EKS Pod Scheduling & Availability Patterns
Written: 2026-02-12 | Updated: 2026-02-14 | Reading time: ~54 min
Reference Environment: EKS 1.30+, Karpenter v1.x, Kubernetes 1.30+
1. Overview
Kubernetes Pod scheduling directly impacts service availability, performance, and cost efficiency. Proper scheduling strategies provide: high availability (failure domain isolation), performance optimization (workload-appropriate node placement), resource efficiency (balanced node utilization), and stable operations (priority-based resource guarantees).
2. Kubernetes Scheduling Fundamentals
3-phase process: Filtering (Predicates) → Scoring (Priorities) → Binding
| Factor | Type | Phase | Enforcement |
|---|---|---|---|
| Node Selector | Pod | Filtering | Hard |
| Node Affinity | Pod | Filtering/Scoring | Hard/Soft |
| Pod Affinity | Pod | Scoring | Hard/Soft |
| Pod Anti-Affinity | Pod | Filtering/Scoring | Hard/Soft |
| Taints/Tolerations | Node+Pod | Filtering | Hard |
| Topology Spread | Pod | Scoring | Hard/Soft |
| PriorityClass | Pod | Preemption | Hard |
3. Node Affinity & Anti-Affinity
Node Selector (basic exact match), Node Affinity (complex conditions with In, NotIn, Exists, DoesNotExist, Gt, Lt operators), Required vs Preferred, weight-based preferences.
4. Pod Affinity & Anti-Affinity
Pod Affinity for co-location (cache locality, data locality). Pod Anti-Affinity for fault isolation (distribute replicas across nodes/AZs). topologyKey: hostname, zone, region, custom.
5-9. Taints/Tolerations, Topology Spread, PDB, PriorityClass, Descheduler, Advanced Patterns
Comprehensive coverage including: dedicated node isolation, GPU/Spot taints, Topology Spread Constraints with minDomains, PDB design, PriorityClass hierarchy, Descheduler+Karpenter combination, and AI/ML workload scheduling patterns.
10. 2025-2026 AWS Innovations
EKS Auto Mode scheduling, Karpenter v1 GA features, Node Readiness Controller.