<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>winunify</title>
  
  <subtitle>winunify</subtitle>
  <link href="https://winunify.com/atom.xml" rel="self"/>
  
  <link href="https://winunify.com/"/>
  <updated>2026-03-20T14:15:39.207Z</updated>
  <id>https://winunify.com/</id>
  
  <author>
    <name>winunify</name>
    
  </author>
  
  <generator uri="https://hexo.io/">Hexo</generator>
  
  <entry>
    <title>构建基于NATS的事件驱动型CI分析管道以解耦单元测试与依赖扫描负载</title>
    <link href="https://winunify.com/7949450658/"/>
    <id>https://winunify.com/7949450658/</id>
    <published>2023-11-16T15:45:00.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;团队内部的 CI 流水线已经慢得像头搁浅的鲸鱼。一个典型的 Go 项目，一次代码提交触发的 &lt;code&gt;gitlab-ci.yml&lt;/code&gt;，串行执行单元测试、代码覆盖率计算、依赖漏洞扫描、静态代码分析，最后是构建和推送镜像。整个过程平均耗时 15 到 20</summary>
        
      
    
    
    
    <category term="DevOps" scheme="https://winunify.com/categories/DevOps/"/>
    
    
    <category term="CI/CD" scheme="https://winunify.com/tags/CI-CD/"/>
    
    <category term="Go" scheme="https://winunify.com/tags/Go/"/>
    
    <category term="NATS" scheme="https://winunify.com/tags/NATS/"/>
    
    <category term="单元测试" scheme="https://winunify.com/tags/%E5%8D%95%E5%85%83%E6%B5%8B%E8%AF%95/"/>
    
    <category term="依赖扫描" scheme="https://winunify.com/tags/%E4%BE%9D%E8%B5%96%E6%89%AB%E6%8F%8F/"/>
    
  </entry>
  
  <entry>
    <title>构建由 Axum 驱动的 Serverless MLOps 管道：集成 OpenFaaS、MLflow 与 SQL 实现 iOS 端模型按需个性化</title>
    <link href="https://winunify.com/1614650352/"/>
    <id>https://winunify.com/1614650352/</id>
    <published>2023-11-15T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.203Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;为移动端应用提供的机器学习模型，其生命力在于个性化。一个静态的、一刀切的模型在部署后很快就会因为用户行为数据的变化而变得迟钝。传统的中心化批量训练模式，周期长、成本高，无法满足对单一用户行为的实时响应。我们面临的挑战是：当特定用户在 iOS</summary>
        
      
    
    
    
    <category term="MLOps" scheme="https://winunify.com/categories/MLOps/"/>
    
    
    <category term="MLflow" scheme="https://winunify.com/tags/MLflow/"/>
    
    <category term="Axum" scheme="https://winunify.com/tags/Axum/"/>
    
    <category term="关系型 (SQL)" scheme="https://winunify.com/tags/%E5%85%B3%E7%B3%BB%E5%9E%8B-SQL/"/>
    
    <category term="iOS 开发" scheme="https://winunify.com/tags/iOS-%E5%BC%80%E5%8F%91/"/>
    
    <category term="OpenFaaS" scheme="https://winunify.com/tags/OpenFaaS/"/>
    
  </entry>
  
  <entry>
    <title>利用 Apache Spark 与 Git LFS 构建基于 SQLite 的原子化 CI 结果数据管道</title>
    <link href="https://winunify.com/2661149783/"/>
    <id>https://winunify.com/2661149783/</id>
    <published>2023-11-15T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.203Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;我们的 monorepo CI 系统正在变成一个性能黑洞。一个看似无害的 PR 能触发数百个独立的构建和测试任务，而定位其中引入的性能衰退，完全依赖工程师的人工排查和直觉。日志散落在各处，缺乏结构，更不用说进行趋势分析了。我们需要一个系统，能将每次 Git提交触发的 CI</summary>
        
      
    
    
    
    <category term="数据工程" scheme="https://winunify.com/categories/%E6%95%B0%E6%8D%AE%E5%B7%A5%E7%A8%8B/"/>
    
    
    <category term="Apache Spark" scheme="https://winunify.com/tags/Apache-Spark/"/>
    
    <category term="SQLite" scheme="https://winunify.com/tags/SQLite/"/>
    
    <category term="Git" scheme="https://winunify.com/tags/Git/"/>
    
    <category term="Turbopack" scheme="https://winunify.com/tags/Turbopack/"/>
    
    <category term="ACID" scheme="https://winunify.com/tags/ACID/"/>
    
  </entry>
  
  <entry>
    <title>构建基于Phoenix与WebRTC的大规模实时音视频数据管道以驱动AI分析</title>
    <link href="https://winunify.com/5400411376/"/>
    <id>https://winunify.com/5400411376/</id>
    <published>2023-10-27T11:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;我们面临的第一个问题不是信令，也不是媒体传输，而是数据出口。当上千路 WebRTC 音视频流在我们的 SFU (Selective Forwarding Unit)</summary>
        
      
    
    
    
    <category term="分布式系统" scheme="https://winunify.com/categories/%E5%88%86%E5%B8%83%E5%BC%8F%E7%B3%BB%E7%BB%9F/"/>
    
    
    <category term="Phoenix" scheme="https://winunify.com/tags/Phoenix/"/>
    
    <category term="WebRTC" scheme="https://winunify.com/tags/WebRTC/"/>
    
    <category term="AI" scheme="https://winunify.com/tags/AI/"/>
    
    <category term="数据科学" scheme="https://winunify.com/tags/%E6%95%B0%E6%8D%AE%E7%A7%91%E5%AD%A6/"/>
    
    <category term="大数据" scheme="https://winunify.com/tags/%E5%A4%A7%E6%95%B0%E6%8D%AE/"/>
    
  </entry>
  
  <entry>
    <title>构建支持 PyTorch 在线推理的低延迟实时特征存储架构</title>
    <link href="https://winunify.com/7152703675/"/>
    <id>https://winunify.com/7152703675/</id>
    <published>2023-10-27T11:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;当推荐系统或风控模型的在线推理（Online Inference）请求 QPS</summary>
        
      
    
    
    
    <category term="数据工程" scheme="https://winunify.com/categories/%E6%95%B0%E6%8D%AE%E5%B7%A5%E7%A8%8B/"/>
    
    
    <category term="Cassandra" scheme="https://winunify.com/tags/Cassandra/"/>
    
    <category term="Google Cloud Pub/Sub" scheme="https://winunify.com/tags/Google-Cloud-Pub-Sub/"/>
    
    <category term="Flask" scheme="https://winunify.com/tags/Flask/"/>
    
    <category term="SSR" scheme="https://winunify.com/tags/SSR/"/>
    
    <category term="PyTorch" scheme="https://winunify.com/tags/PyTorch/"/>
    
  </entry>
  
  <entry>
    <title>利用 Tekton 与 Redis 构建缓存感知的向量嵌入生成流水线</title>
    <link href="https://winunify.com/6080373292/"/>
    <id>https://winunify.com/6080373292/</id>
    <published>2023-10-27T11:15:39.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;团队的向量模型迭代一直是个痛点。每次对预处理逻辑或模型进行微调，都意味着需要对整个数GB的验证数据集重新生成嵌入向量，这个过程动辄数小时。我们的CI/CD流水线，原本是为了加速交付，现在却成了最主要的瓶颈。问题很明确：大量的计算是重复且不必要的。当输入数据和处理逻辑都没变时</summary>
        
      
    
    
    
    <category term="MLOps" scheme="https://winunify.com/categories/MLOps/"/>
    
    
    <category term="CI/CD" scheme="https://winunify.com/tags/CI-CD/"/>
    
    <category term="Tekton" scheme="https://winunify.com/tags/Tekton/"/>
    
    <category term="Redis" scheme="https://winunify.com/tags/Redis/"/>
    
    <category term="Python" scheme="https://winunify.com/tags/Python/"/>
    
    <category term="Vector" scheme="https://winunify.com/tags/Vector/"/>
    
    <category term="缓存" scheme="https://winunify.com/tags/%E7%BC%93%E5%AD%98/"/>
    
  </entry>
  
  <entry>
    <title>构建基于 Sidecar 模式的轻量级服务发现系统以支持 Go 与 Python 异构微服务</title>
    <link href="https://winunify.com/9523703823/"/>
    <id>https://winunify.com/9523703823/</id>
    <published>2023-10-27T11:15:39.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;h3 id=&quot;技术痛点：异构环境下的服务注册与发现难题&quot;&gt;&lt;a href=&quot;#技术痛点：异构环境下的服务注册与发现难题&quot; class=&quot;headerlink&quot;</summary>
        
      
    
    
    
    <category term="分布式架构" scheme="https://winunify.com/categories/%E5%88%86%E5%B8%83%E5%BC%8F%E6%9E%B6%E6%9E%84/"/>
    
    
    <category term="微服务" scheme="https://winunify.com/tags/%E5%BE%AE%E6%9C%8D%E5%8A%A1/"/>
    
    <category term="Python" scheme="https://winunify.com/tags/Python/"/>
    
    <category term="Go" scheme="https://winunify.com/tags/Go/"/>
    
    <category term="服务发现" scheme="https://winunify.com/tags/%E6%9C%8D%E5%8A%A1%E5%8F%91%E7%8E%B0/"/>
    
    <category term="API 与架构" scheme="https://winunify.com/tags/API-%E4%B8%8E%E6%9E%B6%E6%9E%84/"/>
    
    <category term="Sidecar" scheme="https://winunify.com/tags/Sidecar/"/>
    
    <category term="Gossip" scheme="https://winunify.com/tags/Gossip/"/>
    
  </entry>
  
  <entry>
    <title>基于 Apache Flink 与 Clean Architecture 构建流式驱动的静态站点生成管道</title>
    <link href="https://winunify.com/0323418056/"/>
    <id>https://winunify.com/0323418056/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.203Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;一个棘手的需求摆在面前：为高吞吐量的用户行为日志（每秒数万次页面浏览）构建一个近实时的监控仪表盘。传统方案，如使用ELK或直连时序数据库的前端轮询，因其高昂的实时查询成本和复杂的后端维护而被否决。我们的目标是极致的性能、低廉的成本和最小的运维负担。这意味着最终的用户界面必须</summary>
        
      
    
    
    
    <category term="数据工程" scheme="https://winunify.com/categories/%E6%95%B0%E6%8D%AE%E5%B7%A5%E7%A8%8B/"/>
    
    
    <category term="Apache Flink" scheme="https://winunify.com/tags/Apache-Flink/"/>
    
    <category term="Clean Architecture" scheme="https://winunify.com/tags/Clean-Architecture/"/>
    
    <category term="AWS" scheme="https://winunify.com/tags/AWS/"/>
    
    <category term="SSG" scheme="https://winunify.com/tags/SSG/"/>
    
  </entry>
  
  <entry>
    <title>构建金融风控实时特征平台中利用分布式锁保证 Spark 计算一致性的架构权衡</title>
    <link href="https://winunify.com/0415686491/"/>
    <id>https://winunify.com/0415686491/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.203Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;在金融风控场景下，特征计算平台的稳定性和数据一致性是整个系统的基石。我们面临的核心挑战是：如何在一个分布式的环境中，调度数百个 Apache Spark</summary>
        
      
    
    
    
    <category term="数据工程" scheme="https://winunify.com/categories/%E6%95%B0%E6%8D%AE%E5%B7%A5%E7%A8%8B/"/>
    
    
    <category term="Apache Spark" scheme="https://winunify.com/tags/Apache-Spark/"/>
    
    <category term="分布式锁" scheme="https://winunify.com/tags/%E5%88%86%E5%B8%83%E5%BC%8F%E9%94%81/"/>
    
    <category term="GraphQL" scheme="https://winunify.com/tags/GraphQL/"/>
    
    <category term="Apollo Client" scheme="https://winunify.com/tags/Apollo-Client/"/>
    
    <category term="金融风控" scheme="https://winunify.com/tags/%E9%87%91%E8%9E%8D%E9%A3%8E%E6%8E%A7/"/>
    
  </entry>
  
  <entry>
    <title>基于 Flink 与 Redux 构建高吞吐 IoT 平台的端到端状态一致性架构</title>
    <link href="https://winunify.com/1581180707/"/>
    <id>https://winunify.com/1581180707/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.203Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;h3 id=&quot;定义问题：实时风力发电机组监控的状态同步困境&quot;&gt;&lt;a href=&quot;#定义问题：实时风力发电机组监控的状态同步困境&quot; class=&quot;headerlink&quot;</summary>
        
      
    
    
    
    <category term="分布式架构" scheme="https://winunify.com/categories/%E5%88%86%E5%B8%83%E5%BC%8F%E6%9E%B6%E6%9E%84/"/>
    
    
    <category term="Apache Flink" scheme="https://winunify.com/tags/Apache-Flink/"/>
    
    <category term="Node.js" scheme="https://winunify.com/tags/Node-js/"/>
    
    <category term="Redux" scheme="https://winunify.com/tags/Redux/"/>
    
    <category term="测试" scheme="https://winunify.com/tags/%E6%B5%8B%E8%AF%95/"/>
    
    <category term="事件驱动" scheme="https://winunify.com/tags/%E4%BA%8B%E4%BB%B6%E9%A9%B1%E5%8A%A8/"/>
    
  </entry>
  
  <entry>
    <title>构建从PostgreSQL到Elasticsearch的准实时、最终一致性同步管道</title>
    <link href="https://winunify.com/1267129525/"/>
    <id>https://winunify.com/1267129525/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.203Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;我们面临一个在分布式系统中极为常见但又充满挑战的问题：如何维持一个事务型数据库（PostgreSQL）和一个搜索系统（Elasticsearch）之间的数据一致性。业务要求对数据的查询维度非常复杂，单纯依赖PostgreSQL的索引难以满足性能和功能需求，引入Elastic</summary>
        
      
    
    
    
    <category term="数据工程" scheme="https://winunify.com/categories/%E6%95%B0%E6%8D%AE%E5%B7%A5%E7%A8%8B/"/>
    
    
    <category term="消息队列" scheme="https://winunify.com/tags/%E6%B6%88%E6%81%AF%E9%98%9F%E5%88%97/"/>
    
    <category term="搜索" scheme="https://winunify.com/tags/%E6%90%9C%E7%B4%A2/"/>
    
    <category term="分布式一致性" scheme="https://winunify.com/tags/%E5%88%86%E5%B8%83%E5%BC%8F%E4%B8%80%E8%87%B4%E6%80%A7/"/>
    
