INFRASTRUCTURE LEARNINGINFRASTRUCTURE LEARNING
インフラ大全インフラ大全IT INFRASTRUCTURE / LEARNING
Infrastructure learning guide
Find a lesson or lab/

Esc to close · Search terms stay on your device

LEARNING日本語

Topic lessonTopic: Kubernetes and EKS: structure, networking and upgrades

Amazon EKS explained: Kubernetes nodes, Pods and management responsibilities

EKS is an AWS service for using Kubernetes. A Pod is an application deployment unit, a node provides compute for Pods, and the control plane coordinates placement and state. Choosing EKS still leaves application configuration, data and required permissions to decide. Read one report API through separate views of structure and responsibility.

What you will learn

Distinguish the control plane, nodes, Pods and Deployments; explain differences from ECS terminology and the management and cost questions for each compute approach.

Prerequisite

An image defines what to run; a container is an execution using it. No AWS or Kubernetes account is needed. Diagrams and counts are fixed synthetic examples, not observations of a running cluster.

Separate coordination from application execution

Kubernetes manages the placement, replication and updates of containerized applications. A cluster combines the control plane that coordinates them with nodes that run Pods. In EKS, distinguish AWS-managed components from the parts the customer designs and verifies.

Identify four terms by their job
TermJobDo not confuse it with
Control planeAccepts API requests and coordinates control toward the desired state and placement.Not an HTTP proxy that every application request must traverse.
NodeCompute where Pods run; an EC2 instance is a node in an ordinary EC2-backed configuration.Not necessarily one application per node.
PodOne or more containers placed together as a unit.Not an image or the desired replica count itself.
DeploymentDeclares a Pod template and replicas, maintained and updated through controllers.Not proof that every Pod has become Ready.

Controllers reconcile observed state with desired state; the scheduler selects a suitable node for an unscheduled Pod. A Deployment declares the Pod template and replica target. These control functions differ from the application serving a user request.

A declared state differs from an observed state. Three desired replicas do not establish successful placement or readiness: resources, image access, permissions and application startup still matter. Ready is one availability signal, not proof that every business operation succeeds.

Fixed example: replicate one API across three Pods

This is a teaching configuration using EC2-backed nodes. A Deployment requests three report-api Pods. Each Pod template specifies an API container and one helper: two containers per Pod. The example explains roles and counts; it does not recommend a node count, co-location pattern or helper container.

Structural references: lines show desired state and placement, not the user HTTP path
Structural references: lines show desired state and placement, not the user HTTP path1: Deployment: 3 replicas — Declare desired state — EKS control plane. 2: EKS control plane — Controllers and scheduler reconcile — Three Pods. 3: Three Pods — Observe actual placement separately — Node (EC2-backed)123EKS control planeDeployment: 3replicasThree PodsNode (EC2-backed)Structural references: lines show desired state and placement, not the user HTTP path1: Deployment: 3 replicas — Declare desired state — EKS control plane. 2: EKS control plane — Controllers and scheduler reconcile — Three Pods. 3: Three Pods — Observe actual placement separately — Node (EC2-backed)123EKS controlplaneDeployment: 3replicasThree PodsNode (EC2-backed)
  1. Deployment: 3 replicas → EKS control planeDeclare desired state
  2. EKS control plane → Three PodsControllers and scheduler reconcile
  3. Three Pods → Node (EC2-backed)Observe actual placement separately
EKS control plane
API, control and scheduling
Deployment: 3 replicas
report-api Pod template
Three Pods
Each: API + helper container
Node (EC2-backed)
Placement concept; actual placement uncollected
Count template, objects and readiness separately
ValueFixed exampleWhat it establishes
Desired Deployment replicas3A Pod target, not a node count.
Container definitions per PodOne API + one helperIf all three Pods start, this example has six application containers.
Observed PodsTwo Ready; one PendingNot all three Pods are Ready at this observation.
Node count and placementNot collected in this tablePod count alone does not determine node count or AZ distribution.

