YARN是一个集群资源调度框架,其思想是将资源管理和作业调度/监视的功能拆分为单独的守护程序。对于应用开发者来说，几乎不需要自定义YARN本身的程序，对于管理员来说，还需要熟悉如何查看日志、如何配置队列、用户ACL等操作，这些工作实操性比较强。

我在Linux服务器上安装了MIT Kerberos server，然后在Cloudera Manager上通过Web UI向导进行启用Kerberos的初始化。过程执行失败了，刷新CLoudera Manager的首页后发现许多Configuration Issues，提示说各个服务的Kerberos Credentials是missing的状态。

MapReduce既是编程模型又是计算框架，尽管对于我们来说只需要实现map和reduce函数就能完成业务，但是前提是要对MapReduce编程框架的内部机制有清晰的概念。

MapReduce既是一个编程模型，又是一个计算框架。由于我们的数据是存放在分布式文件系统中，自然不能用传统的编程模型来完成任务了。

HDFS Architecture and Explaination of Design

CloudFormation 是 AWS 自家的云服务编排工具，本文介绍了如何使用官方 CloudFormation 模板部署AWS EKS。

本文介绍了如何在Kubernetes中部署Traefik 2.0以及如何配置路由。

By default, Rook Ceph is dynamically set to `fileStore` if a data dir is set and created on the host. This article instructs how to use `blueStore` as `storeType` in Rook Ceph.

Rook Ceph performance testing. Contrast of EC block and Replicated block.

In Kubernetes, StatefulSet is designed to address the topological and storage state of the application. And we usually need to use storage services to provide to StatefulSet to use. Ceph is a popular open source distributed storage solution. Rook can help us deploy and use Ceph in k8s.

Our software technology stack is mainly Java with micro-service frameworks(e.g. Spring Boot, Consul). And we're moving from VM to Kubernetes. Considering cases of putting consumer or/and registry conponents outside of the Kubernetes cluster, we have to make our network plane allows workload accessing Pod IP, Service IP from outside the Kubernetes cluster.

Our on-premise environment consists of several VLAN(layer 2 network), each VLAN is created by VCenter. Physically, every VLAN is based on several bare metal machines on the same floor, the bare metal machines are connected by ethernet switches. And we have several floors of bare metal machines, they're connected by ToR(Top of Rack) routers(layer 3 network).