    <category term="CDC" scheme="https://winunify.com/tags/CDC/"/>
    
    <category term="Debezium" scheme="https://winunify.com/tags/Debezium/"/>
    
  </entry>
  
  <entry>
    <title>使用Jib构建内嵌Hugging Face模型与SQLite向量存储的独立搜索服务</title>
    <link href="https://winunify.com/1872746326/"/>
    <id>https://winunify.com/1872746326/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.203Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;团队内部一个监控看板需要一个简单的日志搜索功能。需求很明确：能在数万条结构化日志中，根据自然语言描述找到相关的异常信息。常规的&lt;code&gt;grep&lt;/code&gt;或者&lt;code&gt;LIKE&lt;/code&gt;查询效果差强人意，而引入Elasticsearch或一个专门的向量数据库，对</summary>
        
      
    
    
    
    <category term="后端架构" scheme="https://winunify.com/categories/%E5%90%8E%E7%AB%AF%E6%9E%B6%E6%9E%84/"/>
    
    
    <category term="LLM" scheme="https://winunify.com/tags/LLM/"/>
    
    <category term="Jib" scheme="https://winunify.com/tags/Jib/"/>
    
    <category term="Search" scheme="https://winunify.com/tags/Search/"/>
    
    <category term="Hugging Face Transformers" scheme="https://winunify.com/tags/Hugging-Face-Transformers/"/>
    
    <category term="SQLite" scheme="https://winunify.com/tags/SQLite/"/>
    
  </entry>
  
  <entry>
    <title>基于Kafka事件流实现HBase在Kubernetes上的自适应区域管理</title>
    <link href="https://winunify.com/2619035752/"/>
    <id>https://winunify.com/2619035752/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.203Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;手动干预 HBase region 热点几乎是每个数据工程师都经历过的噩梦。在业务高峰期，某个 region 的读写请求量飙升，导致整个集群的延迟抖动，而此时我们能做的，往往是连上 shell，执行 &lt;code&gt;split&lt;/code&gt; 或</summary>
        
      
    
    
    
    <category term="云原生" scheme="https://winunify.com/categories/%E4%BA%91%E5%8E%9F%E7%94%9F/"/>
    
    
    <category term="HBase" scheme="https://winunify.com/tags/HBase/"/>
    
    <category term="容器编排" scheme="https://winunify.com/tags/%E5%AE%B9%E5%99%A8%E7%BC%96%E6%8E%92/"/>
    
    <category term="Kafka" scheme="https://winunify.com/tags/Kafka/"/>
    
    <category term="Kubernetes" scheme="https://winunify.com/tags/Kubernetes/"/>
    
    <category term="Operator" scheme="https://winunify.com/tags/Operator/"/>
    
  </entry>
  
  <entry>
    <title>使用Puppet自动化部署由Weaviate、Apache Iceberg与MariaDB构成的混合特征存储架构</title>
    <link href="https://winunify.com/2810506286/"/>
    <id>https://winunify.com/2810506286/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.203Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;我们的机器学习平台最初陷入了一片混乱。特征工程管道的每个组件——离线批处理、在线实时查询、向量相似性检索——都由不同团队手动部署和维护。环境漂移成了家常便饭，开发环境的一个“小”配置更新，在生产环境就可能引发雪崩式的故障。问题的根源在于我们缺乏一个统一、声明式的基础设施管理</summary>
        
      
    
    
    
    <category term="数据工程" scheme="https://winunify.com/categories/%E6%95%B0%E6%8D%AE%E5%B7%A5%E7%A8%8B/"/>
    
    
    <category term="Puppet" scheme="https://winunify.com/tags/Puppet/"/>
    
    <category term="Weaviate" scheme="https://winunify.com/tags/Weaviate/"/>
    
    <category term="Apache Iceberg" scheme="https://winunify.com/tags/Apache-Iceberg/"/>
    
    <category term="MariaDB" scheme="https://winunify.com/tags/MariaDB/"/>
    
    <category term="MLOps" scheme="https://winunify.com/tags/MLOps/"/>
    
    <category term="IaC" scheme="https://winunify.com/tags/IaC/"/>
    
  </entry>
  
  <entry>
    <title>基于Spring Boot与Cassandra构建支持DVC版本追溯的高吞吐实时特征API</title>
    <link href="https://winunify.com/3017668679/"/>
    <id>https://winunify.com/3017668679/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.203Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;在任何严肃的机器学习系统中，训练-服务偏斜（Training-Serving</summary>
        
      
    
    
    
    <category term="数据工程" scheme="https://winunify.com/categories/%E6%95%B0%E6%8D%AE%E5%B7%A5%E7%A8%8B/"/>
    
    
    <category term="Spring Boot" scheme="https://winunify.com/tags/Spring-Boot/"/>
    
    <category term="Loki" scheme="https://winunify.com/tags/Loki/"/>
    
    <category term="API 设计" scheme="https://winunify.com/tags/API-%E8%AE%BE%E8%AE%A1/"/>
    
