OAuth 账号转发上游 OpenAI /v1/responses 时（PreserveToolCallIDs=true），
客户端历史中出现超长工具 ID（如嵌套代理拼接产生的 78 字符 srvtoolu_ ID）
会被原样透传，上游 400 拒绝：
"Invalid 'input[614].call_id': string too long. Expected a string with
maximum length 64, but got a string with length 78 instead."

#4612 为非 preserve 路径增加了 call_id 归一化与压缩，但有意将
PreserveToolCallIDs=true 的路径排除在外。本次在 preserve 路径补上长度护栏：
- len(id) <= 64 时仍原样透传，保持 #4612 承诺的"保持原样"语义；
- 超过 64 时复用已有的 compactCodexCallID 做确定性 SHA-256 压缩。

由于压缩是确定性的，同一逻辑 ID 在 function_call 与 function_call_output
两侧结果一致，tool_use/tool_result 配对不破坏；OAuth 路径 store=false
每轮全量重放历史，跨轮结果亦一致。改写点 fixCallIDPrefix 是
filterCodexInputWithOptions 内所有 call_id 修正的单一入口（含 item_reference
的 call_* 引用），同时覆盖 Messages 桥接（openai_gateway_messages.go）与
/v1/responses 透传（openai_gateway_forward.go）两条 preserve 路径。

Co-authored-by: Cursor <cursoragent@cursor.com>
 2026-07-20 18:49:24 +08:00
li 865128998c fix(billing): 合并 hosted image_generation 工具的图片 token 到 /responses 计费 
/v1/responses 端点使用 hosted image_generation 工具时，上游在
response.tool_usage.image_gen 返回图片 token 用量，但
extractOpenAIUsageFromJSONBytes 只读 response.usage，导致
image_output_tokens 和 image_output_cost 始终为 0。

在 extractOpenAIUsageFromJSONBytes 提取完 base usage 后，合并
tool_usage.image_gen 的 image tokens 维度（仅补充不覆盖），
覆盖 SSE 流式、非流式和 WebSocket 三条路径。
 2026-07-20 16:29:12 +08:00