Do not infer CPU shortage or an application failure from Pending alone. Inspect events, placement constraints, resource requests, image retrieval and the actual reason. The diagram does not supply unobserved placement success or availability.

Management scope also depends on the compute approach

AWS management and customer decisions
ApproachExamples managed by AWSExamples the customer checks
EKS standard + managed node groupsControl plane, with node-group provisioning and update capabilities.Plan and initiate applicable updates; manage workloads, data, access, placement and compatibility.
EKS Auto ModeControl plane plus broader automation of compute and related infrastructure.Supported features and configuration; workloads, data, access and costs.
EKS + FargateUnderlying infrastructure for eligible Pods, without directly managing EC2 nodes.Pod eligibility and networking, storage and feature constraints.
EKS + self-managed nodesControl plane.Broader node maintenance, updates and capacity responsibilities.

More managed infrastructure does not remove application design. Decide what data to store, who may do what, which connections the application needs and how to validate updates. If compute approaches are mixed, identify the infrastructure used by each workload.

Check ECS terminology and separate cost layers

Related roles do not mean identical configuration or APIs
ECS termKubernetes concept hereCaution
Task definition / taskPod template / PodA useful comparison of definition and execution, not identical schemas or lifecycle.
ECS serviceA workload controller such as DeploymentCompare maintaining replicas. A Kubernetes Service serves a different role.
Connectivity with an ALB and other resourcesService, Ingress/controllers and related resourcesInspect references and network conditions; matching names is not a migration plan.

Review costs separately for cluster management, compute running Pods, and storage, networking and related capabilities. Three Pods alone do not determine a bill. Match the chosen approach, resources, duration, Region, support conditions and surrounding resources to current pricing and the actual configuration. This lesson promises neither a price nor free operation.

Read roles, desired state, observations and ownership

  1. Separate control plane, node, Pod and Deployment roles.
  2. Read the Pod template and replica target; do not substitute container or node counts.
  3. Inspect observed state, readiness, placement and events; retain unknowns.
  4. Identify update, access, network, data and cost responsibilities for the chosen infrastructure.

After reading roles and counts, examine how a Service selects destinations, what an Ingress/controller configures and which traffic a NetworkPolicy restricts. Control-plane health and a successful user operation remain separate observations.

Distinguish nodes, Pods and responsibilities

Choose an option to see the result and explanation.

Three Pods each define one API and one helper. If all start, how many application containers are in this example?
Read the answer and reason

Answer: Six; the node count is not determined · Three times two is six. Placement and node count require separate information.

Three desired replicas; two Pods Ready and one Pending. What is established?
Read the answer and reason

Answer: Not all Pods are Ready; investigate the Pending reason · The observation establishes the state, while events and other evidence are needed for the cause.

Even with EKS Auto Mode, what does the customer still decide?
Read the answer and reason

Answer: Application data, required access and configuration conditions · Broader infrastructure management differs from designing the workload.

When comparing the replica-maintenance role of an ECS service, which distinction matters?
Read the answer and reason

Answer: Compare workload controllers such as Deployment; distinguish Kubernetes Service · The comparison explains roles, not identical APIs or configuration.

Can three Pods alone determine the total EKS bill?
Read the answer and reason

Answer: No; inspect compute, duration, management and surrounding resources · Counts alone are insufficient; compare the chosen approach and usage conditions with pricing.

Check, then explain in your own words

  1. Explain the three-Pod, two-containers-per-Pod example.
  2. Read three desired, two Ready and one Pending without inventing a cause.
  3. Identify the compute approach and decisions retained by the customer.
Check your understanding: After requesting three replicas in EKS, what should you verify separately?

The Pod template and container count, observed Pod state and readiness, node placement, and the chosen approach’s management and cost conditions.

A desired replica count establishes neither execution success nor node count. Healthy managed infrastructure alone does not prove all workloads or user operations succeed.

Sources and conditions checked

The official references used in this lesson are collected here with their scope, conditions and corresponding sections. Independent expert review and live AWS comparison have not been performed.