COS … API GW HTTP Timer 调用 函数代码 依赖 本地 开发者 Serverless 介绍 Serverless 业务运维能力 业务运维能力 DevOps 开箱即用 工具建设 工具 ❑ VS Code IDE 插件 ❑ WebIDE ❑ Command Line 功能 ❑ 函数增删改查 ❑ 调试 ❑ 打包部署 ❑ 日志 ❑ … DevOps 解决方案 Network Runtime Serverless vs. IaaS 运维能力 安全保障 Serverless vs. IaaS 运维能力 ❑ 网络隔离 ❑ 执行环境与管理环境隔离 ❑ 函数资源限制 ❑ 文件系统目录限制 ❑ 系统调用限制 ❑ … 腾讯相册微信小程序运维 开发流程 团队 Team 后台开发 * 2 测试 * 1 前端工程师 * 1 运维 * 2 ？weeks

这一步出错会导致和本地文件系统不一致的行为： Linux 接口定义允许 rename 过程中某一段时间存在这样的硬链接（或者 rename 执行到一半断电也会存在） 而 POSIX 接口中提到了该函数得是原子（不断电的情况下，rename 操作不能被其他操作打断，不存在中间状态） 参考： Is rename() atomic? rename(2) - Linux man page However

engineering teams across the world against four key measures: lead time for new features, failure rate of code changes, frequency of deployment and the time it takes to restore service following an outage The Facts 4 4. Change failure rate This can be seen as a measure of quality in the release process. The best performers in the DORA study reported a change failure rate of less than 15%, while the low hour Less than one day Less than one day Between one week and one month Change failure rate 0-15% 0-15% 0-15% 46-60% A surprise: speed and stability support one another Combining the first

and GitOps together Deployment/Cycle Time Deployment Frequency Change Lead Time Change Failure Rate MTTR vs. MTTF (mean time to recover/fail) P.3 Simplify Kubernetes operations using developer

all manually configured (“pets vs. cattle”) Production deployments have a very low success rate I have no visibility or record of configuration changes in environments I can’t easily