Commit Graph
4176 Commits
Author SHA1 Message Date
hansnow 7e579cb28d fix(openai): adapt client tools in WS HTTP bridge 2026-08-18 14:42:30 +08:00
Wesley Liddick 938f1868ae Merge pull request #5714 from wucm667/fix/issue-2733-current-main
fix(ops): avoid single-insert fallback after batch failure
2026-08-18 14:11:42 +08:00
Wesley Liddick 4d19836189 Merge pull request #5716 from wucm667/fix/issue-2695-current-main
fix: skip expiry reminders without SMTP config
2026-08-18 14:06:32 +08:00
Wesley Liddick 1ea4150bf0 Merge pull request #5581 from wucm667/fix/issue-5574-passthrough-model-discovery
fix(gateway): align passthrough model discovery
2026-08-18 13:56:02 +08:00
Wesley Liddick ed2da82396 Merge pull request #5711 from wucm667/fix/issue-5709-antigravity-tool-config
fix(antigravity): preserve mixed Gemini tool config
2026-08-18 13:55:24 +08:00
Wesley Liddick 6259940ef2 Merge pull request #5669 from feeeei/main
feat(openai): OpenAI Team 联动熔断
2026-08-18 13:54:32 +08:00
Wesley Liddick baaf59d417 Merge pull request #5755 from feeeei/feat/gemini
fix(gemini): Skipped 错误策略对齐 OpenAI,上游 4xx 不再硬改 500
2026-08-18 13:54:06 +08:00
Wesley Liddick c0325d24f9 Merge pull request #5759 from o2e/codex/fix-codex-usage-probe-model-clean
[codex] 修复部分账号 Codex 额度查询 400
2026-08-18 13:53:49 +08:00
Wesley Liddick 1a3ecd2b94 Merge pull request #5004 from wucm667/fix/issue-4990-deferred-tool-cache-control
fix(claude): strip cache control from deferred tools
2026-08-18 13:50:57 +08:00
Wesley Liddick cddb03c0f1 Merge pull request #5609 from wucm667/fix/issue-5607-auth-pricing-snapshot
fix: preserve group pricing in auth snapshots
2026-08-18 13:50:13 +08:00
Wesley Liddick a7a321232f Merge pull request #5567 from wucm667/fix/issue-5563-anthropic-sse-overload
fix(gateway): handle Anthropic SSE overload errors
2026-08-18 13:49:59 +08:00
o2e 16e4f7ecc3 修复 Codex 额度探针模型兼容性 2026-08-18 13:08:28 +08:00
feeeei ab0fcd1a0e fix(gemini): Skipped 错误策略对齐 OpenAI,上游 4xx 不再硬改 500
ErrorPolicySkipped(池模式、或自定义错误码未命中)原来在响应写出上
自成一派:v1beta 原生把上游 4xx 硬改 500 后原文透传,/v1/messages
硬传 500 进映射(客户端拿到 502)。下游网关据此把请求级错误当可重
试的服务端故障反复换号,耗尽后改写成 All available accounts
exhausted(2026-08-17 gemini 生产事故链)。现对齐 OpenAI 路径语义:

- Skipped 只豁免账号状态标记,不豁免换号:可 failover 状态码一律
  返回 UpstreamFailoverError(poolModeSkippedFailoverError 泛化为
  skippedErrorPolicyFailoverError,同账号重试标记仍仅池模式携带)
- 池模式的不可 failover 4xx 保真:v1beta 原码+原文透传(新
  writeGeminiNativeUpstreamError 与 ErrorPolicyNone 共用同一写出,
  并补记 ops 事件),/v1/messages 与 chat completions 按真实状态
  码映射
- 自定义错误码未命中且不可 failover:三路径统一 500 + "Upstream
  gateway error" 固定文案,上游细节仅记 ops 错误日志
- 400 属确定性请求错误:mapped 写出回传脱敏后的上游 message,客户
  端可据此定位非法字段
2026-08-18 11:24:39 +08:00
Wesley Liddick 7d633f5fc3 Merge pull request #5742 from heathermhuang/codex/fix-grok-response-model-audit
fix: normalize Grok response model audit aliases
2026-08-18 10:03:26 +08:00
Wesley Liddick b2d1c3859a Merge pull request #5738 from okbexx/fix/codex-identity-snapshot
fix(openai): make Codex convergence identity consistent
2026-08-18 09:57:05 +08:00
shaw 6bf335965a merge main 并修复与 #5730 的语义冲突
main 侧 #5730 新增的 openai_gateway_cn_fixes_test.go 按旧 11 参签名调用
calculateOpenAIRecordUsageCost;本分支为该函数新增了第 12 个参数
pricingAt。文本无冲突但 test build 会失败,此处按本分支对同类测试
调用点的既有处理方式补传 time.Time{}。
2026-08-17 22:27:22 +08:00
Heatherm Huang c46d07ca07 fix: normalize Grok response model audit aliases 2026-08-17 21:15:12 +08:00
Jarl 6793d5ac85 fix(openai): make Codex convergence identity consistent 2026-08-17 20:25:28 +08:00
lyen1688 9f24a55305 功能:支持渠道模型分时倍率定价 2026-08-17 19:45:07 +08:00
shaw 10c8b70203 fix(cn-providers): 修复 CN 分组五项功能缺陷(调度闸门/计费/断开漏记/count_tokens/403)
对已合并 PR #5666 + 分组入口放行后的全量功能审计发现的 P0/P1 修复,
全部先经代码与厂商文档实证再实施:

1. /v1/messages 调度闸门(P0):sanitizeGroupMessagesDispatchFields 对非
   openai 平台恒置 AllowMessagesDispatch=false,而闸门豁免名单只有 grok,
   CN 分组经正常途径创建后恒 403——原生 Anthropic 直通(Claude Code 主用例)
   完全不可达。修复:闸门对 CN 与 grok 同语义豁免;count_tokens 处的内联
   裸检查统一走同一 helper;ResolveMessagesDispatchModel 对 CN 早退,避免
   openai 专属的 gpt-5.x 默认映射发给 CN 上游。

2. 计费候选链(P0):候选链兜底含客户端原始模型名,配合 getFallbackPricing
   的 claude→Sonnet 统一兜底,映射的 CN 模型无价时 CN 流量会按 Claude 原价
   (数倍~数十倍)静默误计,且 usage 日志显示 claude-* 名无从察觉。修复:
   CN 账号的 claude-* 候选仅在显式分组/渠道定价时放行;候选全滤空时按
   ErrModelPricingUnavailable 走零成本+告警落账(顺带修复原空候选错误会
   丢弃整条 usage 记录的次生问题)。

3. 断开/中断漏记(P0,#5148 对齐,惠及 openai 平台):messages/responses/
   chat_completions 三个 handler 的错误路径此前在 err!=nil 时丢弃携带的
   部分 result——客户端断开排水后的完整 usage 被丢,payg 上游照常计费而
   平台漏记(anthropic 网关早有同修复,openai 网关缺失)。修复:错误路径
   result 非空时照常提交 usage;failover 错误恒 result=nil 无重复计费。
   Responses×anthropic 流式转换器同时改为断开后继续排水至流自然结束
   (末尾 message_delta 的 output_tokens 不再丢),finalize 帧补工具名反转
   与客户端工具还原、仅在客户端仍连接时写出。

4. count_tokens(P1,证据修正):经实证三家 Anthropic 兼容层均无
   /v1/messages/count_tokens(DeepSeek 官方文档无此端点且注明
   anthropic-version 被忽略;OpenModel 标注该端点 Anthropic only),
   anthropic 协议转发上游=常态 404,且错误处置缺模型上下文会把不计费的
   探测放大成整账号停调。修复:CN 全协议一律本地 tiktoken 估算(与 Grok
   同方案),删除上游转发死代码。

