github/sub2api
4
0
Fork 0
mirror of https://github.com/Wei-Shaw/sub2api.git synced 2026-10-30 16:23:31 +00:00
Code Issues Packages Projects Releases Wiki Activity
6,994 Commits 49 Branches 65 Tags
e74c690ac42f634797ca14a5847aa4874711b033
Commit Graph
5222 Commits
Author SHA1 Message Date
cyh e74c690ac4 fix(openai): keep transport heartbeats outside semantic output 2026-09-17 19:15:09 +08:00
Wesley Liddick efe9aab1e4 Merge pull request #7246 from laoxiechuzheng/fix/deepseek-native-responses-tool-output-media 
fix(deepseek): lift Responses tool output images
 2026-09-17 17:02:43 +08:00
Wesley Liddick aeab57b9b2 Merge pull request #7256 from zhudaguaneren/fix/antigravity-attribution-429 
fix(antigravity): 过滤 Claude attribution 元数据，修复特定请求的 429
 2026-09-17 16:44:09 +08:00
Wesley Liddick 18328aed05 Merge pull request #7212 from wucm667/feat/issue-7203-redemption-pagination 
feat(redeem): paginate user redemption history
 2026-09-17 16:43:06 +08:00
Wesley Liddick 48a86ec926 Merge pull request #7224 from HypoxanthineOvO/codex/upstream-chat-role-compat 
fix(gateway): normalize developer roles for strict native Chat upstreams
 2026-09-17 16:42:31 +08:00
Wesley Liddick b4e560824a Merge pull request #7177 from xO-ox-ai/codex/fix-openai-response-binding-context 
fix(openai): 客户端断开后仍持久化 Responses 续聊绑定
 2026-09-17 16:42:12 +08:00
Wesley Liddick 9efc167482 Merge pull request #7195 from bitnom/fix/refresh-paused-oauth-accounts 
fix(account): keep refreshing OAuth tokens for paused accounts
 2026-09-17 16:41:54 +08:00
Wesley Liddick c84a3d50a3 Merge pull request #7215 from xz-dev/fix/manifest-double-parse 
fix(openai): parse Codex models manifest envelope once
 2026-09-17 16:41:12 +08:00
Wesley Liddick d52ab27def Merge pull request #7173 from HuntercodeT/fix/gemini-native-mixed-model-listing 
fix(gemini): discover Antigravity mappings in native model lists
 2026-09-17 16:37:38 +08:00
Wesley Liddick 7e4164c1cf Merge pull request #7207 from HuntercodeT/fix/group-usage-summary-seq-scan 
fix: 分组用量汇总不再全表扫 usage_logs（48.7s → 34ms）
 2026-09-17 16:37:21 +08:00
laoxiechuzheng acc05620cb fix(deepseek): keep parallel Responses tool outputs contiguous 2026-09-17 15:45:56 +08:00
HuntercodeT 611c30f04f chore(deps): 升级 google.golang.org/grpc 到 v1.83.2 修复 govulncheck 报告的两条公告 
Security Scan / backend-security 自 2026-09-16 起在所有 PR 上失败，govulncheck 对
google.golang.org/grpc v1.82.1 报：
- GO-2026-6443 Server panic via missing authority or Host headers（修复于 v1.83.2）
- GO-2026-6348 Heap Memory Exhaustion via HTTP/2 DATA Frame Fragmentation（修复于 v1.83.1）
调用链在 pkg/pluginapi/v1/runtime.go 与 internal/service/plugin_runtime.go。

