Posts

Showing posts from 2022

Toward Fairer Data-Driven Performance Management

the most common approach to quantifying someone’s job performance is to rely on subjective ratings, whether by the employee (self-rating of performance) or their manager (supervisory ratings). The typical correlation between self-ratings and supervisory ratings of job performance is merely 0.22, which translates to a trivial 4% overlap between the two. In other words, 96% of the variability in employees’ self-rated job performance is unrelated to how their managers’ view their performance. While employees are generally  too generous  in their self-evaluations of performance, there is not much evidence for the superior accuracy of supervisory ratings in measuring workers’  true  contributed value or output, though aggregating ratings of different managers or sources, including peers, will significantly  boost reliability . Needless to say, it is not just possible, but also desirable, to improve how others see us through factors unrelated to our actual job performance the KPI definition

How to handle to bad reviews

 https://hbr.org/2014/10/what-to-do-after-a-bad-performance-review Reflect before you react  it’s important to “hold your emotions in check,” Look for your blind spots try to seek out those who will be candid with you instead of telling you that the input isn’t true. “making it clear you are interested in honesty, not consolation,”  Ask questions Be careful with your tone: You don’t want to appear as if you’re challenging the review.”   Make it clear you want concrete examples of what you should be doing differently. Make a performance plan The purpose of feedback is to help you improve in your job, and that requires a detailed plan of action. That may involve learning new skills, reprioritizing your tasks, or reevaluating how you come across to colleagues. Agree with your manager on what you need to do to make changes. “Give yourself thirty days or sixty days to experiment with trying to do a couple of things differently,” Give yourself a second score “You could get an F on the exam,

A list of interesting technologies to explore

https://snyk.io/                   Container                    Security sonacube,                             CICD                          Security blackduck                              CICD                     Security  Config map                         K8S                              Manifest RediSearch                            Redis                          RedisJson                               Redis tmux                                        Linux Secreat Management           Security Monokle                                  K8S                              Manifest Helm+Kustomize                 K8S                       Manifest  (Helm for packaging and kustomize for overlay) Grafana as service               Monitoring FinOps                                 FinOps FinOps                                     FinOps PYNQ -- Python for FPGA  Hardware               Python ngrok test                                 Test Automation     Python

Attrition vs Turnover vs Voluntary Turnover

 Attrition:   人员净流失, 就是人走了但是位置没有补 Turnover: 离职的人的总数。不管什么原因走的都算。  Voluntary Turnover:自愿离职的人的总数,辞职,退休都算。  基本 Turnover - Voluntary Turnover = Attrition

This person prioritizes their workload effectively and meets deadlines.

This person prioritizes their workload effectively and meets deadlines. This person communicates clearly and effectively with me and other colleagues. This person exhibits strong leadership skills. This person has strong interpersonal skills and helps everyone feel welcome on the team. This person is always timely and efficient at providing feedback. This person prioritizes teamwork above all else. This person always finds creative solutions and takes initiative when problem-solving. This employee is always open to receiving both negative and positive feedback. This person strongly embodies our  company values . This person values diverse perspectives, even if they are different from their own. Open-Ended 360-Degree Review Sample Questions In order to elicit the qualitative information you’ll want to gather, it's important to include a number of questions that touch on an employee's strengths, as well as their areas for improvement. This can help managers ensure that each 360-d

【读书笔记】 管理其实很简单

https://36kr.com/p/1841625612690692 不要形式大于内容 分清 领导者 vs 管理者 方向: 确保团队可以交付,确保个人可以成功 指导 职业生涯: 目标 ->  愿景 ->  okr ->  优先事项 管理者 自我评价 和 现实 的差距 汇报3个优先事项 你当下的优先事项和季度目标的联系 提供知道,先设定目标,在提供指导。 提供严厉反馈。 做好准备工作。不管是表扬还是批评。  一定要具体。要有具体案例。  如何有效获取反馈 如果要你在工作中取得更大成就,我需要做哪些改变。 你希望在我身上看到或者学到什么,或者不希望看到什么 今天我又那些地方做的不对,可以做的更好 保持倾听,不要反驳 职业生涯规划。以及长期规划 从过去开始,生活中的故事开始,兴趣,爱好。  在谈论未来,梦想。  具体下一步怎么做

Performance Talk

Image
 

[读书笔记] 容器进程调度时是该优先考虑CPU资源还是内存资源?

  https://mp.weixin.qq.com/s/8dXxQi4QOrYLSSEyItSlNA CPU 资源是分时来调度的,每个进程都会得到一些时间片进行执行。所以 A 机器上不管 CPU 有多忙,再加一个的进程来运行话其实影响无非就是所有的进程都运行的更慢了一些。再换个说法,就是 CPU 资源是可以超卖的,是属于可压缩资源。 当机器物理内存不是很充足的时候,如果申请的内存过大,操作系统就可能会挑选在运行的一些进程将其杀掉。

【读书笔记】编码用这 16 个命名规则能让你少写一半以上的注释!