5. 403 处置(P1):CN 此前落入通用 handleAuthError,单次 HTML 403(CDN/
   代理拦截页)即永久禁用,且 403 在 failover 集里会逐账号重放连环禁用
   整组。修复:CN 与 openai 同口径——HTML 豁免 + 3 次累计 + 临时冷却。

新增回归测试 8 项:闸门豁免(含 openai 仍受控断言)、CN 调度映射空返回、
候选过滤三态、空候选零成本落账、断开排水 usage 完整性、HTML-403 零处罚、
结构化 403 首次临时停调。handler/service 全包测试通过。
2026-08-17 17:28:53 +08:00
shaw 7cdca9e495 feat(groups): 放行 kimi/zhipu/deepseek 平台分组创建入口
PR #5666 引入 CN 平台后,路由/调度/前端类型均已支持 CN 平台分组,但分组
创建入口两头缺失:后端 Create/UpdateGroupRequest 的 platform oneof 白名单
与前端 GroupsView 平台选项都没有三平台,导致 CN 账号「无可用分组」、整条
流量链路不通(composite 不能作为替代:CN 不可为 composite 路由目标)。

- group_handler.go: 两处 oneof 加 kimi/zhipu/deepseek;composite 路由目标
  白名单有意不动(DetectModelPlatform/isConcreteRequestPlatform 均无 CN 分支)
- GroupsView: platformOptions/platformFilterOptions 补三项;两处徽章配色链
  按 platformColors.ts 色系补 CN 分支
- i18n: admin.groups.platforms 补 kimi/zhipu/deepseek 键(zh/en),缺键时
  分组徽章/GroupRPM/RateMultipliers 弹窗/ChannelsView 会渲染原始 key
- GroupBadge: badgeClass/labelClass 补 CN 配色
- 新增表驱动测试:9 平台 Create/Update 全放行、非法值(别名/大小写/空格)
  全拒绝、composite target 对 CN 保持拒绝的守卫
2026-08-17 17:28:51 +08:00
wucm667 971544570d test(antigravity): check tool config assertions 2026-08-17 17:11:33 +08:00
wucm667 cb5e03a720 fix(antigravity): preserve mixed Gemini tool config 2026-08-17 12:36:58 +08:00
Randark 4b667ccd45 fix(review): 处理 PR #5666 四个阻断项(B1-B4)
- B1: 移除根目录 docker-compose.yml(个人镜像测试产物混入)
- B2: 新增迁移 224 放宽 user_platform_quotas CHECK 至 8 平台,补
  BulkInsertInitial 国产平台集成测试与迁移内容断言测试
- B3: CC/Responses×anthropic 四个上游读循环接入间隔超时泵
  (gateway.stream_data_interval_timeout,默认 180s):上游挂住 SSE
  不发数据也不断连时结束排水/组装、关闭 resp.Body 归还连接池位,
  排水超时仍按已累计 usage 返回(与 messages 主路径同语义)
- B4: 配额/余额探测发起前过 cnValidateProbeURL(与网关转发同一套
  security.url_allowlist 策略),被拒时零出站、API key 不离开本机
2026-08-17 04:07:06 +00:00
wucm667 79c2eb5020 fix: skip expiry reminders without SMTP config 2026-08-17 10:59:43 +08:00
wucm667 5f19433103 fix(ops): avoid single-insert fallback after batch failure 2026-08-17 10:52:57 +08:00
Randark 901a0439f1 feat: 国产供应商一等支持(Kimi/Zhipu/DeepSeek 多协议 + 配额/余额监控)
后端:
- 协议凭证维度 credentials[api_protocol] ∈ chat_completions(默认)/anthropic/responses(deepseek)
- /v1/messages 零转换直通原生 Anthropic 端点(kimi/zhipu/deepseek),CC/Responses
  入站交叉组合走 apicompat 双向转换链(responses/chat_completions anthropic-native 转发器)
- count_tokens:anthropic 协议透传原生端点;其余 CN 协议本地 tiktoken 估算
- Coding Plan 额度探测(5h/weekly 滚动窗口)+ payg 余额探测(kimi/deepseek),
  deepseek 双币种 CNY+USD 明细,任一币种达标不停调
- 周期任务 [CNBalance] 并发探测 + 预算随工作量放大;响应式 429 冷却到最早窗口
  重置点;余额不足可恢复临时停调;智谱 CREDIT_LIMIT 不污染窗口解析
- CC→anthropic 流式客户端断开后继续排水上游保住 usage 计量

