Skip to content

fix: open bugs 429, 435, 436, 439 and cut v1.3.171 - #448

Merged
dofastted merged 5 commits into
mainfrom
fix/open-bug-issues
Oct 15, 2026
Merged

dofastted merged 5 commits into
mainfrom
fix/open-bug-issues

Conversation

@dofastted

@dofastted dofastted commented Oct 15, 2026 •

Copy link
Copy Markdown
Owner

Summary

Test plan

  • account-quota、codex-slot-pool、handle-codex、stream-keepalive、routing-file、kernel-router、routing-runtime-load、codex-kernel-client 相关单测
  • 忽略 SIGTERM 的子进程不会被当成已退出,socket 保留;正常退出后才删除 socket
  • CI

Note

Medium Risk
Changes affect quota gating, streaming commit/failover timing, Codex slot switching, and ingress/kernel body limits with idle kernel restarts—operational but not auth/data-migration critical.

Overview
v1.3.171 tightens quota accounting, streaming reliability, Codex failover, and configurable request body limits.

Quota (#439): ingestHeaders no longer treats every HTTP 429 as exhausted. Local limits (e.g. fable_requires_max with exhausted: false and no rejected quota headers) leave the 5h window allowed; real upstream rejection still uses rejected headers or exhausted: true.

Streaming (#429): Downstream keepalive is armed before the first upstream byte. After the interval, the gateway can send 200 + SSE ping, which commits the attempt so a later upstream failure cannot switch slots on that socket. Failures before the ping can still failover.

Codex (#435): Uncommitted upstream 503 with server_error is treated as transient failover; committed hops or responses already written to the client are not retried on another slot.

Request body limit (#436): Settings → pool → failover adds 「请求体上限」 (failover.max_request_mib, 8–128 MiB, default 32). inboundBodyLimit caps gateway ingress (≤ KIN_MAX_BODY); maxRequestBytes flows into Claude/Codex kernel configs. Saving routing patches syncs kernel config and reloads Claude kernels when idle; Codex kernels recycle only when the slot has ≤1 in-flight request and the old process has exited (socket removed after exit).

Docs/API and the pool settings UI reflect the new limit behavior.

Reviewed by Cursor Bugbot for commit dcb2b4b. Configure here.

A gateway 429 such as fable_requires_max carries no upstream quota
headers. ingestHeaders was marking the 5h window rejected from the
status alone.

#439
Already-committed hops and responses that have been written downstream
still stop on that slot.

#435
A stream now sends a ping before the first token and commits that
attempt, so a reverse proxy does not time out at 120s. The panel
setting failover.max_request_mib caps both the gateway and the slot
kernels. A Codex kernel recycles only when its slot is otherwise idle,
after the old process has exited.

#429
#436
A recycle deadline is not an exit. SIGTERM, then SIGKILL, and drop the
socket only after the process is gone. If it is still alive, keep the
tracked child and fail the recycle instead of spawning another kernel.
@cursor

cursor Bot commented Oct 15, 2026

Copy link
Copy Markdown

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: 318189e7-4517-421c-a58e-c85fa4f781bd)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant