secrets Checking application health Replicating application instances Using Horizontal Pod Autoscaling Naming and discovering Balancing loads Rolling updates Monitoring resources Accessing and Autoscalers (HPA) 的⽬标。也就是说，ReplicaSet可以由HPA⾃动伸缩。以下是 ⼀个HPA指向我们在上⼀个示例中创建的ReplicaSet的示例。 apiVersion: autoscaling/v1 kind: HorizontalPodAutoscaler metadata: name: frontend-scaler spec: scaleTargetRef: nginx-deployment --replicas=10 deployment "nginx-deployment" scaled 假设您的群集启⽤了 horizontal pod autoscaling 功能，您可以为Deployment设置⼀个autoscaler，并根据现有Pod的 CPU利⽤率，选择要运⾏的Pod的最⼩和最⼤个数。 $ kubectl autoscale deployment

management to optimize resource orchestration efficiency Ø High performance and comprehensive autoscaling What is TKEx Ø Based on TKE (Tencent Kubernetes Engine) & EKS (Tencent Elastic Kubernetes Service) of online and offline services. • Support Service Mesh. • Large-scale and high-performance autoscaling capabilities. • Multi-tenant and quota management. • etc. TKEx Architecture EKS (Elastic Kubernetes ValidatingWebhook to validate pod add request. DynamicQuota Large-scale and high-performance Autoscaling Motivation: Ø Reduce container OOM from bursted traffic. Ø Improve cluster resource utilization

operator design • Pipeline / Stage/ Task / Task Template / Version Control • Logging, monitoring, autoscaling, high availability • Extensibility / Integration • CI/CD examples • Future plan Our DevOps Pipeline/Stage/Task/Task Template/Version Control/UI generation/Volume... • Logging, monitoring, autoscaling, high availability • Extensibility/Integration • CI/CD examples • Future plan Overall Architecture created from template Tasks in same stage can run sequentially or in parallel Logging, Monitoring, Autoscaling, High Availability Kubernetes Cluster Node Node Node Node Job Job Job Job Pod Pod Pod ElasticSearch

GPU or CPU nodes • Massive Scale • OpenAI dedicates up to 10k cores for a single experiment • Autoscaling capabilities: Pay for what you use, scale down when idle • Parallel training instead of sequential: Spin up pods for each variation of hyperparameters • One centralized TensorBoard instance • Autoscaling will create / remove VMs as needed to save cost Demo: Create End to End ML Pipelines with Argo Distributed File Systems • NFS • HDFS • … Classic DevOps solutions: • Containers • CI/CD • Autoscaling • A/B testing and canary release of Models • Comparing Production accuracy vs expected accuracy

Flagger Virtual Service Manual Scaling App CRD HPA Knative Service Cert Canary AutoScaler AutoScaling Route Job Deployment 缺乏交互、复用、可移植能 力。不同重复造轮子只是适 配不同 API 如何基于 K8s ，构建出一个既用户友好，又高可扩展，还 统一、标准化的应用管理平台？ truth) 持续集成 ● Build ● Run Unit Tests ● Build Docker Image ● Push Docker Image Image Registry AutoScaling Controller Rollout Controller GitOps OAM K8s Plugin + CUE Abstraction Processor Kubernetes

分批策略 访问控制 流量配置 Deployment Function - componentName: frontend traits: - trait: apiVersion: autoscaling/v2beta2 kind: HorizontalPodAutoscaler spec: minReplicas: 1 maxReplicas: 10 - trait: apiVersion: spec: components: # 1st component - componentName: frontend traits: - trait: apiVersion: autoscaling/v2beta2 kind: HorizontalPodAutoscaler spec: minReplicas: 1 maxReplicas: 10 - trait: apiVersion:

cpu: 100m # memory: 128Mi # requests: # cpu: 100m # memory: 128Mi autoscaling: enabled: false minReplicas: 1 maxReplicas: 100 targetCPUUtilizationPercentage: cpu: 100m # memory: 128Mi # requests: # cpu: 100m # memory: 128Mi autoscaling: enabled: false minReplicas: 1 maxReplicas: 100 targetCPUUtilizationPercentage:

orchestration, it has been expanded to support for data-intensive and stateful apps. Benefit: l Autoscaling in Cloud l Consolidate online service and offline analysis l Ecosystem( Monitor, logging etc)

containerPort: 80 • kubectl create -f deployment.yaml • kubectl create -f hpa.yaml apiVersion: autoscaling/v2alpha1 kind: HorizontalPodAutoscaler metadata: name: php-apache namespace: default spec: