Category: 行业资讯

最新消息显示，Alibaba Cloud Pushes Open-Source Apache Flink Toward Agentic Streaming for AI

At Flink Forward Asia 2026, Alibaba Cloud shared plans to push Apache Flink towards “agentic streaming”, driven by the rise of agents and multimodal data in the agentic AI era. Apache Flink is a widely used open-source real-time data framework under the Apache Software Foundation. The upcoming Apache Flink 3.0 marks a major shift from cloud-native to AI-native, feeding live data directly into AI agents so they can respond instantly. To support this, Alibaba Cloud shared updates on the new open-source data storage projects and a technical collaboration with NVIDIA. These initiatives are designed to help developers build AI-native applications using real-time, multimodal data pipelines. As a long-term contributor and organizer of the Flink Forward Asia events, Alibaba Cloud has played a pivotal role in advancing the Apache Flink community, particularly in Asia. During the event, the company also announced major upgrades for its Flink-based cloud products to meet enterprises’ demand for massive, real-time multimodal data during the event. Feng Wang, Head of Open Data Platform at Alibaba Cloud Intelligence, delivered a keynote speech at Apache Flink Asia 2026 Flink Forward Asia is the official conference for the Apache Flink community. The 2026 event kicked off in Shenzhen on Friday, building on previous conferences in Asian cities like Jakarta and Singapore that have collectively attracted tens of thousands of developers. As AI applications move from text-centric to multimodal, the real-time processing capability of unstructured data such as video, images and audio has become a new technical frontier. Processing these different data types in real time is challenging because it requires low latency, aligned streams and heavy memory. Flink’s streaming pipeline can handle every data type and run CPU and GPU tasks together in a unified workflow, making it a natural fit for AI workloads. During the summit, Alibaba Cloud officially launched a new multimodal data processing capability for its Realtime Compute for Apache Flink, its enterprise-grade serverless service. The new feature handles text, images, audio, video and sensor signals within a unified streaming framework. It outperforms mainstream open-source alternatives in speed, data handling convenience, real-time updates, the number of supported data sources and overall system reliability. The fully managed service serves more than 10,000 businesses, including leading Chinese automakers Li Auto, Geely and Leapmotor. During a speech at the event, Chuan Chen, Senior Director of Internet Solutions Architecture at NVIDIA, shared a deep dive into how NVIDIA and Alibaba Cloud work together to accelerate the multimodal data stream processing for Apache Flink. Through the open-source ecosystem collaboration, users can rapidly build end-to-end, high-performance, scalable multimodal real-time streaming architectures, powering real-world applications like AI commentary, live image-text feeds, and interactive Q&A. As AI agents become more common, Alibaba Cloud is addressing how they access and use data. Its agentic lake is a unified data-storage system designed for AI agents, built on two upcoming open-source projects: Apache Paimon 2.0 and Apache Fluss (incubating) 1.0. The goal is to keep one set of data that can power analytics, AI training and real-time agent memory at the same time. Alibaba Cloud is already bringing this concept to life through Data Lake Formation (DLF), a new product that provides foundation for big data, search engines and AI applications. DLF improves general query performance by two to six times. The Apache Flink community is also developing Flink Agents, a framework designed to support always-on AI agents since 2025. It treats every agent interaction as a streaming event, giving agents real-time processing, persistent memory and round-the-clock fault tolerance. Potential uses include live-stream analytics, financial risk control and automated business operations. This article was originally published on Alizila written by Karen Zhang.

