re:Invent · architecture
account · prod regions · 3 workloads · ~150 services · ~40

EKS · ArgoCD · App-of-Apps

Multi-cluster hub-spoke sync on Personal lab

One ArgoCD instance in us-east-2 reconciles desired state for two spoke clusters in ap-south-1 and eu-west-1. Karpenter handles compute on every cluster; KEDA drives the agent-orchestrator workload. New Relic and Zenduty close the loop.

▲ HUB · CONTROL PLANE
prod-master
us-east-210.0.0.0/16
G
ArgoCD App-of-Apps · application-controller · repo-server · server
K
Karpenter NodePool: spot + on-demand
⤢
KEDA ScaledObject → agent-orchestrator
W
Workloads ~15 microservices · agent-orchestrator · control-api · workflow-runner
SPOKE · APAC
prod-india
ap-south-110.10.0.0/16
⇆
argocd-agent (sync target) receives App-of-Apps manifests from hub
K
Karpenter NodePool: spot + on-demand
⤢
KEDA ScaledObject → agent-orchestrator
W
Workloads ~13 microservices · regional replicas of core services
SPOKE · EMEA
prod-eu
eu-west-110.20.0.0/16
⇆
argocd-agent (sync target) receives App-of-Apps manifests from hub
K
Karpenter NodePool: spot + on-demand
⤢
KEDA ScaledObject → agent-orchestrator
W
Workloads ~12 microservices · regional replicas of core services
OBSERVABILITY · OPS
3× newrelic-agent → New Relic · single account → Zenduty → ▌ ON-CALL
DaemonSets on all 3 clusters
APM + infra + k8s events
LEGEND Compute · Karpenter Networking · VPC Workloads · services Observability · NR + Zenduty GitOps · ArgoCD