This page is generated from skills/eks-best-practices/SKILL.md. Edit the source, not this page.
EKS Best Practices
Comprehensive guidance for designing, deploying, and operating Amazon EKS clusters. Consolidates guidance from the AWS EKS Best Practices Guide, AWS EKS HA/Resiliency Guide, and terraform-aws-modules/terraform-aws-eks examples.
When to Use This Skill
Activate this skill when:
- Designing a new EKS cluster architecture
- Choosing between EKS compute options (Fargate, MNG, Karpenter, Auto Mode)
- Configuring EKS networking (VPC CNI, ingress, service mesh)
- Implementing EKS security (IAM, pod security, secrets)
- Planning cluster upgrades or migrations
- Reviewing EKS architecture decisions
- Working with terraform-aws-modules/terraform-aws-eks examples
- Optimizing EKS cost or scaling to large clusters
Don't use this skill for:
- Generic Kubernetes concepts (Claude knows these)
- Provider-specific API reference (link to AWS docs)
- Non-EKS container orchestration (ECS, Lambda)
- Step-by-step EKS upgrade execution — this skill covers upgrade strategy and architectural decisions, not the per-version procedures themselves.
EKS Architecture Decision Framework
When to Use EKS
| Requirement | EKS | ECS | Lambda |
|---|---|---|---|
| Kubernetes ecosystem | ✅ Native K8s | ❌ AWS-proprietary | ❌ |
| Portable across clouds | ✅ Standard K8s API | ❌ AWS-only | ❌ AWS-only |
| Long-running services | ✅ | ✅ | ⚠️ 15 min limit |
| Minimal ops overhead | Medium | Low | Lowest |
| GPU/ML workloads | ✅ Best support | Limited | ❌ |
| Complex networking | ✅ Full control | Medium | Limited |
| Team has K8s expertise | Required | Not required | Not required |
EKS Deployment Models
| Model | Description | Operational Overhead | Use When |
|---|---|---|---|
| EKS Standard | Full control over nodes, add-ons, networking | Medium-High | Need full customization |
| EKS Auto Mode | AWS manages nodes, add-ons, scaling | Low | Want minimal ops, standard workloads |
| EKS with Fargate | Serverless pods, per-pod billing | Low | Batch, low-density workloads |
| EKS Hybrid Nodes | Your on-prem/edge nodes on an AWS-managed control plane | Medium-High | Reliable Region link; AWS-managed control plane on your own hardware |
| EKS on Outposts | EKS on AWS-owned hardware in your data center | High | Data residency or low latency; local clusters are disconnect-tolerant (not air-gapped) |
| EKS Anywhere | You run the control plane and nodes on your own infrastructure | Highest | Air-gapped, isolated, or disconnected sites |
For choosing among and operating the on-premises/hybrid models, see: Hybrid & On-Premises Deployments.
Routing: the advisory question "which on-prem/hybrid model should I choose?" stays in this skill; generating a design document or diagram → eks-design; generating a full production cluster Terraform project or an air-gapped build → eks-build (the hybrid cloud/cluster-side Terraform example itself lives in this skill's terraform-examples.md).