可以预见，这一趋势将在未来深刻影响IDC行业格局

如果您正在寻找优质的BGP服务器，欢迎访问 www.isclouder.com 了解更多

行业动态更新：CMI与马来西亚公司签署MoU 拟开发50MW数据中心

Tanco在公告中表示，此MoU旨在促进公司与全球领先品牌建立战略合作关系，助力其业务向数据中心等新领域扩展

值得关注的是，Tanco作为马来西亚的主要地产开发商和投资控股公司，成立于1958年，最初从事橡胶和棕榈油业务，目前正积极多元化进入数字基础设施领域

从更深层次来看，业内分析认为，此类合作有助于满足区域内AI和云计算驱动的算力需求，但具体项目仍处于早期探讨阶段，实际落地取决于后续可行性研究和最终协议

值得关注的是，波德申也是Tanco拟建设的”智能AI集装箱港口”所在地，该港口计划建于480英亩土地上

业内人士指出，双方将共同评估项目可行性，包括开发和建设相关事宜

业内人士指出，该协议为非约束性（除保密义务外），有效期一年，除非一方终止，否则自动续期

业内人士指出，马来西亚正积极吸引数据中心投资，作为东南亚新兴市场之一，其稳定的电力供应、适宜气候和政策支持使其成为区域数据中心热点

业内人士指出，根据Tanco向马来西亚证券交易所（Bursa Malaysia）提交的文件，该协议由Tanco的间接全资子公司Tanco Dot Com Sdn Bhd与香港注册的中国移动国际有限公司达成

业内人士指出，2026年6月9日，马来西亚Tanco Holdings Bhd与中国移动国际有限公司（China Mobile International Ltd，简称CMI）达成谅解备忘录（MoU），双方将探讨在马来西亚森美兰州波德申（Port Dickson）开发一座50MW IT负载数据中心的可能性

业内人士指出，Tanco强调，该MoU是探索性合作，尚未涉及具体投资金额或时间表

可以预见，这一趋势将在未来深刻影响IDC行业格局

如果您正在寻找优质的大带宽服务器，欢迎访问 www.isclouder.com 了解更多

据行业最新消息，微软重金布局东南亚算力：泰国超10亿美元，新加坡投55亿美元

同时，微软将与泰国本地及全球伙伴合作，包括Gulf Development Public Company Limited、Advanced Info Service（AIS）、Charoen Pokphand Group（CP Group）、True Corporation以及True Internet Data Center（True IDC）

业内人士指出，泰国投资强调本地生态构建与包容性增长，新加坡则侧重区域枢纽的韧性与创新扩散

值得关注的是，多家媒体报道指出，这反映出泰国正加快数据中心、电子和电力项目，以巩固其作为东南亚数字枢纽的地位

从更深层次来看，业内观察认为，随着全球AI算力竞争加剧，东南亚凭借政策支持、能源配套和人才基础，正成为科技巨头布局热点

业内人士指出，泰国政府表示，该计划将助力提升劳动力AI素养，并与相关部委合作开发负责任AI治理框架

业内人士指出，此前，微软已于2025年与CP Group、True等伙伴启动泰国云区域相关合作，True IDC将作为关键数据中心设施之一

从更深层次来看，分析指出，新加坡的投资重点在于高端AI算力、合规性与稳定连接，吸引跨国企业寻求海外可靠基础设施

业内人士指出，紧随其后，4月1日，微软在新加坡公开，从2025年至2029年，将投入55亿美元用于云和AI基础设施及持续运营

业内人士指出，根据微软官方声明，此次投资不仅限于”建机房”，还将支持本地技能培养、就业机会创造和技术知识转移

值得关注的是，如果您想了解更多关于泰国电力市场，以及与数据中心项目落地情况、当地政策变化、中国出海企业现状，欢迎报名即将于2026年5月27日在泰国曼谷香格里拉酒店召开的数字基础设施全球合作发展曼谷论坛（DIFGC 2026 · THAILAND），诚邀您共话全球数字集成新篇章