go get google.golang.org/grpc@v1.83.2 && go mod tidy。最小版本选择连带抬升了
golang.org/x/{crypto,mod,net,sync,term,sys,text,tools}、genproto/googleapis/rpc 与
go.opentelemetry.io/otel* 的间接依赖，均为补丁级别。
 2026-09-17 14:33:48 +08:00
zhudaguaneren be4a4990ff fix(antigravity): strip Claude attribution metadata from system prompts 2026-09-17 11:57:53 +08:00
laoxiechuzheng 881ab1b0c6 fix(deepseek): lift Responses tool output images 
DeepSeek native Responses rejects Codex view_image results when function_call_output.output contains input_image parts. Move media into a following user message and keep the tool output as a string so call_id pairing remains valid.
 2026-09-17 01:22:22 +08:00
HypoxanthineOvO 18bfa4bf20 fix(gateway): normalize developer roles for strict Chat upstreams 2026-09-16 16:11:43 +08:00
Xiangzhe 31f3003ff5 fix(openai): preserve manifest key validation 2026-09-16 06:15:31 +00:00
Xiangzhe 2f16e0984e fix(openai): avoid duplicate manifest parsing 2026-09-16 06:15:25 +00:00
wucm667 a9ed218989 feat(redeem): paginate user redemption history 2026-09-16 13:19:15 +08:00
HuntercodeT 18d483c2a1 fix: 分组用量汇总不再全表扫 usage_logs 
GET /api/v1/admin/groups/usage-summary 在有一定数据量的库上会退化成
usage_logs 全表扫。根因是 getAllGroupUsageSummaryFromRollups 里，tail CTE
的 created_at 下界 state.tail_start 由同一条 SQL 的 state CTE 算出、再用
CROSS JOIN 交给 tail：这个值对 planner 来说要到运行期才知道，既进不了
index cond，也估不准选择性，于是只能顺序扫全表再逐行过滤。

生产实测（1721 万行 usage_logs / PostgreSQL 17）：

    Nested Loop  (actual time=28161..48334 rows=34788)
      Join Filter: (ul.created_at >= state_1.tail_start)
      Rows Removed by Join Filter: 17178082
      ->  Seq Scan on usage_logs  (actual time=0.030..47567 rows=17212870)
          Buffers: shared hit=12863 read=951875
    Execution Time: 48720 ms

改成先读一次水位行、在 Go 侧算出 tail_start / closed_before / retained
日期，再把它们当查询参数传进来。同一套数据上：

    Index Scan using idx_usage_logs_created_at on usage_logs
      Index Cond: (created_at >= $7)
    Execution Time: 34 ms

水位有效性的判定（恰好一行、时区名与配置一致、水位不在未来）和无效时
退化成从 epoch 全量重算的行为，与原先 SQL 里 state_values / CASE WHEN
valid 的分支逐条对应，结果口径不变。

另外读取水位的语句不加锁，不参与 #6976 那条 FOR UPDATE / FOR KEY SHARE
的阻塞链。
 2026-09-16 10:11:18 +08:00
bitnom 8e34ca5e31 fix(account): keep refreshing OAuth tokens for paused accounts 
Paused (schedulable=false) active accounts were excluded from the
background token refresh candidates, so their stored access_token
silently expired. The admin usage-window quota probe uses the stored
token verbatim (no token provider / no refresh), so the probe then gets
HTTP 401 UNAUTHENTICATED and the dashboard reports a false "needs
re-auth" for a perfectly healthy account, while Test Connection and live
traffic (which refresh on demand via the token provider) keep working.

Pause is not permanent rejection: error accounts are already excluded by
the status = 'active' filter, and accounts whose refresh genuinely fails
are rate-limited by the retry-cooldown exclusion (10-minute retry-
exhausted cooldown). Refresh success only persists credentials and never
touches schedulable, so a paused account cannot be silently un-paused.

Drop the schedulable filter and update the SQL-filter test, the
integration pagination test, and the service-level candidate fake.
 2026-09-15 10:00:26 -04:00
xO-ox-ai d7ee1ab6be fix(openai): persist response affinity after client cancel 2026-09-15 18:44:50 +08:00
github-actions[bot] 881f320269 chore: sync VERSION to 0.2.5 [skip ci] 2026-09-15 09:02:51 +00:00
HuntercodeT da74bf13a4 fix(gemini): discover antigravity mappings in native model lists 2026-09-15 16:50:46 +08:00
Wesley Liddick 86f93c28ee Merge pull request #6977 from alfadb/fix/deepseek-model-name-validation 
fix(deepseek): 未配置 model_mapping 时按官方模型名单校验，不再透传未知模型名
 2026-09-15 16:20:31 +08:00
Wesley Liddick 30ed40a56a Merge pull request #7074 from Brisbanehuang/codex/fix-codex-quota-windows 
fix(scheduler): 按规范 5h/7d 窗口读取 Codex 配额与重置时间
 2026-09-15 16:17:37 +08:00
Wesley Liddick b8275209e8 Merge pull request #7162 from HuntercodeT/fix/antigravity-gemini-sse-separator 
fix(antigravity): gemini 透传 SSE 事件之间不再多写一个空行（Antigravity CLI invalid stream chunk）
 2026-09-15 16:17:25 +08:00
Wesley Liddick 25c21aedda Merge pull request #7129 from kk-418/fix/grok-build-responses-sequence-number 
fix(responses): 始终在 Responses SSE 中写出 sequence_number 避免反序列化失败
 2026-09-15 15:28:33 +08:00
HuntercodeT 8447bdd368 fix(antigravity): gemini 透传 SSE 事件之间不再多写一个空行 
handleGeminiStreamingResponse 把上游每条 data 行改写成 "data: ...\n\n"，
紧接着又把上游的空分隔行原样透传成第三个 "\n"，下游看到的事件之间是
"\n\n\n"。google-genai 的 Go SDK（Antigravity CLI 在用）按 "\n\n" 切事件，
多出的 "\n" 粘到下一个事件开头，前缀变成 "\ndata" 而报
"iterateResponseStream: invalid stream chunk"，每条流都在第二个事件中断。

空分隔行不再透传（data 分支已经写过 "\n\n"），附回归测试：两个事件
（LF 与 CRLF 各一）经过后输出必须恰好是一个空行分隔，并按 "\n\n" 切帧
逐个校验前缀。

Fixes #7161
 2026-09-15 15:10:45 +08:00
alfadb a939553c87 fix(deepseek): 模型名校验不再默认兜底，未配置映射时按官方名单校验 
- 未配置 model_mapping 的 DeepSeek 账号不再放行任意模型名，改为官方名单精确匹配
  （deepseek-flash / deepseek-v4-pro / deepseek-v4-flash /
   deepseek-v4-flash-vision-exp / deepseek-v4-pro-0813）
- 校验前先归一化：剥离 Claude Code 客户端上下文后缀 [1m]（含重复形式）并统一小写，
  使官方 Claude Code 接入写法 deepseek-flash[1m] 继续可用
- 未知模型名在调度阶段即拒绝，返回 404 model_not_found，避免透传上游后触发
  per-(账号,模型) 冷却、并避免上游把不认识的模型名静默兜底成 deepseek-flash
- 出站前剥离泄漏的 [1m] 后缀，仅作用于 DeepSeek 平台
- 配置了 model_mapping 的账号语义不变（映射优先）
 2026-09-15 10:42:29 +08:00
Wesley Liddick 2061508ad8 Merge pull request #7072 from Brisbanehuang/codex/fix-scheduler-metrics 
fix(scheduler): 修复选号耗时返回值和粘性命中率重复计数
 2026-09-15 08:58:58 +08:00
Wesley Liddick fb90414c05 Merge pull request #7126 from xO-ox-ai/codex/fix-responses-lite-namespace 
fix(openai): 保留 Responses Lite namespace 工具调用
 2026-09-15 08:58:25 +08:00
Wesley Liddick 9fade3e8c2 Merge pull request #6769 from alfadb/feature/ollama-rate-limit-reset 
feat(ollama): 根据用量窗口异步更新限流恢复时间
 2026-09-15 08:57:43 +08:00
Wesley Liddick 1920a7d687 Merge pull request #7091 from SavitarC/feat/subscription-bulk-actions 
feat: 支持订阅管理批量操作
 2026-09-15 08:57:27 +08:00
Wesley Liddick 150c52b8de Merge pull request #6689 from tunsuy/fix/antigravity-codex-web-search-mixed-tools 
fix(antigravity): drop mixed web_search when client tools present
 2026-09-15 08:54:40 +08:00
Wesley Liddick 3e4b6a09b2 Merge pull request #6925 from reed-yang/fix/recover-responses-text 
fix(apicompat): recover Responses text from done and terminal events
 2026-09-15 08:54:21 +08:00
Wesley Liddick dd2e6b3680 Merge pull request #7076 from Shen-An/fix/antigravity-refresh-warning 
fix(admin): surface partial Antigravity refresh warnings
 2026-09-15 08:54:06 +08:00
Wesley Liddick d4439129a7 Merge pull request #7094 from liuyuzhou666/fix/responses-chat-developer-role-mapping 
fix(apicompat): 修复 Responses 降级 Chat Completions 时的 system 消息顺序问题
 2026-09-15 08:53:50 +08:00
Wesley Liddick 1e1d15cca5 Merge pull request #7082 from ggllzz04/fix/antigravity-token-cache-isolation 
fix(antigravity): isolate OAuth token cache key by account ID to prevent 429 collision
 2026-09-15 08:53:32 +08:00
Wesley Liddick f7e959ae5f Merge pull request #7111 from ouchihao/codex/bug-proxy-clear-credentials 
fix(proxy): allow explicitly clearing stored credentials
 2026-09-15 08:52:58 +08:00
Wesley Liddick 008391f699 Merge pull request #7085 from r266-tech/codex/sub2api-f352188cb784 
fix(openai): strip internal metadata from OAuth input items
 2026-09-15 08:52:30 +08:00
xO-ox-ai 13c1d4cde2 test(openai): 覆盖 Lite 命名空间工具回传 2026-09-14 11:46:41 +08:00
cikenerdandCursor 087987070b fix(responses): 为 grok-build 始终写出 sequence_number 
网关合成/重编的 Responses SSE 因 omitempty 或缺字段丢掉 sequence_number，
grok-build 会整轮反序列化失败。合成失败帧、compact 桥接和转换 wire JSON
一律写出该字段，0 也保留。

Fixes #7128

Co-authored-by: Cursor <cursoragent@cursor.com>
 2026-09-14 11:45:53 +08:00
xO-ox-ai 4bd7132ba9 fix(openai): 保留 Responses Lite namespace 调用 2026-09-14 11:35:03 +08:00
chihao.ou 28a4ea914d fix(proxy): allow explicitly clearing stored credentials 2026-09-13 17:23:24 +08:00
liuyuzhou666 52558381c6 fix(apicompat): merge leading system messages and downgrade mid-conversation system from Responses bridge 
The Responses to Chat Completions bridge mapped both the instructions field and role developer items to system messages. Codex always sends both, and also injects developer notices mid-conversation when switching models, so the forwarded request contained consecutive leading system messages and mid-conversation system messages. Strict Qwen-family upstreams reject this with HTTP 400 "System message must be at the beginning."

Leading system/developer messages are now merged into a single leading system message, and later system/developer messages are downgraded to user so the content is preserved without violating upstream role ordering.
 2026-09-13 14:08:57 +08:00
SavitarC e3cce574d7 fix: serialize subscription extensions 2026-09-13 12:33:35 +08:00
SavitarC 3d6c207723 feat: 支持订阅管理批量操作 2026-09-13 11:49:03 +08:00
alfadb a1d5968b24 feat(ollama): async rate-limit reset for Ollama Cloud usage windows 
Real Ollama Cloud 429s are driven by ollama.com usage windows, not by
response headers that OpenAI/Anthropic/CN rate-limit parsing understands.
Handle them separately: apply a never-shortening temporary cooldown first,
then schedule an async probe that learns the true usage-window reset point
with a controlled, coalesced per-api_key refresh.

- ratelimit_service: route real Ollama Cloud usage accounts to a dedicated
  429 path before the CN-provider branch.
- ollama_cloud_usage: add a background probe coordinator (started in Start,
  waited by Stop) that fans out model-429 probes to one controlled refresh
  per api_key group, reusing the recent group snapshot for coalesced members.
- repository/account_repo: add SetRateLimitedIfUnchanged, a write-back CAS
  counterpart to ClearRateLimitIfObserved, so an async reset never overwrites
  a newer 429, an admin clear, a re-armed generation, or a key/state change
  observed by another writer between read and write.
- wire: inject OllamaCloudUsageService into RateLimitService.

Tests: unit tests for service and repository pass; targeted race tests pass;
build and vet clean; git diff --check clean. The CAS integration test compiles
but is not run (no Docker available).
 2026-09-13 10:41:47 +08:00
r266-tech e5272c130c test(openai): check input type before metadata assertion 2026-09-13 07:13:31 +08:00
r266-tech 196c15b5cd fix(openai): strip internal metadata from OAuth input items 2026-09-13 07:00:50 +08:00