Shared Responsibility
| Component | AWS Manages | You Manage |
|---|---|---|
| Control plane | API server, etcd, HA, patching | RBAC, admission control, audit logging |
| Data plane (MNG) | AMI updates, node health | Instance type, scaling, pod scheduling |
| Data plane (Fargate) | Everything | Pod spec, resource requests |
| Data plane (Auto Mode) | Node lifecycle, OS patching | Workload definitions |
| Networking | ENI attachment, VPC CNI releases | Subnet design, IP planning, ingress |
| Security | Control plane auth | IAM, pod security, secrets, network policies |
Compute Selection Matrix
Decision Table
| Factor | Fargate | MNG | Karpenter | Auto Mode | Self-Managed |
|---|---|---|---|---|---|
| Best for | Batch, small scale | Stable, predictable | Dynamic, varied | Minimal ops | Custom AMI/kernel |
| Scaling | Per-pod | ASG-based | Fast, flexible | AWS-managed | Manual ASG |
| Spot support | ❌ | ✅ | ✅ Native | ✅ | ✅ |
| GPU support | ❌ | ✅ | ✅ | ✅ | ✅ |
| DaemonSets | ❌ | ✅ | ✅ | ✅ | ✅ |
| Cost model | Per vCPU/GB/hr | Per EC2 instance | Per EC2 instance | Per EC2 instance | Per EC2 instance |
| Max pods/node | 1 | ENI-based | ENI-based | AWS-managed | ENI-based |
| Node SSH | ❌ | ✅ | ✅ | ❌ | ✅ |
| Operational | Lowest | Low | Low | Lowest | Highest |
Quick Decision Guide
- Default choice: Karpenter — best balance of flexibility, cost, and automation
- Zero ops priority: EKS Auto Mode — AWS manages nodes, add-ons, and scaling via managed Karpenter. Best for teams that want Kubernetes benefits without operational overhead around upgrades, autoscaling, load balancing, and storage
- Serverless/batch: Fargate — no nodes to manage, per-pod billing
- Predictable, stable: MNG — familiar ASG model, managed updates
- Custom requirements: Self-managed — full control, highest overhead
✅ DO:
- Use Karpenter as the default node autoscaler for new clusters
- Run system components (CoreDNS, Karpenter) on MNG or Fargate
- Use multiple instance types for availability and cost optimization
❌ DON'T:
- Use self-managed nodes without a specific technical requirement
- Run Fargate for GPU or DaemonSet-dependent workloads
- Mix Karpenter and Cluster Autoscaler on the same node groups
Hybrid & On-Premises Deployments
Three models run EKS outside a standard in-Region cluster. Pick by who owns the control plane and whether the site stays connected:
| Model | Control plane | Reliable Region link? | Choose when |
|---|---|---|---|
| EKS Hybrid Nodes | AWS-managed, in-Region | Required | You want an AWS-managed control plane over your own on-prem/edge hardware |
| EKS Anywhere | You run it on your infra | Not required | Air-gapped, isolated, or disconnected sites; you own the full stack |
| EKS on Outposts | AWS-managed (in-Region for extended, on the Outpost for local) | Required (extended) / tolerates disconnect (local, racks only) | AWS-owned hardware for data residency or low latency |
Critical rule: split by why the site is disconnected. Permanently air-gapped/isolated → EKS Anywhere (the only fully air-gapped model). A connected site that must keep operating through Region outages on AWS-owned hardware (data residency/sovereignty) → Outposts local clusters (Outposts racks only) — disconnect-tolerant, not air-gapped (IAM/IRSA/KMS/EBS-PV/Route 53 are unavailable offline). EKS Hybrid Nodes and Outposts extended clusters both depend on a reliable connection to an AWS Region.
For detailed hybrid-deployment guidance, see: Hybrid & On-Premises Deployments
Networking Quick Reference
VPC CNI Mode Decision
| Mode | Use When | Pod Density |
|---|---|---|
| Secondary IP (default) | Most workloads, simple setup | Limited by ENI × IPs per ENI |
| Prefix Delegation | >30 pods/node, IP-constrained VPC | 4-16× more pods per node |
| Custom Networking | Pods need different CIDR than nodes | Same as underlying mode |
Ingress Pattern Selection
| Pattern | Best For | Key Feature |
|---|---|---|
| ALB (via LBC) | HTTP/HTTPS web apps | Native WAF, Cognito auth |
| NLB (via LBC) | TCP/UDP, gRPC, low latency | Static IPs, source IP preservation |
| Gateway API | Multi-team, new deployments | ✅ Recommended standard |
| VPC Lattice | Cross-VPC service-to-service | No sidecar, IAM auth |
IPv4 vs IPv6
| Factor | IPv4 | IPv6 |
|---|---|---|
| Default choice | ✅ Yes | When facing IP exhaustion |
| AWS service support | Full | Most (check specific services) |
| Complexity | Standard | Requires dual-stack VPC |
For detailed networking guidance, see: Networking — VPC CNI & IP | Networking — Ingress & DNS
Security Essentials
IAM Strategy
| Approach | Use When | Setup |
|---|---|---|