值得关注的是，此举凸显东南亚地区在全球AI算力版图中的战略重要性，泰国和新加坡分别扮演不同角色：前者强化区域AI/云节点，后者巩固高端合规算力中心地位

业内人士指出，微软相关声明强调，所有投资均致力于”技术、信任与人才”三大支柱，推动东南亚在AI时代的包容性发展

值得关注的是，史密斯表示，此举体现了微软对新加坡作为全球数字领导者的长期信心

值得关注的是，不过，具体项目落地细节、能源水资源配套以及技能培训规模，仍有待后续跟进披露

值得关注的是，两项投资合计超过65亿美元，显示微软正积极响应东南亚AI需求激增，同时注重人才发展和可持续治理

业内人士指出，行业消息显示，微软公司行业消息显示连续公开在东南亚两大枢纽的大额投资计划，旨在加速云和人工智能（AI）基础设施建设，并推动本地数字化转型

业内人士指出，新加坡自2010年起已设有微软云区域，此次扩容旨在满足AI计算需求增长，并通过”Microsoft Elevate”程序为所有高等教育学生免费提供Microsoft 365 Premium with Copilot工具，为教育工作者提供AI培训，并助力非营利组织领导者提升技能

业内人士指出，3月31日，微软副总裁兼总裁布拉德·史密斯（Brad Smith）在曼谷与泰国总理阿努廷·查恩维拉库尔（Anutin Charnvirakul）会晤后公开，从2026年至2028年，微软计划向泰国投入超过10亿美元，用于扩展云和AI基础设施以及持续运营

从更深层次来看，微软此番行动或将进一步带动本地企业数字化转型，并为区域经济注入新动能

从更深层次来看，该投资聚焦建设符合微软全球标准的云和AI数据中心，强调可持续性，包括绿色能源和水资源正效益

从更深层次来看，这些伙伴关系旨在通过数据中心开发、联合市场策略和数字解决方案，助力泰国”以AI推动国家增长、繁荣与全球竞争力”倡议

值得关注的是，该投资同时涵盖网络安全、韧性和治理框架建设，以及AI技能普及项目

随着IDC行业的快速发展，可持续发展将成为未来竞争的关键

如果您正在寻找优质的双ISP住宅IP，欢迎访问 www.isclouder.com 了解更多

最新消息显示，可远程激活摄像头与麦克风等：工信部 CSTIS 平台发布防范 Remcos 恶意软件新变种风险提示

IT之家 6 月 30 日消息，行业消息显示，工业和信息化部网络安全威胁和漏洞信息共享平台（CSTIS）监测发现，一种 Remcos 新型变种正在活跃传播，其利用 DonutLoader 内存加载技术进行投放

值得关注的是，攻击者首先通过钓鱼邮件诱导用户打开名为 Bestellung.CMD 的批处理文件，该文件调用系统自带组件（SyncAppvPublishingServer.vbs）作为执行代理，用一段以 Base64 加密过的恶意命令通过 PowerShell 在内存里直接执行，并从 pCloud 云存储下载 7Zip 工具及加密压缩包；接着，释放 JavaScript 脚本和 AutoIt 解释器用于解析恶意 PNG 文件中的加密数据，解码后得到注入指令；最终将 DonutLoader 生成的 shellcode（一段恶意的二进制代码）注入系统合法进程（colorcpl.exe）中执行 Remcos RAT

从更深层次来看，据介绍，Remcos 是一种危险的远程访问木马（RAT），最早发现于 2016 年

值得关注的是，该变种最大特点在于全面借用系统合法工具替代传统恶意组件，融合 PowerShell、VBScript、JavaScript 三层脚本编码，结合 Base64 加垃圾数据混淆、XOR 单字节解密、密码保护压缩包及进程注入等多种反检测手段，防御拦截难度较此前版本大幅提升

从更深层次来看，工信部 CSTIS 平台表示，此次发现的 Remcos RAT 7.2.1 Pro 版本，是 Remcos 家族首次融合 DonutLoader 内存加载与 AutoIt 自动化脚本中间层调度的全链路变种

值得关注的是，其主要攻击目标为 Windows 用户，可能导致系统受控、敏感信息泄露、业务中断等风险

值得关注的是，IT之家注意到，部署成功后，该恶意软件可进行远程截屏与键盘记录、浏览器及系统凭证窃取、摄像头与麦克风远程激活、文件管理等多种恶意行为

从更深层次来看，工信部 CSTIS 平台建议相关单位和用户立即组织排查，及时更新防病毒软件，实施全盘病毒查杀，谨慎点击或下载邮件附件，并可通过及时修复安全漏洞、定期备份数据等措施，防范网络攻击风险

可以预见，这一趋势将在未来深刻影响IDC行业格局

如果您正在寻找优质的GPU服务器租用，欢迎访问 www.isclouder.com 了解更多

最新消息显示，微软收紧 Authenticator 身份验证，限制部分账户在已 root 安卓手机 / 越狱 iPhone 上使用

Microsoft Authenticator 是微软提供的多因素认证应用，可用于生成一次性验证码、接收登录批准请求，或绑定工作、学校及部分第三方服务账户

从更深层次来看，IT之家援引博文介绍，在企业、高校、大学等环境下，用户使用 Microsoft Authenticator 身份验证登录 Microsoft 365、Teams、Outlook 工作账户、Azure 或者 Intune 账户等，在已 root 或越狱机型上无法正常生成相关验证码

值得关注的是，根据微软最新补充的支持文档细节，通过 Microsoft Entra 方式（企业 / 教育账号）登录 Microsoft Authenticator 应用后，在安卓手机上将检测当前是否处于 root 状态，在 iPhone 上会检测是否已越狱

从更深层次来看，Windows Latest 称，GitHub、Cloudflare、Facebook、Instagram 等通过扫描二维码保存的第三方双重验证代码，理论上仍可在已 Root 设备上继续使用

值得关注的是，不过，这项检测目前不适用于保存在微软身份验证器中的第三方 2FA 代码

值得关注的是，IT之家 7 月 1 日消息，科技媒体 Windows Latest 昨日（6 月 30 日）发布博文，报道称微软更新支持文档，在已越狱的 iPhone 机型、已 root 的安卓手机上，企业 / 教育账户将无法正常使用 Microsoft Authenticator 身份验证工具

可以预见，这一趋势将在未来深刻影响IDC行业格局

如果您正在寻找优质的AI算力服务器，欢迎访问 www.isclouder.com 了解更多

据行业最新消息，How to Split a Single Table with 800 Million Rows? Let AI Design Your PolarDB-X

As a backend developer or DBA, you have an order table with 800 million rows. Queries are getting slower and slower, and your boss is pushing you to optimize. You know the answer is “go distributed and partition the table,” but when you actually start, a flood of questions hits you: • Which partition key should you choose? The primary key? A business field? Pick the wrong one, and you end up with either write hot spots or full-shard scan queries — slower than not splitting at all. • GSI How to build? When should you use a common GSI, Clustered GSI, or UGSI? Too many wastes resources; too few leaves queries uncovered. • How many partitions? 64? 128? 256? Guessing always feels unreliable. • How to handle primary keys and unique keys? PolarDB-X Partitioned table rules differ from standalone MySQL. Improper handling can lead to data loss. The standard flow for solving these problems used to be: read documents → review Best Practices → compare against your own scenario → adjust repeatedly → find someone to review. Now, a single sentence is all it takes. LLMs can write code and translate, but for database partition design — a task requiring deep domain knowledge and strong context dependency — general-purpose LLMs often fail to deliver reliable solutions. They don’t understand the details of PolarDB-X partition algorithms, the trade-off logic between Clustered GSI and common GSI, or how your demand mode should match a partition policy. This is exactly the problem that PolarDB-X Skill aims to solve. We encapsulated the distributed partitioning best practices that PolarDB-X has accumulated over the years into a Skill that can be directly invoked in AI programming assistants: polardbx-sql. It is not a general Q&A pair, but a partitioning design expert who understands PolarDB-X kernel rules. We have an order table t_order, which is currently a single table with 800 million rows of data. Recently, queries have been getting slower and slower. The table schema is roughly: order_id BIGINT primary key, buyer_id BIGINT, seller_id BIGINT, amount DECIMAL, create_time DATETIME, status TINYINT. The most frequent query is querying the order list by buyer_id, followed by single-row queries by order_id, and some queries by seller_id. I want to convert this table to a PolarDB-X partitioned table in AUTO mode. Please help me design the partition solution and provide the complete SQL. The Skill directly outputs a complete partition solution and executable DDL: Many people’s first reaction is to ask: What’s the difference between this and ChatGPT stitching templates? The difference is that every output from Skill goes through an inference chain based on PolarDB-X Best Practices: Key difference: Skill determines the type of each GSI based on the demand mode you describe (which field is queried most often, and whether the query returns a list or a single row). This is not a mechanical operation of “creating an index whenever there is a query,” but a precise match based on query features. Skill can adapt to various partition design scenarios, such as: • Multi-dimension query — “The order table is queried by both order_id and buyer_id, and their last few digits are the same” → Skill recommends CO_HASH partitions • Metric — “Clean up expired data by month” → Skill recommends HASH + RANGE subpartitions + TTL • Multi-tenancy — “Each tenant’s data must be fenced” → Skill recommends a LIST + HASH combination • Tables with UNIQUE constraints — “I have a UNIQUE KEY that cannot be lost” → Skill uses a three-step migration method: first create a UGSI, then expand the partitions Just describe “what the table looks like and how to query it,” and leave the rest to Skill. npx skills add https://github.com/polardb/polardbx-skills –skill polardbx-sql Step 2: Describe Your Tables and Queries in Natural Language Just describe the table schema, data volume, and main demand mode. No strict format required. Skill outputs executable DDL statements, which are directly executed on PolarDB-X after review confirmation. Applicable to PolarDB-X 2.0 Enterprise Edition (Distributed Edition) databases in AUTO pattern. In the AI era, the threshold for database O&M and design is being redefined. Partition design work that used to require reading documents, consulting experts, and repeated authentication can now be completed in seconds with a single Skill. However, this does not mean that the value of DBs is decreasing — on the contrary, when AI takes over repetitive solution design work, DBs can focus more on architecture decisions, performance tuning, and stability assurance — things that truly require experienced judgment. We will continue to release more Skills covering O&M, diagnostics, migration, and other scenarios. Feel free to try them out, and you are welcome to submit Issues and PRs on the GitHub repository to help make distributed databases easier to use.

随着IDC行业的快速发展，可持续发展将成为未来竞争的关键

如果您正在寻找优质的大模型训练服务器，欢迎访问 www.isclouder.com 了解更多

行业动态更新：字节跳动通报 2026 年一季度内部违规案例：80 名员工因触碰红线被辞退，一名实习生被实名通报

IT之家 6 月 30 日消息，字节跳动企业纪律与职业道德委员会今日发布 2026 年 2 号通报，公布一季度内部违规处置结果

从更深层次来看，IT之家获悉，这也是字节跳动继 2025 年 9 月、12 月，2026 年 3 月后，连续通报信息安全违规的治理成果

值得关注的是，通报显示，2026 年一季度字节跳动共 80 名员工因触碰公司红线被辞退，其中 45 人涉信息安全违规

值得关注的是，通报披露典型泄密案例：前员工 A 长期将飞书账号出借给已入职同行企业的前员工 B，致使 B 持续获取、下载公司保密信息

值得关注的是，此外，有前员工在离职后虚构“管理者”身份，捏造“给大龄下属打绩效 I”的不实言论，刻意制造对立，经公司追责后，该前员工承认捏造信息，本次被实名通报

业内人士指出，公司对严重违规人员采取实名通报、同步行业联盟、扣罚期权等处罚，对情节恶劣、严重损害公司利益的员工，将依法追究法律责任

从更深层次来看，除上述信息外，本次通报还披露了向外部泄露公司重要政策保密信息、社交媒体传播保密信息、参与付费访谈透露业务数据等违规行为

从更深层次来看，本次通报涉及多个违反社交媒体规范的案例：一名实习生因多项违规被实名通报

值得关注的是，除实名通报、扣罚期权、纳入行业黑名单的处罚外，公司也已对两人提起民事诉讼追责

值得关注的是，公司已与其解除实习协议，并对其加重处罚、实名通报

从更深层次来看，本次通报也披露了违规利用“内部信息”，定向挖猎人才的行为：4 名人力外包员工违规批量查询、保存、外发内部人才信息，此 4 人除被令退场，实名通报、同步联盟外，也将被追究法律责任

