Examples

Each example is a self-contained deployable pattern with its own README explaining the "why" alongside the "how." Deploy the base cluster once, then apply individual examples to explore.

Compute Patterns

Example	Description
Graviton	ARM64 workloads on cost-effective Graviton instances
Spot	Fault-tolerant workloads on EC2 Spot with diverse instance families
GPU	GPU-accelerated ML inference (Qwen 3 on NVIDIA GPUs)
Neuron	ML inference on AWS Inferentia2 (DeepSeek-R1 served by vLLM)

Cost Optimization

Example	Description
Cost Optimization	OD/Spot mixed pools with weighted priorities and pause-pod overprovision

Advanced Scheduling

Example	Description
Capacity Reservation	Pin workloads to On-Demand Capacity Reservations (ODCRs)
Static Capacity	Fixed fleet of always-on nodes using spec.replicas
Batch Jobs	Protect long-running jobs from eviction with do-not-disrupt
Disruption Budgets	Limit simultaneous node drains during consolidation

Autoscaling

Example	Description
Pod Autoscaling	HPA for CPU-based scaling + KEDA for event-driven scaling

Observability

Example	Description
Observability	CloudWatch Container Insights integration