组合使用deepseek模型时，failover报错upstream_server_error

[hnyyghk]

Client or integration

Codex App

Area

Proxy and routing

Summary

组合使用deepseek模型时，failover报错upstream_server_error

Reproduction

第一个模型报错rate_limit_exceeded时，failover切换到第二个deepseek官方api模型，502报错：
upstream_server_error
Upstream stream ended unexpectedly without a terminal event (adapter_eof)

后面codex客户端的自动重试也一直503报错
server_is_overloaded
No available targets for combo: deepseek-flash-combo

不知道这个信息有没有帮助，我是前置先用的qwen3.8-max，额度用尽了之后429报错，再切换的自己配置的deepseek组合模型
rate_limit_exceeded
Provider error 429: Your token-plan 1-week quota has been exhausted. The quota will reset at MM-dd HH:mm:ss UTC.

Version

2.60.0

Operating system

macOS 13.5.1

Provider and model

No response

Logs or error output

Screenshots and supporting files

No response

Redacted configuration

Checks

• I searched existing issues and documentation.
• I removed secrets, tokens, account details, request credentials, and personal data.

[lidge-jun]

The part of this that OpenCodex controls is fixed on dev in #6062 (merge bf6c57c0d7, commit 9fdc0c4582). A 429 saying the token-plan quota "has been exhausted" gave the combo target a 60-second cooldown. The dead target was offered again every minute, and together with one failure on the DeepSeek target, every request in between got 503 No available targets. That target now waits the ten-minute exhaustion hold. Why the official DeepSeek stream ended without a terminal event (adapter_eof) cannot be seen from the repository. If it still happens on the next release, please open a new issue with the proxy log lines for that request (redacted).