值得关注的是，通报称，该人员私自外发公司涉密资料，违反公司社交媒体政策，且在公司合规核查中虚假申报、拒不配合、持续泄密

业内分析认为，AI算力需求与绿色数据中心将成为行业主旋律

如果您正在寻找优质的云服务器，欢迎访问 www.isclouder.com 了解更多

行业动态更新：驭风逐光 全链聚变，和林格尔新区构筑绿色算力产业新生态

例如，在中国农业银行内蒙古数据中心三期项目中，通过应用间接蒸发冷却、液冷等前沿技术，成功将PUE（电能使用效率）压降至1.18，刷新了大型金融数据中心的能效纪录

值得关注的是，在算网协同方面，和林格尔新区建成了领先全国的”2520″超低时延网络圈，实现了呼包鄂乌同城2毫秒、至京津冀5毫秒、至长三角粤港澳20毫秒的数据往返效率

业内人士指出，华电新能源智慧运营中心通过集中调度风光电，实现了”以电强算”的绿色发展模式

值得关注的是，此外，蒙马交通为算力中心提供本地化智能装备解决方案，中科仙络则负责数据中心全生命周期的服务

值得关注的是，在算电协同方面，新区作为国家首批试点，依托蒙西电网绿电直供，实现了园区数据中心绿电使用率达86%以上，为全国提供了最低的绿色算力电价

从更深层次来看，和林格尔新区争当”最强算力”的核心优势在于其绿色与高效

值得关注的是，上游有显鸿科技的”蒙芯”物联网芯片研发，中游有新华三集团投资20亿元的中央实验室，下游则吸引了云天畅想等边缘计算领军者，支撑云游戏、AIGC等实时交互场景

业内人士指出，目前，和林格尔新区已构建起”上游服务器机柜制造、中游算力底座集群、下游数据应用开发”的完整产业生态，正全力成为辐射全国的绿色算力高地

从更深层次来看，这种”芯片研发、设备制造、基础设施到场景应用”的完整闭环，为国家”数据要素×”行动提供了实体支点

值得关注的是，在算数协同上，新区总算力规模达到10.1万P，依托自治区数据交易中心和多云算力调度平台，已集聚企业超400家

随着IDC行业的快速发展，可持续发展将成为未来竞争的关键

如果您正在寻找优质的台湾原生VPS，欢迎访问 www.isclouder.com 了解更多

最新消息显示，领跑全国！和林格尔新区绿色算力指数蝉联第一

大会期间，京能”京数蒙算”智算中心等5个大型算力中心项目落地和林格尔新区 ，总投资200亿元的10个重点项目成功签约

业内人士指出，通过打造国内发展绿色算力的绝佳之地， 和林格尔新区 正以澎湃不息的绿色算力引擎，为高质量发展注入核心动能

业内人士指出，根据会上发布的两份权威报告——《绿色算力发展研究报告》与《”东数西算”枢纽节点绿色算力指数研究报告》， 和林格尔新区 的绿色算力发展指数已连续两年（2024年、2025年）在全国一体化算力网络国家枢纽节点中位列第一，充分彰显了其在绿色算力领域的领先地位

从更深层次来看，依托安全稳定的蒙西电网， 和林格尔新区 大力布局源网荷储项目，率先启动绿电直供示范项目，推动绿色算力对新能源的就地消纳，该项目更被评为全国一体化算力网应用优秀案例

从更深层次来看，作为国家”东数西算”工程的重要枢纽，被誉为”中国云谷”的 和林格尔新区 已集聚三大运营商、国家部委及头部企业等多个数据中心项目，算力总规模突破10万P