| Pod Identity | ✅ New workloads | EKS add-on + association |
| IRSA | Older clusters, Fargate | OIDC provider + trust policy |
Key rules:
- ✅ Use Pod Identity for new workloads — simpler setup, session tags, role chaining
- ✅ Use EKS access entries (API mode) over aws-auth ConfigMap
- ✅ Move VPC CNI permissions from node role to Pod Identity/IRSA
- ❌ Don't use wildcard conditions in IRSA trust policies
- ❌ Don't attach application permissions to node IAM roles
Pod Security Baseline
Apply Pod Security Admission (PSA) labels to all namespaces:
# Minimum: enforce baseline, warn on restricted
metadata:
labels:
pod-security.kubernetes.io/enforce: baseline
pod-security.kubernetes.io/warn: restricted
Secrets Management
| Approach | Complexity | Best For |
|---|---|---|
| External Secrets Operator | Medium | ✅ GitOps workflows |
| Secrets Store CSI | Medium | Mount secrets as volumes |
| KMS envelope encryption | Low | Encrypt etcd secrets |
Envelope encryption is on by default (≥1.28) and covers all Kubernetes API data; layer your own KMS CMK when you need key control and CloudTrail visibility. Source: https://docs.aws.amazon.com/eks/latest/userguide/envelope-encryption.html
For detailed security guidance, see: Security Reference | Runtime & Network | Supply Chain & Compliance
Reliability Essentials
Pod Disruption Budgets
Create PDBs for every production workload with >1 replica:
| Workload | Recommended PDB |
|---|---|
| Stateless (3+ replicas) | minAvailable: "50%" |
| Stateful quorum (3) | maxUnavailable: 1 |
| Batch/job | maxUnavailable: "50%" |
| Singleton | No PDB (would block all disruptions) |
Health Probe Strategy
| Probe | Purpose | Key Rule |
|---|---|---|
| Startup | Wait for slow init | Use for apps >10s startup |
| Readiness | Traffic routing | ✅ Check dependencies here |
| Liveness | Detect deadlocks | ❌ Never check dependencies |
Critical rule: Liveness probes must NOT check external dependencies. If the database goes down and liveness checks the DB, ALL pods restart — causing cascading failure.
Graceful Shutdown Pattern
spec:
terminationGracePeriodSeconds: 60
containers:
- lifecycle:
preStop:
exec:
command: ["/bin/sh", "-c", "sleep 15"]
Why sleep 15: Gives kube-proxy and load balancer time to remove the pod from traffic routing before SIGTERM.
Multi-AZ Distribution
topologySpreadConstraints:
- maxSkew: 1
topologyKey: topology.kubernetes.io/zone
whenUnsatisfiable: DoNotSchedule
For detailed reliability guidance, see: Reliability & Resiliency — Core (see also reliability-advanced.md for DR, deployment strategies, and large-cluster guidance)
Cluster Upgrade Strategy
Upgrade Sequence (Strict Order)
1. Control Plane → 2. EKS Add-ons → 3. Data Plane → 4. Custom Add-ons
Pre-Upgrade Checklist
- Check EKS Cluster Insights for upgrade readiness
- Scan for deprecated APIs (Pluto, kube-no-trouble)
- Verify add-on compatibility with target version
- Test in non-prod environment first
- Ensure PDBs are configured for graceful node drain
- Back up cluster state (Velero or GitOps repo)
Upgrade Strategy Decision
| Factor | In-Place | Blue-Green |
|---|---|---|
| Risk | Low-Medium | Lowest |
| Cost | No extra | 2× during migration |
| Rollback | ✅ To N-1 within 7 days of upgrade | ✅ Switch back (any window) |
| Use when | ✅ Most upgrades | Post-7-day rollback, multi-minor jumps, data-plane isolation |
Native in-place rollback to N-1 is supported within 7 days of an in-place upgrade (all regions; Auto Mode rolls back nodes automatically; rollback into an extended-support version requires setting the cluster upgrade policy to EXTENDED first). update-cluster-version supports type VersionRollback, gated by a ROLLBACK_READINESS insight; add-ons/data-plane roll back separately. Blue-green retains independent rationale for post-7-day windows, multi-minor jumps, and data-plane isolation — not "because you can't roll back."
Data Plane with Karpenter
Karpenter automatically replaces nodes via drift detection after control plane upgrade. Control the speed with disruption.budgets:
disruption:
budgets:
- nodes: "10%" # Max 10% of nodes replaced at a time
For detailed upgrade guidance, see: Cluster Upgrades Reference
Autoscaling Quick Reference
Node Autoscaler Selection
| Karpenter | Cluster Autoscaler | Auto Mode | |
|---|---|---|---|
| Default choice | ✅ Yes | Legacy/Outposts | Minimal ops |
| Scale-up speed | ~30s | ~60-90s | AWS-managed |
| Consolidation | ✅ Built-in | ❌ | ✅ |
| Customization | High | Medium | Low |
Cluster Autoscaler's ~60-90s scale-up assumes cold EC2 launches; MNG EC2 warm pools (2026-04) pre-initialize instances and cut that latency substantially for MNG-backed node groups.
Pod Autoscaler Selection
| Scaler | Trigger | Use Case |
|---|---|---|
| HPA | CPU, memory, custom | Stateless services |
| VPA | Historical usage | Right-sizing (recommendation mode) |
| KEDA | External events (SQS, Kafka) | Event-driven workloads |
For detailed autoscaling guidance, see: Autoscaling Reference | Karpenter Reference
Preparing for a Known Traffic Peak
For a planned peak — a flash sale, marketing push, product launch, or seasonal event — the arrival is a near-instantaneous step at a known clock time, so favor scheduled pre-scaling (set floors ahead of the trigger) over relying on reactive autoscaling to catch up. Pre-warm all three layers (control plane, nodes, pods) and hold the floor across the whole event window rather than scaling down between peaks.
Critical rule: A load test that passes on request rate can still miss the failure that hits production — if it never reproduces the resource-consumption pattern (memory working-set, connection count) of the event's actual access patterns, especially for new features. Test consumption shapes, not just throughput, and act on findings before the event.
For detailed surge-readiness guidance, see: Surge Readiness Reference
Terraform Examples Quick Start
Based on terraform-aws-modules/terraform-aws-eks.
Example Selection
| Starting Point | Recommended Example |
|---|---|
| General production | karpenter (MNG for system + Karpenter for workloads) |
| Minimal ops | eks-auto-mode |
| Managed nodes | eks-managed-node-group (AL2023 or Bottlerocket) |
| Full node control | self-managed-node-group |
| Platform capabilities | eks-capabilities (ArgoCD, ACK, KRO) |
| Hybrid/edge | eks-hybrid-nodes |
Common Deployment Topologies
Private cluster with Karpenter:
VPC (3 AZs, terraform-aws-modules/vpc/aws)
├── Private subnets → EKS nodes (MNG for system, Karpenter for workloads)
├── Public subnets → ALB (internet-facing)
├── Intra subnets → EKS control plane ENIs
└── NAT Gateway → 1 per AZ for production
Multi-tenant platform:
EKS Cluster (terraform-aws-modules/eks/aws)
├── kube-system (platform: CoreDNS, kube-proxy, VPC CNI)
├── karpenter (Karpenter controller on MNG)
├── monitoring (shared: Prometheus, Grafana)
├── ingress (shared: AWS LBC)
├── team-a namespace (RBAC, NetworkPolicy, ResourceQuota)
├── team-b namespace (RBAC, NetworkPolicy, ResourceQuota)
└── team-c namespace (RBAC, NetworkPolicy, ResourceQuota)
For detailed examples and terraform patterns, see: Terraform Examples Reference
Cost Optimization Quick Wins
| Action | Savings | Effort |
|---|---|---|
| Graviton (arm64) | 20-40% | Low |
| Spot for non-critical | 60-90% | Low |
| Karpenter consolidation | 20-30% | Low |
| VPA right-sizing | 15-30% | Medium |
| gp3 over gp2 | 20% on EBS | Low |
| VPC endpoints | Eliminate NAT costs | Low |
For detailed cost guidance, see: Cost Optimization Reference | For scalability guidance, see: Scalability Reference
Observability Quick Reference
| Pillar | AWS-Managed | Open Source |
|---|---|---|
| Metrics | Container Insights | AMP + Grafana |
| Logs | CloudWatch Logs | OpenSearch, Loki |
| Traces | X-Ray | ADOT + Jaeger/Tempo |
Essential: Enable EKS audit logging and GuardDuty EKS Runtime Monitoring for security visibility.
For detailed observability guidance, see: Observability Reference
EKS Capabilities
EKS Capabilities are AWS-managed features installed and updated as part of the EKS platform. They run in AWS-owned infrastructure separate from your clusters, with AWS handling scaling, patching, and upgrading.
| Capability | What It Does | When to Use Managed | When to Self-Manage |
|---|---|---|---|
| ArgoCD | GitOps continuous delivery | Multi-account hub-and-spoke, IAM IDC integration, minimal ops | Custom plugins, air-gapped, existing ArgoCD investment |
| ACK | Manage AWS resources via K8s CRDs (S3, RDS, IAM, etc.) | Standard AWS resource management | Specific controller version pinning, custom config |
| KRO | Platform abstractions via ResourceGroupDefinitions | Golden path templates, multi-resource compositions | Early adoption risk concerns, custom reconciliation logic |
Combined pattern: ArgoCD deploys ACK resources + KRO compositions via GitOps, providing a single workflow for both infrastructure and applications.