https://xie.infoq.cn/article/76e43e489d1898ce4500da5b8 保持命名形式的一致性 避免缩写 将描述事物最准确的名词放在最后 让代码像语句一样连贯 对于非布尔类型的属性或变量使用名词 对于代表可执行某类操作布尔属性,优先使用能动词 对于布尔类型的参数,不要使用动词 对于布尔型属性和变量,应该使用正向名称 使用祈使动词来命名函数或那些产生其他效果的方法 对于主要目的是返回一个值的函数优先使用名词或非祈使动词 不要在方法名称开头加 get 如果方法是将一个对象复制成为另一种表现形式,那么使用 toX() 形式 如果返回的是原对象的不同的表现形式,那么使用 asX() 形式 不要在方法名称上重复参数名 对于泛型,遵循通用的助记符

Setup Strikesthrough shortcut in Outlook

 https://techwelkin.com/strikethrough-shortcut-key-for-ms-word-excel-outlook Ctrl+D,   Ctrl+Alt+('+' in numpad)

在bash里并行运行命令

要求: 一个job,在server上同时运行几个process,只要有一个process失败 ,整个job就标记成失败。 实现: command_api &  process_api=$! command_worker &  process_worker=$! command_hworker &  process_hworker=$! wait $process_worker exit1=$? wait $process_api exit2=$? wait $process_hworker exit3=$? exit $(($exit1+$exit2+$exit3)) 参考 https://blog.csdn.net/fengfengzai0101/article/details/104047419

Strategy is not a plan

https://youtu.be/iuYlGRnC7J8 Strategy -->  from Why to how,  must be a coherent inside.  Plan doen't have this kind of coherent. 

How to deal with passive-agressive coworker

https://youtu.be/-W4_I4nX2uI *  Don't label the feeling, and behavior,  * Try to focus what the person really try to say.  give a chance to let them to say directly ,  not focus on their behaviour * Do not take bait * Get the support from the team.  

Database Log comparision

  Oracle MySql     Sql Server  Explaination Redo Log      transaction log / Redo Log             Transaction log                  all changes that happened to your data. Alert Log      error log find various messages about your database Archive Log         Binary Log Archived REDO Log          

[读书笔记] The 16-point checklist for GitOps success

https://www.cncf.io/blog/2022/07/08/the-16-point-checklist-for-gitops-success/   We have adopted Kubernetes for container and infrastructure management :  -- Done  We have documented a clear workflow between Application Development teams and the Platform team :  -- Not ready We have trained teams on the new workflows and tooling :   --  Not ready We have identified which changes can be automatically deployed to production, and which require a manual pull request :   --  Not ready We have declared everything in Git (this includes applications, infrastructure, networking, and configuration) :   --  Part Ready We have decided on an initial structure for our Git repositories   --   Part Ready We have selected the appropriate tooling that makes up our GitOps pipeline (Flux, Helm, Flagger, etc) :   --  Done  We have connected GitOps toolings like Flux, Helm, and Kustomize to our Git repositories :   --   Part Ready We have configured Git webhook for build triggers :    --   Part Ready We ha

[读书笔记] 上云 IT 成本治理之路

https://mp.weixin.qq.com/s/FUQmLZmj35bPu_ApmfY3mw 拆分系统 ->  全链路压力测试,评估反馈 ->   搭建模型,定义资源浪费标准 ->  分时混部在线业务与临时业务错峰使用 ->  使用弹性伸缩

【读书笔记】服务器 SMP、NUMA、MPP 三大体系结构

Image
https://mp.weixin.qq.com/s/WCIV5gCw2YkU2hi1m-bQaw SMP 主要特征是共享,系统中所有资源 (CPU 、内存、 I/O 等 ) 都是共享的。也正是由于这种特征,导致了SMP 服务器的主要问题,那就是它的扩展能力非常有限。 而最受限制的则是内存。由于每个 CPU 必须通过相同的内存总线访问相同的内存资源,因此随着 CPU 数量的增加,内存访问冲突将迅速增加,最终会造成 CPU 资源的浪费,使 CPU 性能的有效性大大降低。实验证明, SMP 服务器 CPU 利用率最好的情况是 2 至 4 个 CPU 。 NUMA 每个 CPU 可以访问整个系统的内存  访问本地内存的速度将远远高于访问远地内存  由于访问远地内存的延时远远超过本地内存,因此当 CPU 数量增加时,系统性能无法线性增加。 SWAP问题,即:NUMA0内存已经用完都开始用SWAP空间了,NUMA1还有很大的内存free。 MPP(Massive Parallel Processing) 由多个 SMP 服务器 ( 每个 SMP 服务器称节点 ) 通过节点互联网络连接而成,每个节点只访问自己的本地资源 ( 内存、存储等 ) ,是一种完全无共享 (Share Nothing) 结构,因而扩展能力最好,理论上其扩展无限制 MPP 服务器需要一种复杂的机制来调度和平衡各个节点的负载和并行处理过程

[读书笔记] Gartner:2022—2023年八大网络安全趋势预测

  https://mp.weixin.qq.com/s/kguBo1LJ00PVtTHrAma5YA Zero Trust Security Service Edge --   单一厂商的SSE平台访问网络、云服务和专用应用的策略。 公司和第三方进行商业业务的时候,必须要把供应商的安全能力作为一项决定因素

