Скачать или смотреть Platform engineering on AWS EKS

‪@BranFolio‬ Starting a series on multi-tenant platform engineering — what works, what doesn’t, straight from the trenches. This story is of one cluster that learned to host a 1000 strangers without ever flinching.
Imagine standing on the roof of a datacenter in Montreal, Quebec, part of AWS's ca-central-1 region, at 03:14 UTC. Below you, rows of blinking servers hum like a single, giant lung. Somewhere inside one of those racks lives an Amazon EKS cluster whose name is simply “aihq-prod-1.” It is not the largest cluster AWS has ever hosted—barely 350 nodes at peak—but it is the most crowded apartment building in the cloud: eight hundred and forty-three tenants, thirty-two micro-services, six AI pipelines, two compliance scanners, and one on-call engineer who is currently asleep because the cluster has never called them in the middle of the night.

This article is the biography of that apartment building: how we designed it, how we keep the neighbors from fighting, and how we replace the plumbing while the water is still running. #Kubernetes #AWS #AWSEKS #multi-tenant #GPU #AI #ML #Kubeflow #pipeline #spotinstance #cloudzero #aiHQ #CPU #DevOps #DevSecOps #FinOps #ingress #multitenant #multi-tenancy #conformalcloud #contextscloud #Karpenter #cert-manager #clusterautoscaler #velero #route53 #cluster #istio #servicemesh #argocd #GitOps #IRSA #RBAC #networkpolicy #CVE #platformengineering