For detailed ArgoCD patterns, see: ArgoCD Patterns Reference
Sources:
- EKS Capabilities Documentation
- AWS Blog — Deep dive: Simplifying resource orchestration with Amazon EKS Capabilities
Detailed References
This skill uses progressive disclosure — essential guidance is in this main file, detailed reference material is loaded on demand:
- Security — IAM, Cluster Access Manager, Pod Identity, IRSA, pod security standards, multi-tenancy, secrets management, data encryption
- Security — Runtime & Network — Runtime threat detection (GuardDuty, seccomp, AppArmor, Falco), network policies, SG for pods, encryption in transit, detective controls
- Security — Supply Chain & Compliance — Image security (SBOMs, attestations, ECR hardening), infrastructure hardening (Bottlerocket, CIS benchmarks), regulatory compliance, incident response
- Networking — VPC CNI modes (secondary IP, prefix delegation, custom networking), subnet/CIDR planning, IPv4 vs IPv6, Security Groups for Pods, IP address management
- Networking — Ingress & DNS — Ingress patterns (ALB, NLB, Gateway API), AWS Load Balancer Controller, service mesh, DNS/CoreDNS tuning, private cluster connectivity
- Reliability & Resiliency — Core — HA patterns, PDBs, health probes, load balancer health checks, lifecycle hooks, topology spread, resource management
- Reliability & Resiliency — Advanced — disaster recovery, zonal shift, deployment strategies, large cluster guidance, chaos engineering, admission-controller topology enforcement
- Surge Readiness — preparing for a known traffic peak (flash sale, marketing push, product launch, seasonal event): spike shape vs magnitude, scheduled pre-scaling, capacity assurance (ODCR, quotas, LB warm-up), load-test realism, graceful degradation, and a descriptive pre-event readiness checklist
- Autoscaling — Autoscaler selection, Cluster Autoscaler (IAM, Spot, overprovisioning, parameter tuning), HPA, VPA, KEDA, CoreDNS autoscaling
- Karpenter — Operational best practices, NodePools, EC2NodeClass, Spot/interruption handling, consolidation, multiple NodePool strategy, cost controls, resource management, private clusters, CoreDNS with Karpenter
- Cluster Upgrades — In-place and blue-green upgrades, pre-upgrade validation, add-on management, API deprecation detection, version skew policy, Bottlerocket updates, rollback procedures
- Cost Optimization — CFM framework, compute/networking/storage cost strategies, observability cost management, Spot, Graviton, tagging, Kubecost
- Scalability — Scaling theory (churn rate, QPS), control plane (APF, config parameters, monitoring), data plane (node sizing, diversity), cluster services (CoreDNS, Metrics Server), workload patterns, IPVS, large-cluster guidance
- Observability — Observability strategy, CloudWatch Container Insights & Application Signals, Prometheus/Grafana, control plane monitoring, network performance monitoring, logging architecture, distributed tracing, GPU/AI-ML observability, detective controls, alerting patterns
- Terraform Examples — terraform-aws-modules/terraform-aws-eks examples, submodules, add-on management, Provisioned Control Plane, EFA, VPC patterns, deployment topologies
- ArgoCD Patterns — ArgoCD architecture, App of Apps, ApplicationSets, GitOps Bridge, multi-cluster patterns (hub-and-spoke, decentralized, hybrid), EKS ArgoCD Capability (managed vs self-managed, migration), ACK/KRO integration, multi-tenant RBAC
- Container Registry — ECR architecture, operating models, image promotion, vulnerability scanning, base image curation, lifecycle policies, pull-through cache, repository creation templates, managed signing (AWS Signer), archival storage class, registry configuration
- EKS Auto Mode — Auto Mode architecture, managed NodePools/NodeClasses, migration from standard EKS, comparison with self-managed Karpenter, limitations and FAQ
- Hybrid & On-Premises Deployments — EKS Hybrid Nodes, EKS Anywhere, and EKS on Outposts: model-selection decision boundary, per-model networking/CIDR/CNI, disconnection behavior, compute & autoscaling support, identity, lifecycle, and a cross-model support matrix
How to use: When you need detailed information on a topic, reference the appropriate guide. Claude will load it on demand.