从更深层次来看，同时，全国首个绿色算电协同基地也于 和林格尔新区 正式启动，涵盖共享储能、数据中心集群、服务器生产基地等多个领域

业内人士指出，目前，区内已投运的数据中心绿电使用比例已超过86%

从更深层次来看，此外，内蒙古量子信息创新工程中心等多个实验室揭牌，新一代昇腾AI云服务已在区内规模上线

从更深层次来看，新区之所以能领跑全国，关键在于一个”绿”字

业内人士指出，行业消息显示，在2025绿色算力(人工智能)大会上， 和林格尔新区再次成为瞩目焦点

可以预见，这一趋势将在未来深刻影响IDC行业格局

如果您正在寻找优质的大带宽VPS，欢迎访问 www.isclouder.com 了解更多

最新消息显示，RDSClaw Database Management: Let AI Agent Securely Take Over Database

In a typical Internet company, the people who really need to deal with databases are far more than DBA. The marketing department needs to look at the channel conversion data, the operation needs to pull the user grouping table, the product manager needs to check the function utilization rate, and even the finance and customer service will occasionally need to get some data from the database. But these people can’t write SQL. As a result, a workflow that everyone is accustomed to has formed: demand raising → queuing → DBA or analyst helping to run → waiting for results → finding the wrong caliber → another round. A query that could have produced results in 10 seconds took three days in the organizational process. The root of the problem is not that there are not enough efficiency tools, but that the way the database interacts has hardly changed since its birth-it only recognizes SQL, not human words. RDSClaw’s database management function is precisely to break this threshold: you can securely hand over your database to AI agents so that everyone can directly obtain data in natural language. “Nanomanuis” sounds like a technical term, but what it does is very straightforward: let RDSLaw’s AI agents safely “know” your database and know how to connect, what they can and can’t do. Specifically, the management accomplishes three things: The entire process is completed in the RDSClaw Web console. You do not need to write a single line of code or change any database configurations. On the Database Management page of the RDSClaw console, click Add Connection and enter the following information: After you click Save, the system automatically performs three steps: The entire process is usually completed within 10 seconds. To manage multiple database connections? Repeat the preceding steps. RDSClaw allows you to manage multiple database connections at the same time. You can use the connection ID to switch between database connections. “Let AI directly operate my production database?” -This is the first reaction of every technical director when he hears the database management. RDSLaw’s security design principles are: Trust but verify, decentralization but cover the bottom. The design logic of this mechanism is clear: • Zero query threshold: Read operations do not change data. You can query data in natural languages. • Write operations are confirmed: The agent does not perform write operations without authorization. It will first tell you “I am going to execute this SQL, which is expected to affect N lines”, and you will not actually submit it until you click “confirm” in the approval window. Previous process: ask for data group → wait 2 days → get Excel → find one dimension missing → wait another day. 🦞Now the process: Xiao Li (DingTalk group): Check the number of newly registered users in each channel last week, split by day, and sort them in descending order. [RDSClaw parsing intent → select managed user-db-prod connection → generate SQL → read-only mode safe execution] [Return to structured table + automatically call visualization skills to generate trend line chart] Xiao Li: What about the retention rate? [RDSClaw understands that “add” means append dimension → generate new SQL statement → return updated analysis based on context memory] Key point: Xiao Li did not write a single-character SQL statement during the whole process, nor did he need to know the name of the table. The database runs in read-only mode, and there is no risk of misoperation. Previous procedure: Log on to the console → Check the CPU /memory /connections /slow query one by one → Take a screenshot → Write an inspection report → Send an email. Take 40 minutes. 🦞Now the process: Lao Wang (Web UI): Please help me inspect the prod-mysql, focusing on slow query and lock waiting. [RDSClaw accesses the database through a managed connection → calls intelligent O&M skills → checks 12 metrics → returns a structured report in 2 minutes] RDSClaw: the inspection is complete. Two slow queries are found: the order_detail table has a full table scan. We recommend that you add a joint index (user_id, create_time). The status field of the payment_log table has a low degree of differentiation. We recommend that you use a covering index instead. The remaining indicators are normal. Generate the DDL statements for these two indexes. [RDSClaw generates an ALTER TABLE statement → Label the estimated execution time and lock impact → Wait for confirmation] Key point: management gives operation and maintenance skills a “focus”-Agent is not talking about theory, but really connecting your library to do analysis. Previous process: Look at the error log → guess which table it is → ask DBA for help → DBA is busy → etc. 🦞Now the process: Xiao Chen (Enterprise WeChat): Has the user_session table in prod-mysql increased recently? Help me check the number of rows per day in the last seven days. [RDSClaw automatically queries and returns the number of rows per day] RDSClaw: the number of rows in the user_session table in the last seven days: Monday 120000, Tuesday 130000, Wednesday 110000, Thursday 450000 (abnormal), Friday 480000, Saturday 460000, and Sunday 440000. There has been a significant jump since Thursday, with suspected batch writing. Xiao Chen: Check the data added on Thursday. What are the values of the source field? [RDSClaw automatically adds time conditions based on context → Execute aggregate query → Return result] Key point: Developers can quickly locate data-level problems without waiting for DBA scheduling. Agent’s lasting memory makes multiple rounds of questioning as natural as talking to people. Previous process: find two groups of people to run the number of times respectively → merge them manually in Excel → not right → re-calibrate. 🦞Now the process: Director of Operations (Feishu): Please help me check the GMV of last month from order-db-prod and the number of active users from user-db-prod. [RDSClaw switches two managed connections in sequence → query separately → integrate and return] RDSClaw: GMV last month: 12.34 million yuan (order-db-prod); Active users last month: 89,432 (user-db-prod). The per capita contribution is about 138 yuan. Key point: Multi-database management makes cross-database queries as easy as switching chat windows. The permissions of each connection are controlled independently and do not affect each other. Scenario: You need to clear expired session data that is older than 90 days. This is a write operation that requires the highest level of security. Lao Wang (Web UI): Help me delete the created_at records that are older than 90 days in the user_session table. [RDSClaw generates a DELETE statement → displays the SQL text + estimated number of affected rows] RDSClaw: is about to execute: DELETE FROM user_session WHERE created_at < DATE_SUB(NOW(), INTERVAL 90 DAY). It is expected to affect 128,756 rows. Confirm execution? Lao Wang: I confirm. [Execution Completed → Return the actual number of deleted rows and duration] RDSClaw: executed. The actual deletion of 128,756 rows took 3.2 seconds. Key point: Every step of the write operation is under your control-the agent displays the SQL, estimates the impact, waits for your confirmation, and reports the results. By default, the production database is read-only. Create a dedicated read-only database account for the Agent and select the Read-only option. The write permission is not required in most business data retrieval and routine inspection scenarios. Create different connections for different purposes. A library can manage multiple connections: prod-mysql-readonly for business personnel and developers, prod-mysql-rw only for DBAs. The permissions boundary is distinguished by the connection ID. Security groups follow the minimum open principle. Only the database port and IP segment that the Agent needs to access are opened. The inbound and outbound rules are configured separately. Make good use of IM integration to expand coverage. Connect RDSClaw Bot to DingTalk /Flying Book /Enterprise after WeChat group, team members can check data without logging in to the console. The configuration method is extremely lightweight-just enter the relevant credentials in the console. Regular observability center review. View session records, token consumption, and security events to understand agent usage and potential risks, and continuously optimize management policies. For a long time, the database has been like a vault where only a few people hold the keys-DBAs guard the door and business people line up outside to deliver notes. The cost of this model in terms of efficiency and cost is well known to everyone, but it is used to it. RDSClaw’s database management is essentially a change of lock: “only people who can write SQL can enter” to “one word can enter, but every step of the operation has a safety fence”. It allows colleagues in the marketing department to get the data in 10 seconds, so that developers do not have to wait in line for DBA, so that DBA can release their energy from repetitive work to do more valuable architectural work. The prerequisite for all of this- security, from credential encryption to read /write control to secondary confirmation – is that RDSLaw ensures layers of security. You do not need to understand SQL, do not need to wait for scheduling, and do not need to worry about security. Tell me what you want and RDSClaw will help you.

随着IDC行业的快速发展，可持续发展将成为未来竞争的关键

如果您正在寻找优质的香港云服务器，欢迎访问 www.isclouder.com 了解更多