[Cherry-Pick][RL] Allow configuring update weights timeout (#8073)

[jackyYang6]

Motivation

Support configuring the timeout for update_weights requests. This helps large weight updates avoid timing out with the fixed default timeout.

Modifications

• Add timeout parameter validation in the OpenAI API update_weights endpoint.
• Use the configured timeout when updating worker weights and cache-transfer metadata.
• Keep timeout as a control-only parameter and do not forward it to workers.
• Add unit tests for default timeout, custom timeout, and invalid timeout values.

Usage or Command

Example request body:

{
  "timeout": 120
}

If timeout is not provided, the default value is 180 seconds.

Accuracy Tests

Not applicable. This PR only changes control request timeout handling and does not affect model outputs.

Checklist

• Add at least a tag in the PR title.
  • Suggested title: [Engine][APIServer] Allow configuring update weights timeout
• Format your code, run pre-commit before commit.
• Add unit tests.
• Provide accuracy results.
• If the current PR is submitting to the release branch, make sure the PR has been submitted to the develop branch, then cherry-pick it to the release branch with the [Cherry-Pick] PR tag.

[PaddlePaddle-bot]

🤖 Paddle-CI-Agent | pr_review | 2026-06-24 13:37:18

📋 Review 摘要

PR 概述：为 OpenAI update_weights 增加 timeout 参数校验，并在 Engine/Cache 更新权重时使用该超时。
变更范围：fastdeploy/entrypoints/openai/api_server.py、fastdeploy/engine/common_engine.py 及相关单测
影响面 Tag：APIServer Engine

问题

级别	文件	概述
🔴 Bug	fastdeploy/entrypoints/openai/api_server.py:486	timeout 大于 600 秒时仍会被 EngineClient.run_control_method 固定 600 秒提前截断，配置不会真正生效

📝 PR 规范检查

PR 描述结构符合模板；标题包含 [Cherry-Pick] 和官方 Tag，本轮未基于上下文确认原 develop PR 号。

总体评价

核心路径已正确避免把 timeout 透传给 worker/cache-transfer，但 API 入口到 engine 的等待超时仍未联动，导致大于 600 秒的配置场景会返回超时错误，需要修复后再合入。

[PaddlePaddle-bot]

🔴 Bug 这里允许用户把 timeout 配成任意正整数，但当前请求链路里 EngineClient.run_control_method() 仍固定 asyncio.wait_for(..., timeout=600)。

当用户为大权重更新配置 timeout > 600 时，engine 内部会继续按该值等待 worker/cache-transfer，但 OpenAI API 侧会在 600 秒先返回 Timeout waiting for control method response，配置的超时没有真正生效，甚至可能出现 HTTP 已失败而后台仍在更新权重的状态。

建议修复方式：
在发送 ControlRequest 后，让 EngineClient.run_control_method 对 update_weights 使用 request.args["timeout"] 派生的等待时间（建议加少量 buffer，例如 timeout + 10，或 max(600, timeout + 10)），并补一个覆盖 timeout > 600 的 engine_client/API 单测。

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (release/2.6@8be4aab). Learn more about missing BASE report.

Additional details and impacted files

@@              Coverage Diff               @@
##             release/2.6    #8074   +/-   ##
==============================================
  Coverage               ?   71.60%           
==============================================
  Files                  ?      386           
  Lines                  ?    55773           
  Branches               ?     8762           
==============================================
  Hits                   ?    39937           
  Misses                 ?    13021           
  Partials               ?     2815           

Flag	Coverage Δ
GPU	71.60% <100.00%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[mitu626]

lgtm