[读书笔记]:分布式一致性算法协议 Paxos

Image
 https://mp.weixin.qq.com/s/9vH8y-P7nU-kBGteUVxGiA 目的 Paxos算法是基于 消息传递 且具有 高度容错特性 的 一致性算法 Paxos算法需要解决的问题就是 如何在一个可能发生上述异常的分布式系统中,快速且正确地在集群内部对 某个数据的值 达成 一致 ,并且保证不论发生以上任何异常,都不会破坏整个系统的一致性。 概念 Proposer  ,  Acceptor,    Learners     Proposal

[读书笔记] 我在Uber创立SRE团队的故事

https://mp.weixin.qq.com/s/vgIb0anRke3FZ8b0JRpulg\ 一个描述我们为其他团队所做工作的YAML文件,以及一个将该文件编译成UI的Python Flask应用程序,该UI在可能的情况下自动执行这些请求,并生成包含完成任务所需信息的结构良好的票据, 包括完成任务的必要信息。 们把重点放在了我们剩下的最好的工具,即自动化。 由于缺乏监督或标准化,我们也有过既好笑又可怕的招聘经历。 在新领导加入并重新实施他们之前的设置后,通常情况下,关于事情如何运作的重要背景会被掩盖。我也没有很好地解释我们在优步的SRE组织中解决的具体问题。直到今天,我认为这些新领导人对优步的SRE组织正在解决的问题(使自下而上的组织具有高度流动性的优先权,而不是分解成一个单一的优先事项列表)有一个根本性的误解,即从他们以前的经验的角度来看待它。因此,SRE很快从嵌入式模式转变为完全不同的模式。 名字没变,但很快就变成了一个截然不同的团队。这个名字的延续掩盖了一个重要的领导和文化转变的开始,

[读书笔记] 管理宠物到管理牛群,DevOps 场景下效率难题如何解决

Image
https://www.infoq.cn/article/jTJGTtu2AgX74GkGif8Y 整个 CI/CD 过程就是一个不断将可变变为不可变的过程。 每一步都是要锁定一部分内容,做到不可变。有了不可变,才能随时得到不同资源组合的稳定版本,然后才能重复部署,或者安全及准确的回退到之前的版本。 组织团队之间的通讯和交流的模式,最终会映射成软件系统或者产品的形态。 独立部署是微服务的精髓,服务部署时还要依赖其他团队,本质上就不是微服务了 ( 不过是一个很昂贵的解决方案 ) 两个 container 可以共享一个存储,叫做 volume;通过 Init container 把数据准备到这个共享存储空间的路径上,Main container 就可以在这个存储上加载数据。这样一来,Main container 就会与数据加载以及初始化的方式彻底解耦。 如果在 pod 中包含多个 container,当我们需要某个 container 先完成启动就绪,就绪完成后才继续下一个 container 的创建,那么就可以在前一个 container 中加一个 postStart。这个 postStart 过程只是检查自己是否就绪,如果一直没有就绪,退出后返回不正常时整个 Pod 会失败,无法创建成功。如果创建成功,意味着已经就绪,就可以继续启动下一个 container。 当一个 Node 上有很多 Pod 在运行,可能会出现 Node 硬件不够用的情况,那么哪些 Pod 会被杀死?哪些 Pod 会被保留下来?这个叫 QoS。 Affinity 亲和性和 Taint 这两个概念有很多人都搞不清楚。所谓亲和性是从 Pod 的视角去看,什么样的 Node 适合我来运行。Taint 是从 Node 的角度来看自己可以运行什么样的 Pod,兼容了 node 上的 taint 的 pod 才可以被调度到 node 上。 引入了容器之后,带来的改变绝对不仅仅是构建、部署和维护的方式这么简单。其实每一次软件设计的变更都涉及到新增的构建块,带来新的设计思考。

[读书笔记] 5 ways to build the team

 https://leadershipfreak.blog/2022/06/03/5-ways-to-build-the-team-you-want/ " Brings up tough issues; has conflict and resolves it"  " If you want to build the team,  dedicate time to building . You can’t build a team when you’re buried in day-to-day execution."  "#4. Stop helping competent people. Overhelpfulness creates weakness in others and frustration for you."

领导矩阵管理

Image
 文章一 Tweet Post Share Save Buy Copies Print What if you upgraded your IT network to a new operating system that promised faster, better and leaner results — only to find that the vast majority of your leaders did not have the necessary competencies to use that new system? At Hay Group,  our research  suggests that many companies are now stuck in that metaphorical boat. Determined to get closer to their customers and become more agile and innovative, global organizations like IBM and GE are embracing  the matrix organizational model . But, the transition to this new model has not been all smooth sailing. In matrix organizations, leaders suddenly find themselves having to master the challenges of managing cross-divisional, international teams over whom they have little formal authority. Not surprisingly, the skills required to effectively navigate the matrix are different than those needed to succeed in the old, hierarchical organizational model. Leaders who lack these skills often find