Lecture 243: Amazon EKS Architecture
- 课程：[NEW] AWS Certified Advanced Networking Specialty 2026
- 视频时长：03:47
- 来源：Udemy Lecture 35813104
- Slide 数量：3
- 画面切换：01:29、02:51
扫描说明
逐条读取了 Transcript 面板中的 94 条英文 cue，并按播放器进度同步全长约 03:47 的视频。依据 Managed Control Plane、EC2 Node Options、Fargate/Networking Responsibility 划分为 3 个小节。3 张截图来自真实 EKS Architecture Slide，统一为 1364×770。
Transcript 来自 Udemy 页面自带的英文自动字幕，没有使用 ASR。下面保留页面原文，包括其中的语法或识别错误。
1. EKS 托管跨三 AZ 的 Kubernetes Control Plane，Customer 负责 Data Plane
时间范围：00:00–01:29
总结 / Summary
中文
EKS 保留 Kubernetes Control Plane/Data Plane 架构，但重新划分责任。AWS 在多个 Availability Zone 中运行和维护 kube-apiserver、etcd 与 Controller 等 Control Plane Component，负责 High Availability、Replacement、Patching 与 Control Plane Operation；用户不直接管理 Master Node。Customer 仍决定 Data Plane 的 Compute Type、Capacity、Subnet/AZ Placement、Pod Scheduling Constraint 与 Application Lifecycle。Managed Control Plane 不等于整个 Cluster Fully Managed，Node、Add-on、Workload 与 Network Policy 仍有客户责任。
English
EKS preserves the Kubernetes control-plane and data-plane architecture but changes ownership. AWS runs and maintains kube-apiserver, etcd, controllers, and other control-plane components across multiple Availability Zones, handling high availability, replacement, patching, and control-plane operations; customers do not manage master nodes. Customers still choose data-plane compute, capacity, subnet and AZ placement, pod scheduling constraints, and application lifecycle. A managed control plane does not make the entire cluster fully managed; nodes, add-ons, workloads, and network policy retain customer responsibility.
Transcript
00:00 Hello and welcome to this lecture.
00:02 So in the earlier lecture,
00:03 we saw the components of Kubernetes cluster.
00:06 Essentially we saw the control plane, and data plane,
00:09 and what are the important services
00:11 which run inside the control plane, and data plane.
00:14 Now, when it comes to the EKS,
00:16 of course architecture remains the same,
00:18 but the responsibility
00:19 of control plane, and data plane, changes.
00:22 And as you might have guessed,
00:24 it makes more sense
00:25 that AWS takes care of the difficult thing,
00:29 that is managing the control plane.
00:31 And then you take care of the data plane,
00:33 where you decide which kind of nodes you want to run,
00:35 how many ports you want to run.
00:37 That is all part of your data plane.
00:39 So the architecture is similar to what we saw earlier.
00:43 As you can see here, there is a control plane on the top,
00:46 and it is across three availability zones, of course,
00:49 for high availability, and resiliency.
00:51 And this control plane is completely taken care of by AWS.
00:55 And below is the data plane.
00:57 And then you can decide
00:58 how many availability zones you want to use,
01:00 and then you can spin up those nodes
01:03 across multiple availability zones.
01:06 Now this is the Kubernetes architecture.
01:09 And now if you look at the EKS,
01:11 so you don't care about the control plane high availability.
01:14 You know that there are different components
01:17 like for example, APIs, server, SD, data store
01:20 cube controller manager, and all this
01:22 has been taken care of.
01:24 So with EKS, you get a managed control plane.
2. EC2 Data Plane 可选 Self-managed Node 或 Managed Node Group
时间范围：01:29–02:51
总结 / Summary
中文
Self-managed Node 由客户选择 AMI/Instance Type、创建 Auto Scaling Group、执行 Bootstrap 让 EC2 注册为 Kubernetes Node，并负责 Upgrade、Drain、Replacement 与 Scaling Workflow。EKS Managed Node Group 仍以客户 VPC 中的 EC2 提供 Capacity，但 AWS/EKS 自动化 Provisioning、Cluster Registration、Rolling Update、Drain/Termination 等 Node Lifecycle，显著降低运维工作。两者都属于 Node Group：一组承担相似 Role/Label/Taint/Capacity Profile 的 EC2 Worker。
English
With self-managed nodes, customers choose AMIs and instance types, create Auto Scaling groups, bootstrap EC2 instances into the cluster, and own upgrades, draining, replacement, and scaling workflows. EKS managed node groups still use EC2 capacity in the customer VPC, but EKS automates provisioning, cluster registration, rolling updates, draining, and termination, reducing operational work. Both are node groups: collections of EC2 workers sharing similar roles, labels, taints, and capacity profiles.
Transcript
01:29 Now, when it comes to the data plane,
01:31 there are multiple options
01:32 for the way you would launch your nodes.
01:35 Now these nodes are typically your EC2 machines,
01:37 but then Kubernetes
01:38 also offers you serverless compute options
01:41 in the form of AWS Fargate.
01:43 Now, when it comes to the EC2 options,
01:45 there are two ways
01:46 in which you can launch this EC2 machine.
01:49 One is self-managed nodes.
01:51 That means you launch this EC2 machine.
01:53 You can put those into the auto-scaling group,
01:55 but then you decide which AMIs to use,
01:58 which instance types to use.
02:00 Once these instances are launched,
02:02 then you want to make sure that the instance gets registered
02:05 into the Kubernetes cluster as a node.
02:07 And for that you need to run a few commands.
02:09 And if for some reason, the instance is unhealthy,
02:12 then you would want to take this out of your cluster.
02:14 So all this has to be managed by you,
02:16 if you're using self-managed nodes.
02:19 Now, EKS has simplified this operation,
02:21 in the form of managed node groups.
02:24 That means provisioning the nodes,
02:26 and then adding it to the Kubernetes cluster,
02:29 as well as draining those nodes,
02:31 in case you want to reduce the capacity,
02:33 or adding new nodes when you want to increase the capacity.
02:36 All this has been taken care of,
02:38 if you're using manage node groups.
02:41 Now, node groups are nothing but a collection
02:43 of EC2 machines, whether self-managed, or managed.
02:46 But then underlying compute capacity
02:48 is provided by EC2 instances.
3. Fargate 提供 Pod-level Serverless Compute，但所有 Data Plane 仍落在 VPC Network
时间范围：02:51–03:47
总结 / Summary
中文
AWS Fargate 让用户声明 Pod 需要的 CPU/Memory，由 AWS Provision 隔离的底层 Compute，不必管理 EC2 Worker Node；Pod 通过 Fargate Profile 选择 Namespace/Label 与目标 Subnet。它适合不希望操作 Node 的 Workload，但在 DaemonSet、Privileged/Host Feature、Storage、Add-on 与 Platform Support 上有不同约束。无论使用 EC2 还是 Fargate，Data Plane 都连接客户 VPC，因此 Node/Pod IP、Subnet Capacity、Security Group、Control-plane Reachability 与 Service Exposure 仍是 EKS Networking 的核心。
English
AWS Fargate lets users declare pod CPU and memory while AWS provisions isolated underlying compute, eliminating EC2 worker-node management. Fargate profiles select namespaces and labels and choose target subnets. It suits workloads that should avoid node operations but has different constraints around DaemonSets, privileged or host features, storage, add-ons, and platform support. Whether the data plane uses EC2 or Fargate, it connects to the customer VPC, so node and pod IPs, subnet capacity, security groups, control-plane reachability, and service exposure remain central EKS networking concerns.
Transcript
02:51 There is a third option, which is AWS Fargate.
02:54 Now, as you already know,
02:55 AWS Fargate is a serverless compute offering from AWS.
02:59 That means there are no machines to manage for you.
03:02 Now with respect to EKS,
03:05 you just define your pods, and container capacity,
03:08 that they need,
03:09 and then Fargate will provision
03:11 the underlying compute capacity
03:12 for running those containers,
03:14 and the pods.
03:15 So that's a very brief about EKS architecture,
03:17 very similar to the Kubernetes architecture.
03:20 And then now we will focus more on the networking part.
03:23 Now you know that AWS has VPC.
03:26 That means your cluster will be inside a VPC,
03:29 or will be launching this data plane nodes inside a VPC.
03:33 So it's interesting to see how these nodes communicate,
03:36 how these ports communicate,
03:37 how you expose these services outside of the cluster.
03:41 So all that will be covered in the next lecture.
03:44 So stay tuned, and thanks for watching.
整讲总结 / Overall Summary
中文
EKS 的价值是 AWS 托管高可用 Kubernetes Control Plane，而 Data Plane 由客户在 VPC 中选择 Self-managed EC2、Managed Node Group 或 Fargate。托管程度不同，但 Network Design 始终涉及 Subnet/AZ、IP Address、Security Group、Control Plane Connectivity 和对外 Service Exposure。
English
EKS has AWS manage a highly available Kubernetes control plane while customers choose a VPC data plane of self-managed EC2 nodes, managed node groups, or Fargate. Operational ownership differs across options, but network design always involves subnets and AZs, IP addresses, security groups, control-plane connectivity, and external service exposure.
返回顶部 ↑