前端:
- 创建/编辑弹窗 account_mode + api_protocol + base_url 联动预设(含 watcher 竞态防护)
- 用量单元格:kimi/zhipu coding 显示 5h/weekly 窗口,kimi/deepseek payg 显示余额,
  多币种并列展示;探测失败保留快照;挂载自动探测 5min 去抖
- 调度阈值设置面板补 kimi/zhipu 平台(对齐后端 AllowedSchedulingThresholdPlatforms)
2026-08-15 10:37:51 +00:00
feeeei d677d67dda feat: OpenAI Team 联动熔断
OpenAI OAuth 账户收到 402 deactivated_workspace(ChatGPT Team 工作区被停用)时,
将同一 Team(credentials.chatgpt_account_id 相同)的其余 active 账户一并置为
error 并进程内立即熔断,让调度瞬间切走,避免流量继续打到已死的兄弟账户。

- 触发条件仅限 402 + detail.code == deactivated_workspace;通用 402 与 401 行为不变
- 调用点两处:fastpath 各类早退之前 + HandleUpstreamError 顶部,进程内 60s 按 Team 去重
- 逐账户 SetError(错误信息标注触发账户),单账户失败不中断其余
- 默认生效,无配置开关
2026-08-15 16:48:40 +08:00
shaw 4d9fedee20 fix(test): check type assertions in turn-state provenance tests
errcheck runs with check-type-assertions enabled, so the single-value
assertions on the provenance map values fail lint.
2026-08-15 16:44:30 +08:00
shaw fce41e318f fix(openai): make Codex fingerprint convergence opt-in and cover passthrough
Default codex_fingerprint_mode to off. v0.1.175 treated a missing key as
"session", so upgrading silently rewrote installation/session/thread/turn/
window identifiers for every existing OAuth account that had never configured
this field. The quota regressions in #5555, #5556 and #5582 line up with that
version boundary, with A/B reports that rolling back to v0.1.173 restores
quota. Convergence is now explicit opt-in (#5610).

Only accounts that never set the field change behaviour; explicit off /
device / session / full keep working exactly as configured. That required
flipping the persistence condition in all three account modals from
"!== 'session'" to "!== 'off'": the old rule deleted the key when it equalled
the default, which after the flip would have silently discarded an
administrator's explicit opt-in to session.

Also extend convergence to the passthrough path, which previously left client
identifiers untouched:

- resolve the ids once in forwardOpenAIPassthrough and rewrite
  client_metadata on the raw bytes (gjson extract + sjson splice) because
  passthrough is a hot path that must not fully unmarshal multi-MB bodies;
  a shared core keeps the raw and map variants from drifting
- both request builders apply the staged ids at the same relative position
  (after session isolation, before identity enforcement) so headers and body
  share one id set and turn_id stays consistent
- stage the ids unconditionally, including nil: a failover from a converged
  account to an off account must not leave the previous account's ids behind
2026-08-15 16:35:26 +08:00
shaw 8ae6d8f67e fix(openai): send session-level beta features and probe native compaction v2
OpenAI sunset the legacy unary /responses/compact endpoint (404, #5598,
#5624), so the account "compact probe" in the admin UI kept failing even for
healthy accounts, and the beta-feature negotiation header was only attached
to compaction turns.

