云原生大规模应用落地指南 (2020阿里巴巴双11云原生技术实践) (李响 等)（Z-Library）

【One-Line Pitch】 A practitioner's field report on how Alibaba rebuilt its Double 11 core systems into a fully cloud-native engine in 2020, useful for engineers and architects who want concrete patterns for running Kubernetes, Serverless, and middleware at extreme scale. 【Book Arc】 - **Opening (~0%–10%)**: Sets the strategic frame — why cloud-native is the shortest path to digital transformation, and how Alibaba moved from "Double 11 on the cloud" to a "cloud-native Double 11," with headline results across elasticity, compute, middleware, and Serverless. - **Early (~10%–35%)**: Drills into the technical foundation: Kubernetes/containers as the new cloud interface, the Sigma-to-ACK migration, Serverless landing challenges (cold start, dev workflow, middleware connectivity), RocketMQ's zero-fault messaging, and Sentinel Go for flow control and degradation. - **Middle (~35%–55%)**: Moves to operating large clusters — multi-cluster scale, change velocity, risk control, OpenKruise as the deployment substrate, and the ASI scheduler's mixed scheduling of online, batch, and best-effort workloads. - **Late (~55% onward)**: The excerpts do not cover the later chapters in detail; based on the table of contents and author list, the book continues into further cloud-native product and practice topics, but the provided material stops around the scheduling discussion. 【Key Takeaways】 - **Cloud-native is framed as an evolution, not a rewrite** (Opening): the book positions the journey as moving from Cloud Hosting to Cloud Native, with Alibaba's own path from 2008 distributed middleware through 2011 containerization to 2020 full core-system cloud-nativization. - **Kubernetes is the new cloud interface** (Early): containers standardize application delivery and decouple apps from infrastructure, while Kubernetes standardizes scheduling and hides underlying differences — this is the conceptual anchor for everything that follows. - **Scale demands specific engineering responses** (Middle): the book details concrete techniques for large clusters — using application names as namespaces, optimizing ETCD storage and read paths, and protecting the APIServer with inflight request limits. - **Serverless adoption hinges on solving cold start and workflow gaps** (Early): the book candidly lists three pain points — cold start stretching to seconds in core chains, disconnect from testing/gray-release/disaster-recovery workflows, and middleware connection overhead — then describes reserved-plus-on-demand modes and tooling as the response. - **Middleware is treated as a first-class cloud-native concern** (Early): RocketMQ's zero-fault Double 11 record and Sentinel's flow control, degradation, and adaptive protection illustrate how messaging and resilience are engineered for peak traffic. - **Deployment automation is the substrate for scale** (Middle): OpenKruise extends Kubernetes workloads to handle deployment, upgrade, scaling, QoS, health checks, and migration for millions of containers, and its "three-in-one" open-source/commercial/internal model is a recurring theme. - **Scheduling is a multi-layer problem, not just a central scheduler** (Middle): the book describes a "broad scheduler" combining central, single-machine, kernel, rescheduling, and orchestration layers, with differentiated SLO tiers (Product, Batch, Best Effort) to improve utilization. - **Stability is designed in, not bolted on** (Middle): default stability configurations, standardized probes, centralized self-healing, and resource isolation for sidecars are presented as practices that make stability "like water, electricity, and coal." 【Reading Tips】 - **Read the Opening for context, then jump to chapters matching your stack.** The strategic framing is useful once; the real value is in the Kubernetes, Serverless, middleware, and scheduling chapters. - **Treat the Serverless chapter as a decision aid.** The three pain points and the reserved/on-demand solution are the most transferable part for teams evaluating FaaS for latency-sensitive workloads. - **Skim the product-marketing passages.** Some sections read as Alibaba Cloud product promotion; extract the architectural patterns and skip the feature lists. - **Pay attention to the "three-in-one" theme.** The recurring idea of internal practice feeding open source and commercial products is a useful lens for evaluating any vendor's cloud-native claims. - **Use the scheduling chapter for utilization thinking.** The Product/Batch/Best Effort tiering is a concrete model for mixing workloads to raise cluster efficiency. 【Coverage Limits】 This guide is based on stratified excerpts covering roughly the first half of the book (through the scheduling discussion). Later chapters and specific implementation details beyond the provided material are not covered.