    <category term="Cassandra" scheme="https://winunify.com/tags/Cassandra/"/>
    
    <category term="DVC" scheme="https://winunify.com/tags/DVC/"/>
    
  </entry>
  
  <entry>
    <title>构建基于Phoenix、InfluxDB与Ant Design的统一实时指标网关</title>
    <link href="https://winunify.com/4364980208/"/>
    <id>https://winunify.com/4364980208/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.203Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;我们面临的第一个问题是指标孤岛。数十个微服务各自通过不同的方式暴露Prometheus端点、写入日志或直接推送数据到消息队列。运维团队需要维护一个庞杂的监控栈，而开发团队想要排查一个跨服务的请求链路问题，则需要在多个系统之间来回跳转。我们需要一个统一的入口，一个高性能的指标</summary>
        
      
    
    
    
    <category term="后端架构" scheme="https://winunify.com/categories/%E5%90%8E%E7%AB%AF%E6%9E%B6%E6%9E%84/"/>
    
    
    <category term="Phoenix" scheme="https://winunify.com/tags/Phoenix/"/>
    
    <category term="InfluxDB" scheme="https://winunify.com/tags/InfluxDB/"/>
    
    <category term="Ant Design" scheme="https://winunify.com/tags/Ant-Design/"/>
    
    <category term="网关与代理" scheme="https://winunify.com/tags/%E7%BD%91%E5%85%B3%E4%B8%8E%E4%BB%A3%E7%90%86/"/>
    
    <category term="状态管理" scheme="https://winunify.com/tags/%E7%8A%B6%E6%80%81%E7%AE%A1%E7%90%86/"/>
    
  </entry>
  
  <entry>
    <title>基于 Pulsar 不可变日志与 TimescaleDB 读模型构建 Azure 高基数 IoT 事件溯源管道</title>
    <link href="https://winunify.com/4600880425/"/>
    <id>https://winunify.com/4600880425/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.203Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;处理工业物联网（IIoT）数据流的挑战不在于其总量，而在于其结构。一个典型的场景是数百万台设备，每台设备每秒上报数十个遥测点。这种“高基数”特性，即拥有大量唯一时间序列标识符（如设备ID），会迅速摧毁传统时序数据库的索引性能。更棘手的问题是，业务需求不仅仅是存储和查询最新状</summary>
        
      
    
    
    
    <category term="数据工程与存储" scheme="https://winunify.com/categories/%E6%95%B0%E6%8D%AE%E5%B7%A5%E7%A8%8B%E4%B8%8E%E5%AD%98%E5%82%A8/"/>
    
    
    <category term="Pulsar" scheme="https://winunify.com/tags/Pulsar/"/>
    
    <category term="TimescaleDB" scheme="https://winunify.com/tags/TimescaleDB/"/>
    
    <category term="Azure" scheme="https://winunify.com/tags/Azure/"/>
    
    <category term="EventSourcing" scheme="https://winunify.com/tags/EventSourcing/"/>
    
    <category term="IoT" scheme="https://winunify.com/tags/IoT/"/>
    
  </entry>
  
  <entry>
    <title>构建ClickHouse高吞吐异步写入客户端的C++实践及其GitOps声明式管理</title>
    <link href="https://winunify.com/4702474191/"/>
    <id>https://winunify.com/4702474191/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;最初的问题很简单：我们需要将海量的遥测事件从C++服务集群实时写入ClickHouse。最初的实现也同样简单，每个事件都通过一个HTTP</summary>
        
      
    
    
    
    <category term="后端架构" scheme="https://winunify.com/categories/%E5%90%8E%E7%AB%AF%E6%9E%B6%E6%9E%84/"/>
    
    
    <category term="C++" scheme="https://winunify.com/tags/C/"/>
    
    <category term="ClickHouse" scheme="https://winunify.com/tags/ClickHouse/"/>
    
    <category term="CI/CD" scheme="https://winunify.com/tags/CI-CD/"/>
    
    <category term="GitOps" scheme="https://winunify.com/tags/GitOps/"/>
    
    <category term="高性能" scheme="https://winunify.com/tags/%E9%AB%98%E6%80%A7%E8%83%BD/"/>
    
  </entry>
  
  <entry>
    <title>构建从 Pandas 到 Zustand 的实时数据流架构：一种 Node.js 驱动的交互式分析前端实现</title>
    <link href="https://winunify.com/5553967003/"/>
    <id>https://winunify.com/5553967003/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;在处理大规模数据集的交互式分析场景中，传统的请求-响应模型往往会遭遇瓶颈。用户在前端界面调整一个筛选参数，可能需要等待后端完成数秒甚至数分钟的完整计算，才能看到结果。这种延迟严重破坏了数据探索的流畅性。我们的目标是构建一个架构，让前端的数据展现能实时响应后端的重计算过程，用</summary>
        
      
    
    
    
    <category term="全栈架构" scheme="https://winunify.com/categories/%E5%85%A8%E6%A0%88%E6%9E%B6%E6%9E%84/"/>
    
