GPU Resource Management On JDOS 梁永清 liangyongqing1@jd.com 提供的服务 1. 用于实验的 GPU 容器 2.基于 Kubeflow 的机器学习训练服务 3.模型管理和模型 Serving 服务 Experiment Training Serving 均基于容器，不对业务方直接提供 GPU 物理机 GPU 实验 JDOS 常规的容器服务

Node Operator: Kubernetes Node Management Made Simple 陈俊(Joe), Ant Financial Agenda • Background and Motivation • Introduction of Operators • Node-Operator • Advanced Topic: • Upgrade Master & Node Components reliably • Canary Rollout • Master & Node Component Versions Management Motivation: Work Order Deployment Worker Order • Upgrade Nodes Versions • Upgrade Node 10.10 • Non-failure-aware • Complicated architecture Work order deployment system can not meet the requirements of resource management. Operator Observe Action Analyze • Observe: watch desired resource

placement options, for both control plane and worker nodes. 2 levels of scheduling and resource management are active. ​Currently no automatic scheduling integration occurs, that is, Kubernetes is not to solve potential issues with CPU and memory intensive workloads ​Kubernetes default resource management ​How it works ​Extending the functionality of Kubernetes ​Using vSphere DRS with Kubernetes Memory intensive workloads Nearly all database servers (e.g. Oracle, MongoDB), present a workload which will attempt to detect and consume as much of the system’s memory as possible. Where does this

placement options, for both control plane and worker nodes. 2 levels of scheduling and resource management are active. Currently no automatic scheduling integration occurs, that is, Kubernetes is not to solve potential issues with CPU and memory intensive workloads Kubernetes default resource management How it works Extending the functionality of Kubernetes Using vSphere DRS with Kubernetes High Memory intensive workloads Nearly all database servers (e.g. Oracle, MongoDB), present a workload which will attempt to detect and consume as much of the system’s memory as possible. Where does this

Dashboard 执⾏： kubectl proxy 02-安装单机版Kubernetes 8 访问： http://localhost:8001/api/v1/namespaces/kube-system/services/https:kubernetes-dashboard:/proxy/#!/overview? namespace=default 参考： https://kubernetes addons（插件） Addon是实现集群功能的Pod和Service。Pod可由Deployment、ReplicationController等进⾏管理。Namespace的插件 对象则是在 kube-system 这个namespace中被创建的。 Addon manager创建并维护addon的资源。详⻅这⾥： here 。 DNS 虽然其他Addon不是严格要求的，但所有Kubernetes集群都应该有 STATUS AGE default Active 1d kube-system Active 1d Kubernetes初始有两个Namespace： default ：对于没有其他Namespace的对象的默认Namespace kube-system ：由Kubernetes系统所创建的对象的Namespace 为请求设置Namespace

2010 2015- Now 2010- Now 2012- Now Bare metals Way to Kubernetes Search Grid Hadoop PoP Database Frontend VM Kubernetes plays magic api etcd kind: metadata: spec: control loop control loop Kubernetes You have your compute node now, all you need is to configure it by a configuration management orchestration. We use SaltStack. Let’s model a datacenter running Kubernetes Onboard Provision KafkaCluster, HadoopCluster, MongoDB, ESCluster …… Fleet (Compute, Network, Storage) Configuration Management Infrastructure Service Application Service Recap We are hiring! xnxin@ebay.com cmei@ebay.com

cgroupdriver=systemd" ] } # mkdir -p /etc/systemd/system/docker.service.d # docker info ★docker会修改防火墙规则，导致pod网络不通 # vi /usr/lib/systemd/system/docker.service #在[Service]下的ExecStart=/usr/bin/dockerd 置文件并编辑 # vi /etc/kubeadm-init.yaml apiVersion: kubeadm.k8s.io/v1beta2 bootstrapTokens: - groups: - system:bootstrappers:kubeadm:default-node-token token: abcdef.0123456789abcdef �l: 24h0m0s usages: 置文件并编辑 # vi /etc/kubeadm-init.yaml apiVersion: kubeadm.k8s.io/v1beta3 bootstrapTokens: - groups: - system:bootstrappers:kubeadm:default-node-token token: abcdef.0123456789abcdef �l: 24h0m0s usages:

Scalable Kubernetes Applications • Scalable Infrastructure for Applications Application Operating System Physical Infrastructure Platform Containers as Enabler Fast Boot Environments Rapidly Portable Needed Application Operating System Physical Infrastructure Containers and VMs - A Practical Comparison Containers Containers virtualize the operating system limiting the the number of application applications on the same OS Allows you to run multiple OS on the same hardware Application Operating System Physical Infrastructure Containers VMware Hypervisor VMs Docker Containers User Cases 9

li@coreos.com | Head of distributed system Self driving infrastructure Topics ● Cluster management systems ● Today’s problems with operating cluster management systems ● A self-driving approach components ○ dynamic dependencies ○ fast deployment iteration ● Solution: automation Cluster management system ● Automation ○ Scheduling ○ Deployment ○ Healing ○ Discovery/load balancing ○ Scaling Scheduling Kubernetes? ● Operational expertise around app management in k8s extends to k8s itself ○ E.g. scaling ● Bootstrapping simplified ● Simply cluster life cycle management ○ E.g. updates ● Upstream improvements

Accessible by users who shouldn’t have access, e.g., CEO ○ Stored in public storage buckets Secret management requirements Identity Require strong identities and least privilege Auditing Verify the use Separate where secrets are used vs managed Encryption at different layers (or turtles) disks file system etcd Recommendation: Use two-layers of encryption, e.g., full-disk & application-layer … then security against penetration. Similarly, poor key management may easily compromise strong algorithms.” NIST SP 800-57, Recommendation for Key Management Keys get old Key rotation ● Key rotation is meant