我们在一个拉取请求中发现了每月1200美元的云成本回归问题。
我们最近在一次拉取请求审查中发现了云成本回归(约 $1,200/月)。<p>并没有什么特别的情况:
- 添加了一个 NAT 网关
- RDS 副本数量增加
- 没有触发警报
- 没有更改仪表板<p>这个拉取请求本可以像其他基础设施变更一样合并。<p>我们开发了一个小工具,可以在拉取请求中直接进行成本检查:
- 显示合并前后的成本估算
- 标记高风险模式(无限循环、频繁调用的 API、过度配置的基础设施)
- 可选地根据政策阈值阻止合并<p>我们整理了一个公开的演示仓库,您可以在其中打开拉取请求并查看成本反馈:<p>https://github.com/CloudVerse-Pte-Ltd/devx-cost-regression-demo<p>想了解其他人在代码审查时如何处理成本回归。
查看原文
We recently caught a cloud cost regression (~$1,200/month) during a pull request review.<p>Nothing exotic:
- A NAT gateway was added
- An RDS replica count increased
- No alarms fired
- No dashboards changed<p>The PR would have merged like any other infra change.<p>We built a small tool that runs cost checks directly in PRs:
- Shows a before → after cost estimate
- Flags high-risk patterns (unbounded loops, chatty APIs, over-provisioned infra)
- Optionally blocks merges on policy thresholds<p>We put together a public demo repo where you can open a PR and see the cost feedback:<p>https://github.com/CloudVerse-Pte-Ltd/devx-cost-regression-demo<p>Curious how others handle cost regressions at code-review time.