    
    <category term="Node.js" scheme="https://winunify.com/tags/Node-js/"/>
    
    <category term="Zustand" scheme="https://winunify.com/tags/Zustand/"/>
    
    <category term="Headless UI" scheme="https://winunify.com/tags/Headless-UI/"/>
    
    <category term="Pandas" scheme="https://winunify.com/tags/Pandas/"/>
    
    <category term="WebSocket" scheme="https://winunify.com/tags/WebSocket/"/>
    
  </entry>
  
  <entry>
    <title>基于 etcd Watch 构建实时配置管道驱动 Pinia 状态动态更新</title>
    <link href="https://winunify.com/4739756994/"/>
    <id>https://winunify.com/4739756994/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;我们团队的微服务体系深度依赖 &lt;code&gt;etcd&lt;/code&gt; 进行服务发现和配置管理。最近，前端团队遇到了一个棘手的状态同步问题。一些关键的业务功能开关、A/B测试的分流配置，都存储在 &lt;code&gt;etcd&lt;/code&gt;</summary>
        
      
    
    
    
    <category term="后端架构" scheme="https://winunify.com/categories/%E5%90%8E%E7%AB%AF%E6%9E%B6%E6%9E%84/"/>
    
    
    <category term="WebSocket" scheme="https://winunify.com/tags/WebSocket/"/>
    
    <category term="Pinia" scheme="https://winunify.com/tags/Pinia/"/>
    
    <category term="微服务" scheme="https://winunify.com/tags/%E5%BE%AE%E6%9C%8D%E5%8A%A1/"/>
    
    <category term="etcd" scheme="https://winunify.com/tags/etcd/"/>
    
    <category term="Golang" scheme="https://winunify.com/tags/Golang/"/>
    
  </entry>
  
  <entry>
    <title>利用 Pulsar Topic 策略解决读写分离架构下的会话一致性问题</title>
    <link href="https://winunify.com/5573346278/"/>
    <id>https://winunify.com/5573346278/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;用户更新了个人资料，点击保存，刷新页面后却看到了旧的信息。这个场景在任何采用数据库读写分离架构的系统中都可能发生。问题的根源在于主库写入成功后，数据同步到从库存在延迟。当用户的下一个读请求被路由到尚未同步完成的从库时，返回的就是旧数据。这严重破坏了用户体验中的“我写即我见”</summary>
        
      
    
    
    
    <category term="分布式架构" scheme="https://winunify.com/categories/%E5%88%86%E5%B8%83%E5%BC%8F%E6%9E%B6%E6%9E%84/"/>
    
    
    <category term="读写分离" scheme="https://winunify.com/tags/%E8%AF%BB%E5%86%99%E5%88%86%E7%A6%BB/"/>
    
    <category term="Pulsar" scheme="https://winunify.com/tags/Pulsar/"/>
    
    <category term="WebSocket" scheme="https://winunify.com/tags/WebSocket/"/>
    
    <category term="MobX" scheme="https://winunify.com/tags/MobX/"/>
    
    <category term="最终一致性" scheme="https://winunify.com/tags/%E6%9C%80%E7%BB%88%E4%B8%80%E8%87%B4%E6%80%A7/"/>
    
  </entry>
  
  <entry>
    <title>基于事件溯源与IaC构建BentoML模型的声明式部署基础设施</title>
    <link href="https://winunify.com/5599875093/"/>
    <id>https://winunify.com/5599875093/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;管理机器学习模型的生命周期是一项比想象中更复杂的任务。当团队从几个模型扩展到几十上百个时，依赖CI脚本、手动配置和环境变更记录的传统方式会迅速演变成一场灾难。模型版本、部署环境、资源配置之间的关系变得模糊不清，每一次发布都伴随着风险，而且几乎没有任何可靠的审计日志来追溯某个</summary>
        
      
    
    
    
    <category term="MLOps" scheme="https://winunify.com/categories/MLOps/"/>
    
    
    <category term="BentoML" scheme="https://winunify.com/tags/BentoML/"/>
    
    <category term="基础设施即代码 (IaC)" scheme="https://winunify.com/tags/%E5%9F%BA%E7%A1%80%E8%AE%BE%E6%96%BD%E5%8D%B3%E4%BB%A3%E7%A0%81-IaC/"/>
    
    <category term="Event Sourcing" scheme="https://winunify.com/tags/Event-Sourcing/"/>
    
    <category term="分布式与中间件" scheme="https://winunify.com/tags/%E5%88%86%E5%B8%83%E5%BC%8F%E4%B8%8E%E4%B8%AD%E9%97%B4%E4%BB%B6/"/>
    
    <category term="Kit" scheme="https://winunify.com/tags/Kit/"/>
    
  </entry>
  
  <entry>
    <title>使用Go、Nginx、Pulsar与SQLite构建高可用的边缘数据采集网关</title>
    <link href="https://winunify.com/6324720261/"/>
    <id>https://winunify.com/6324720261/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;我们面临一个棘手的工程问题：需要从部署在全球上千个边缘节点的设备上，实时采集大量的遥测数据。这些节点所处的网络环境极不稳定，从时常抖动的Wi-Fi到信号微弱的4G网络，无所不有。业务要求是数据绝对不能丢失，即便边缘节点与中心云端的连接中断数小时甚至数天。&lt;/p&gt;
&lt;p&gt;一个</summary>
        
