feat(gateway): Codex/OpenAI Responses WebSocket 代理模式

在 /v1/responses 上支持 WebSocket 升级,把客户端帧中继到上游 Codex /
OpenAI Responses WebSocket 端点,同时保持既有的路由、鉴权、配额与用量
语义:

- 路由与准入:control/route/ai.rs 识别 WebSocket 升级请求;
  websocket/ingress.rs 复用 API Key 鉴权、IP 规则与并发许可,并引入
  独立的 WebSocket 连接许可
- 中继:websocket/responses/* 按 connection / session / turn 分层,
  帧解析归一化、socket 写入有界、continuation 保持调度亲和性
- 配额:orchestration/codex_quota_breaker.rs 在账号配额耗尽时熔断并
  自动恢复,不再直接断开客户端连接
- 用量:每个 turn 的终态用量落库,request_metadata 记录
  websocket_mode / websocket_transport,管理端与 usage 视图暴露
  is_websocket
- 管理端:provider 可配置 Responses WebSocket 开关
This commit is contained in:
AAEE86
2026-08-17 14:50:33 +08:00
committed by ZheFox
parent 9a0d346ff3
commit 71b54070e8
72 changed files with 10441 additions and 108 deletions
@@ -68,12 +68,14 @@ impl ProviderPoolInFlightGuard {
if self.released {
return;
}
self.released = true;
match &mut self.kind {
ProviderPoolInFlightGuardKind::Local {
provider_id,
counter,
} => decrement_local_provider_in_flight(provider_id, counter),
} => {
self.released = true;
decrement_local_provider_in_flight(provider_id, counter);
}
ProviderPoolInFlightGuardKind::Runtime {
runtime,
tokens_key,
@@ -85,11 +87,17 @@ impl ProviderPoolInFlightGuard {
if let Some(handle) = renew_handle.take() {
handle.abort();
}
if let Err(err) = runtime.score_remove(tokens_key, token).await {
debug!(
error = ?err,
"gateway provider pool demand: failed to release in-flight token"
);
// Mark the guard released only after Redis confirms removal.
// If this future is cancelled, Drop still schedules the same
// idempotent cleanup instead of leaving the token until TTL.
match runtime.score_remove(tokens_key, token).await {
Ok(_) => self.released = true,
Err(err) => {
debug!(
error = ?err,
"gateway provider pool demand: failed to release in-flight token; scheduling drop fallback"
);
}
}
}
}