Cloud & DevOps / Reference implementation

WhiteFriday Engine

An event-traffic scaling lab combining EKS, Karpenter, application autoscaling, load tests, and disruption scenarios.

Explore the architecture ↓View repository ↗

Focus: Infrastructure & platform engineering
Project type: Reference implementation
Core technologies: Terraform · EKS · Karpenter · Prometheus

WhiteFriday Engine concept interface illustration — Concept interface illustration

Context

The engineering problem

Large sales events create uneven demand. Capacity planning needs application signals, node provisioning, cost controls, and tests for infrastructure disruption.

System architecture

Architecture at a glance

A simplified implementation map based on the linked repository. Arrows show the principal relationships; deployment details live in the source.

clientk6 / Locust traffic
Load scenarios
serviceEKS demo services
serviceEKS demo services
Metrics endpoint
infraPrometheus metrics
infraPrometheus metrics
Scaling signals
infraHPA replicas
infraHPA replicas
Pending workload demand
infraKarpenter capacity
infraKarpenter capacity
Node provisioning
infraSpot / On-Demand

All system components (6)

k6 / Locust trafficclient
EKS demo servicesservice
Prometheus metricsinfra
HPA replicasinfra
Karpenter capacityinfra
Spot / On-Demandinfra

Approach

Implementation decisions

Terraform
EKS
Karpenter
Prometheus
k6
Locust
Litmus

Scale at two levels

HPA configuration addresses workload replicas while Karpenter node pools address underlying compute capacity.

Mix capacity deliberately

Separate Spot and On-Demand node-pool definitions make capacity tradeoffs explicit.

Exercise failure paths

Load-test scenarios, zone disruption configuration, and rollback scripts make reliability work inspectable.

Explore the implementation

Start with the source.

These entry points connect the architecture to the implementation.

Implementation scope

Scaling and resilience reference environment with an in-memory demo order service. The Karpenter manifests target an older API, and the k6 scenario includes an undefined metric. Both need correction for the chosen cluster and test harness before a measured load run.