      
    
    
    
    <category term="分布式系统" scheme="https://winunify.com/categories/%E5%88%86%E5%B8%83%E5%BC%8F%E7%B3%BB%E7%BB%9F/"/>
    
    
    <category term="SQLite" scheme="https://winunify.com/tags/SQLite/"/>
    
    <category term="Pulsar" scheme="https://winunify.com/tags/Pulsar/"/>
    
    <category term="Go" scheme="https://winunify.com/tags/Go/"/>
    
    <category term="Nginx" scheme="https://winunify.com/tags/Nginx/"/>
    
  </entry>
  
  <entry>
    <title>构建基于 Lua 微内核与 Docker 的可热插拔高并发 WebSocket 网关</title>
    <link href="https://winunify.com/6577308556/"/>
    <id>https://winunify.com/6577308556/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;项目初期，各个业务线对实时消息推送的需求开始涌现：实时仪表盘、在线协作、消息通知。最初的方案是每个业务团队各自维护一套 WebSocket</summary>
        
      
    
    
    
    <category term="后端架构" scheme="https://winunify.com/categories/%E5%90%8E%E7%AB%AF%E6%9E%B6%E6%9E%84/"/>
    
    
    <category term="API 设计" scheme="https://winunify.com/tags/API-%E8%AE%BE%E8%AE%A1/"/>
    
    <category term="Docker" scheme="https://winunify.com/tags/Docker/"/>
    
    <category term="WebSockets" scheme="https://winunify.com/tags/WebSockets/"/>
    
    <category term="Lua" scheme="https://winunify.com/tags/Lua/"/>
    
  </entry>
  
  <entry>
    <title>构建基于Playwright、NATS JetStream与数据湖的弹性非结构化数据采集总线</title>
    <link href="https://winunify.com/6689847412/"/>
    <id>https://winunify.com/6689847412/</id>
    <published>2023-10-27T10:30:00.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;获取高质量、结构化的API数据是一回事，但现实世界中，大量关键信息被锁定在动态渲染、需要复杂用户交互才能访问的Web应用里。单纯依赖API，我们丢失了用户视角下的真实页面呈现、第三方脚本行为以及通过交互才能触发的数据。最初我们尝试构建一个简单的爬虫集群，用一个中央调度器通过</summary>
        
      
    
    
    
    <category term="数据工程" scheme="https://winunify.com/categories/%E6%95%B0%E6%8D%AE%E5%B7%A5%E7%A8%8B/"/>
    
    
    <category term="Playwright" scheme="https://winunify.com/tags/Playwright/"/>
    
    <category term="Data Lake" scheme="https://winunify.com/tags/Data-Lake/"/>
    
    <category term="NATS" scheme="https://winunify.com/tags/NATS/"/>
    
    <category term="JetStream" scheme="https://winunify.com/tags/JetStream/"/>
    
    <category term="TypeScript" scheme="https://winunify.com/tags/TypeScript/"/>
    
    <category term="MinIO" scheme="https://winunify.com/tags/MinIO/"/>
    
  </entry>
  
  <entry>
    <title>构建基于Java与Zig混合架构的GraphQL感知型深度安全网关</title>
    <link href="https://winunify.com/5197178275/"/>
    <id>https://winunify.com/5197178275/</id>
    <published>2023-10-27T10:15:30.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;h3 id=&quot;一、问题的根源：传统WAF在GraphQL面前的失效&quot;&gt;&lt;a href=&quot;#一、问题的根源：传统WAF在GraphQL面前的失效&quot; class=&quot;headerlink&quot;</summary>
        
      
    
    
    
    <category term="后端架构" scheme="https://winunify.com/categories/%E5%90%8E%E7%AB%AF%E6%9E%B6%E6%9E%84/"/>
    
    
    <category term="GraphQL" scheme="https://winunify.com/tags/GraphQL/"/>
    
    <category term="Java" scheme="https://winunify.com/tags/Java/"/>
    
    <category term="防火墙" scheme="https://winunify.com/tags/%E9%98%B2%E7%81%AB%E5%A2%99/"/>
    
    <category term="Zig" scheme="https://winunify.com/tags/Zig/"/>
    
    <category term="架构设计" scheme="https://winunify.com/tags/%E6%9E%B6%E6%9E%84%E8%AE%BE%E8%AE%A1/"/>
    
  </entry>
  
  <entry>
    <title>基于 AWS EKS 构建利用 Memcached 加速状态流转的 Saga 分布式事务协调器</title>
    <link href="https://winunify.com/6159115739/"/>
    <id>https://winunify.com/6159115739/</id>
    <published>2023-10-27T10:15:21.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;一个看似简单的下单操作，背后联动了订单、库存、积分三个微服务。在一次大促压测中，积分服务因为负载过高出现短暂不可用，导致大量用户的订单创建成功、库存也扣减了，但积分发放失败。数据不一致的烂摊子，花了我和团队整整两天时间手动修复，这个教训足够深刻。传统的两阶段提交（2PC）在</summary>
        
