1. EKS Account:

• The Managed Control Plane includes API Server instances and etcd instances for storing cluster data.
• The Managed Capabilities provide core services like storage, compute, and load balancing essential for Kubernetes operations.

• Encompasses user-managed AWS services such as Application Load Balancers, EBS Volumes, and Network Load Balancers.
• Includes EC2 Managed Instances for containerized workloads, offering users control over worker node configurations.

Relation to Auto Mode:

• Auto Mode leverages the managed capabilities within the EKS account while simplifying node management and scaling within the customer account.
• AWS automatically handles updates, right-sizing, and scalability, ensuring high availability without user intervention.

Relation to Fargate:

• Fargate eliminates the EC2 Managed Instances entirely by providing a serverless approach.
• Workloads are defined and run directly on Fargate, simplifying the architecture but introducing specific limitations like the lack of DaemonSets and GPU support.

1. Infrastructure Management:

• Requires provisioning and managing the control plane.
• Involves installing plug-ins and selecting appropriate compute resources.

• Enables deploying applications, monitoring, repairing, and upgrading clusters.
• Provides tools to analyze and optimize workloads effectively.

• Supports exposing pods through services like LoadBalancer and Ingress.
• Heavily relies on DaemonSets for background processes.
• Allows running stateful applications by leveraging EFS for persistent storage.

• Offers workload-specific EC2 instance selection, including GPU instances.
• Supports deployment in both public and private subnets.
• Allows HostPort/HostNetwork configurations and CNI custom networking.

• Entails control plane and EC2 worker node costs.
• Can incur idle capacity costs if not managed efficiently.

EKS Auto Mode: Simplifying Cluster Management

Key Features:

• Application-Ready Clusters: Automatically provisions the infrastructure with essential Kubernetes capabilities.
• Managed Add-ons: AWS takes care of installing and updating core Kubernetes add-ons, such as EBS drivers, ingress controllers, and CoreDNS. These add-ons scale with the control plane.
• Node Management:
• Automatically right-sizes EC2 worker nodes based on workload requirements.
• Supports the use of reserved and spot instances for cost efficiency.
• Updates worker nodes in a rolling deployment fashion every 15–21 days.
• Cost Optimization: Continuously monitors and optimizes instances to minimize costs.

Advantages:

• Reduces operational overhead by managing add-ons and control plane updates.
• Ensures scalability and high availability of worker nodes.
• Ideal for workloads requiring DaemonSets, public/private subnets, or custom networking configurations.

Costs:

• Control plane cost + worker EC2 instance cost (optimized).

EKS with Fargate: Serverless Kubernetes

Key Features:

• Pod-Centric Management:
• Deploy and define pods directly to Fargate.
• Allows mixing Fargate and EC2-based workloads within the same cluster.
• Serverless Simplicity: No worker nodes to manage, significantly reducing operational overhead.
• Scaling: Horizontal Pod Autoscaling (HPA) is supported for better scalability.

Limitations:

• Feature Constraints:
• DaemonSets are not supported; sidecars must be used as alternatives.
• Stateful applications are not recommended due to lack of GPU support and a maximum pod size of 4 vCPU/30GB memory.
• Networking:
• Only supports private subnets.
• HostPort/HostNetwork configurations and CNI custom networking are not available.
• Load Balancer Limitations:
• Classic Load Balancers (CLB) and Network Load Balancers (NLB) are unsupported; pods can only be exposed using Ingress.

Costs:

• Control plane cost + pay-per-pod pricing (based on vCPU, memory, and runtime).

Use Cases:

• Ideal for stateless applications.
• Suited for teams seeking to reduce management overhead while running containerized workloads.

Key Differences: EKS Auto Mode vs. Fargate

1. EKS Auto Mode: Uses managed EC2 instances.
2. Fargate: Serverless; no worker nodes required.

1. EKS Auto Mode: Automatically right-sizes EC2 instances.
2. Fargate: Scales pods directly via Horizontal Pod Autoscaler (HPA).

1. EKS Auto Mode: Supported.
2. Fargate: Unsupported; requires sidecar containers instead.

1. EKS Auto Mode: Supported (e.g., databases, persistent storage).
2. Fargate: Not recommended for stateful workloads.

1. EKS Auto Mode: Supports public and private subnets.
2. Fargate: Limited to private subnets.

1. EKS Auto Mode: Compatible with Classic (CLB), Network (NLB), and Ingress.
2. Fargate: Only Ingress-supported.

1. EKS Auto Mode: Supports Container Networking Interface (CNI).
2. Fargate: CNI not supported.

1. EKS Auto Mode: Pay for control plane + EC2 instance costs.
2. Fargate: Pay for control plane + per-pod runtime costs.

1. EKS Auto Mode: Moderate (AWS manages add-ons/updates; you manage nodes).
2. Fargate: Low (serverless; no node management).

1. EKS Auto Mode: Flexible for mixed workloads (stateful, batch, custom apps).
2. Fargate: Ideal for stateless, event-driven, or serverless applications.

Conclusion

• EKS Auto Mode is for more control over infrastructure, support for DaemonSets, stateful applications, and advanced networking.
• Fargate is used for its simplicity, serverless operation, and suitability for stateless workloads.