Beta features (codex-rs session/mod.rs build_model_client_beta_features_header
+ client.rs build_responses_headers): the header is a session-level constant
attached to every /responses request, the WS handshake and /responses/compact.
Enumerating FEATURES shows no Experimental feature is enabled by default, so a
default install sends exactly "remote_compaction_v2". Mirror that:

- OAuth requests without a client-declared header get the default shape, so we
  no longer produce a "header only on compaction turns" pattern real Codex
  never emits (#5586 chains that strip the header)
- a client-declared header is preserved as-is: non-empty without v2 means the
  user disabled the feature and the gateway must not rewrite that
- native v2 turns (compaction_trigger in body) always ensure v2 is present
- non-OAuth upstreams keep the compaction-turn-only behaviour
- the WS injection sits outside the client-header copy block so prewarm and
  turn handshakes cannot land in different pool compatibility buckets

Compact probe now exercises native v2 (streaming /responses +
compaction_trigger) instead of the dead endpoint. Success requires an actual
compaction output item — scanning output_item.done/added, the terminal
response.output[] and the whole-JSON fallback — so a 2xx that silently drops
the trigger is reported as unsupported (the "got 0 items" class, #5478,
#5648). Probe identity is now UUID-shaped and applies the account's
convergence, matching real traffic on the same endpoint.
2026-08-15 16:35:08 +08:00
shaw 8219dcfc87 fix(openai): relay x-codex-turn-state and guard cross-account echo
Codex captures x-codex-turn-state from /responses SSE, /responses/compact
JSON and the WS handshake (codex-api sse/responses.rs, endpoint/compact.rs),
then echoes it back on later requests of the same turn. The HTTP path dropped
it because the header is not in the generic response allowlist, while the WS
path already relayed it — an inconsistency that broke the protocol chain.

Relay it explicitly at every commit point instead of widening the global
allowlist (which would leak it into Anthropic/Gemini responses):

- streaming, non-streaming and SSE-to-JSON handlers relay it, clearing any
  value left over by a previous failover attempt when upstream sends none
- under the first-output guard the header is only staged; provenance is
  recorded when applyAttemptResponseHeaders actually writes it, because a
  first-output timeout discards the staged headers and the client never
  receives that blob
- record (api key + client session) -> minting account, TTL-bounded with an
  opportunistic sweep, and strip echoes known to come from another account
  before they go upstream. Stripping only: injection is the Claude bridge's
  job. Same-account or unknown provenance passes through unchanged.
2026-08-15 16:34:47 +08:00
Wesley Liddick 1d3b9665c8 Merge pull request #5641 from InCerryGit/fix/issue-5624-remote-compaction-v2
fix(openai): preserve remote compaction v2 responses endpoint
2026-08-15 13:46:31 +08:00
lyen1688 45dcce0e49 修复:隔离分组用量仓储测试时区
分组用量仓储测试原先固定恢复上海时区,覆盖集成测试包初始化的 UTC,并导致后续今日统计多算。统一保存并恢复测试进入前的时区,避免 unit 和 integration 用例相互污染。
2026-08-14 23:48:03 +08:00
lyen1688 89d826be29 修复:解决分组用量 PR 的 CI 失败
恢复分组时区测试进入前的全局时区,避免污染高峰计费和支付租约测试。同步升级 Go 1.26.6 与 nanoid 3.3.18,清除后端和前端安全门禁。
2026-08-14 23:25:31 +08:00
lyen1688 cb7b03795d feat: 优化分组用量统计 2026-08-14 22:34:43 +08:00
InCerryGit a8b9ea22b7 fix(openai): separate native and legacy compaction routing 2026-08-14 18:18:53 +08:00
InCerryGit 9662cff2e7 fix(openai): preserve remote compaction v2 responses endpoint 2026-08-14 16:23:52 +08:00
wucm667 674570ca17 fix: preserve group pricing in auth snapshots 2026-08-14 00:25:55 +08:00
wucm667 a288bab73a fix(gateway): align passthrough model discovery 2026-08-13 12:40:06 +08:00
Wesley Liddick fbfdcef818 Merge pull request #5573 from IanShaw027/fix/grok-long-context-and-media-fallback
fix(grok): 长上下文阶梯不再被 OpenAI 账号开关否决,媒体 ID 不继承文本价
2026-08-13 10:32:07 +08:00
Wesley Liddick e803e3851c Merge pull request #5559 from seng1e/fix/scheduled-backup-leader-lock
fix(backup): 定时备份加 leader 锁,避免多实例重复备份
2026-08-13 09:24:11 +08:00
Wesley Liddick 5912ae960c Merge pull request #5543 from luckydududu/fix/invalidate-channel-cache-on-group-platform-change
fix(group): 改分组 platform 后失效渠道缓存(否则最长 10 分钟按旧平台计价)
2026-08-13 09:24:00 +08:00
Wesley Liddick e77b4b260e Merge pull request #5504 from fengshao1227/fix/responses-probe-inconclusive-keeps-unknown
fix(openai): 探测响应未跑完时不再落标为「上游不支持 Responses」
2026-08-13 09:23:50 +08:00
IanShaw027 e29b93a1fb fix: 未知 Grok 文本兜底排除 image/video/audio 等媒体族
带版本号的媒体 ID(grok-2-image-1212、grok-5-video)会被
grok-N* 白名单当成文本,错吃 4.5 token 价。vision 多模态
仍按 token 计;HasIdentifiedTokenPricing 也不放行这些 ID。
2026-08-13 09:22:38 +08:00
IanShaw027 fd82dfd52d fix: Grok 长上下文只跟分组开关,不受 OpenAI 账号开关否决
openai_long_context_billing_enabled 是 OpenAI 账号设置,Grok
账号无法打开。按账号硬传 false 会让分组开关失效,官方 200k
阶梯和渠道多档区间全部塌到第一档。非 OpenAI 不再传账号门闩。
2026-08-13 09:22:34 +08:00
Wesley Liddick 4853662f99 Merge pull request #5540 from luckydududu/fix/channel-pricing-model-name-normalization
fix(channel): 定价冲突检测与定价缓存键的归一化对齐(#4754 修复残留)
2026-08-13 09:12:41 +08:00
IanShaw027 b61e4bcc45 fix: 对齐 JWT 常量 gofmt,并补全 available groups 契约字段
golangci-lint 要求 const 块按类型对齐;分组 DTO 新增
long_context_pricing_enabled 后,契约夹具未同步导致比较失败。
2026-08-13 08:55:56 +08:00
IanShaw027 0ae151a238 fix: 徽章改读 grok_usage_snapshot,增量刷新比较 Grok 快照
后端写入 grok_usage_snapshot,列表却读无 writer 的 grok_quota_snapshot;
增量刷新不比 Grok 快照,档位更新后行不替换。

- 正式快照优先 usage,quota 仅兜底;计划字段只收非空字符串
- buildGrokUsageRefreshKey 接入自动刷新
- 注明网关 Extra 写入依赖每请求独立 Account 副本
2026-08-13 08:38:10 +08:00
IanShaw027 b830bc14d6 fix: 长上下文默认保持开启,并与 OpenAI 账号开关取交集
迁移默认 false 会让存量分组静默丢掉 ≥200k 阶梯与渠道多档价。
OpenAI 路径传入 Group 后,账号级 openai_long_context_billing_enabled 被顶掉。

- 列默认 TRUE,并对存量行回填
- 分组价卡不再手填 token 区间,勾选走官方/预设阶梯
- 分组开关与账号开关 AND;价卡 JSON 损坏时打 warn
2026-08-13 08:38:10 +08:00