      
    
    
    
    <category term="分布式系统" scheme="https://winunify.com/categories/%E5%88%86%E5%B8%83%E5%BC%8F%E7%B3%BB%E7%BB%9F/"/>
    
    
    <category term="Kubernetes" scheme="https://winunify.com/tags/Kubernetes/"/>
    
    <category term="Ant Design" scheme="https://winunify.com/tags/Ant-Design/"/>
    
    <category term="分布式事务" scheme="https://winunify.com/tags/%E5%88%86%E5%B8%83%E5%BC%8F%E4%BA%8B%E5%8A%A1/"/>
    
    <category term="AWS EKS" scheme="https://winunify.com/tags/AWS-EKS/"/>
    
    <category term="Memcached" scheme="https://winunify.com/tags/Memcached/"/>
    
  </entry>
  
  <entry>
    <title>构建基于 Lit 的 ISR 渲染服务以解决读写分离架构下的会话一致性难题</title>
    <link href="https://winunify.com/0451145073/"/>
    <id>https://winunify.com/0451145073/</id>
    <published>2023-10-27T10:15:20.000Z</published>
    <updated>2026-03-20T14:15:39.203Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;项目上线后，我们收到了第一个紧急工单：“我刚刚更新了商品描述，刷新页面后看到的还是旧内容，要等一分多钟才能看到变化！”。这个反馈并不意外，它精准地击中了我们新架构的阿喀琉斯之踵。为了应对日益增长的读取压力，我们将商品详情页从传统的服务端渲染（SSR）迁移到了增量静态再生（I</summary>
        
      
    
    
    
    <category term="后端架构" scheme="https://winunify.com/categories/%E5%90%8E%E7%AB%AF%E6%9E%B6%E6%9E%84/"/>
    
    
    <category term="Lit" scheme="https://winunify.com/tags/Lit/"/>
    
    <category term="ISR" scheme="https://winunify.com/tags/ISR/"/>
    
    <category term="读写分离" scheme="https://winunify.com/tags/%E8%AF%BB%E5%86%99%E5%88%86%E7%A6%BB/"/>
    
    <category term="BASE" scheme="https://winunify.com/tags/BASE/"/>
    
    <category term="Node.js" scheme="https://winunify.com/tags/Node-js/"/>
    
  </entry>
  
  <entry>
    <title>基于Monorepo架构构建PyTorch模型交互式分析前端的技术选型与实现</title>
    <link href="https://winunify.com/8671873458/"/>
    <id>https://winunify.com/8671873458/</id>
    <published>2023-10-27T10:15:20.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;一个日益普遍的技术难题摆在面前：数据科学团队产出了大量精密的PyTorch模型，但这些模型的价值却被束缚在Jupyter</summary>
        
      
    
    
    
    <category term="全栈架构" scheme="https://winunify.com/categories/%E5%85%A8%E6%A0%88%E6%9E%B6%E6%9E%84/"/>
    
    
    <category term="Zustand" scheme="https://winunify.com/tags/Zustand/"/>
    
    <category term="PyTorch" scheme="https://winunify.com/tags/PyTorch/"/>
    
    <category term="CSS Modules" scheme="https://winunify.com/tags/CSS-Modules/"/>
    
    <category term="Monorepo" scheme="https://winunify.com/tags/Monorepo/"/>
    
  </entry>
  
  <entry>
    <title>构建基于NLP与全链路追踪的实时API安全审计系统</title>
    <link href="https://winunify.com/4652455221/"/>
    <id>https://winunify.com/4652455221/</id>
    <published>2023-10-27T10:00:00.000Z</published>
    <updated>2026-03-20T14:15:39.207Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;传统的Web应用防火墙（WAF）和基于正则表达式的日志扫描，在面对经过编码、混淆或利用业务逻辑漏洞的攻击时，越来越显得力不从心。在我们的微服务体系中，安全日志散落在各个服务，而分布式追踪数据（Trace）则独立存在于Zipkin中。当安全事件发生时，分析师需要手动关联日志和</summary>
        
      
    
    
    
    <category term="安全架构" scheme="https://winunify.com/categories/%E5%AE%89%E5%85%A8%E6%9E%B6%E6%9E%84/"/>
    
    
    <category term="网络安全" scheme="https://winunify.com/tags/%E7%BD%91%E7%BB%9C%E5%AE%89%E5%85%A8/"/>
    
    <category term="NLP" scheme="https://winunify.com/tags/NLP/"/>
    
    <category term="Zipkin" scheme="https://winunify.com/tags/Zipkin/"/>
    
    <category term="Fluentd" scheme="https://winunify.com/tags/Fluentd/"/>
    
    <category term="Valtio" scheme="https://winunify.com/tags/Valtio/"/>
    
  </entry>
  
</feed>
