AWS Dependencies
AWS-specific requirements only. Everything cloud-agnostic (PostgreSQL extensions, artifact access, domain names, secrets handling) is defined once in the Environment Dependencies registry; this page refines those rows for EKS.
Cluster
| Dependency | Requirement | Why It Is Needed | Verify |
|---|---|---|---|
| EKS cluster | A provisioned EKS cluster with the VPC CNI, CoreDNS, and Amazon EBS CSI Driver add-ons installed. | These add-ons back pod networking, service discovery, and volume provisioning for the platform. | kubectl get pods -n kube-system shows VPC CNI, CoreDNS, and EBS CSI pods Running; a test PersistentVolume provisions successfully. |
| IAM for service accounts | IRSA (IAM Roles for Service Accounts) configured for workloads that access S3. | Pods assume scoped IAM roles instead of using static keys. | Service accounts carry the eks.amazonaws.com/role-arn annotation and a test pod can list its bucket. |
Compute
| Dependency | Requirement | Why It Is Needed | Verify |
|---|---|---|---|
| GPU nodes | p4d.24xlarge instances (p4de.24xlarge is an accepted substitute) provisioned according to DynamoAI guidance. | DynamoGuard inference and training require these GPU classes (see the core CUDA requirement). | GPU nodes join the cluster and advertise nvidia.com/gpu capacity. |
| CPU nodes | Node capacity from the m5, c7i, c8i, or c5n families. | DynamoGuard base functions and DynamoEval run on these families. | Autoscaling provisions nodes from the approved families on demand. |
| Karpenter prerequisites | If Karpenter performs autoscaling: the IAM role and SQS interruption queue it requires must be provisioned through infrastructure-as-code before installation. | Karpenter crash-loops when its AWS-side resources do not exist. | Karpenter controller pods are Running and a node pool update produces a new Ready node. |
Databases and Storage
| Dependency | Requirement | Why It Is Needed | Verify |
|---|---|---|---|
| RDS PostgreSQL | Multi-AZ RDS PostgreSQL with automated failover, sized to the projected regional workload, satisfying the core PostgreSQL row (extensions, dfl and keycloak databases). | Regional HA for the primary relational store. | Multi-AZ shown in the RDS console; endpoint reachable from the cluster. |
| S3 buckets | Two buckets, one for model storage and one for platform/application artifacts, with least-privilege IAM access from the cluster. | Concrete AWS form of the core object-storage requirement. | Bucket names and ARNs documented; a test read/write succeeds from a pod via IRSA. |
Networking and Ingress
| Dependency | Requirement | Why It Is Needed | Verify |
|---|---|---|---|
| AWS Load Balancer Controller | The AWS Load Balancer Controller (v2.13.x) provisioned, with ingress for the three platform endpoints (apps., auth., api.) and their certificates. It can be installed through the DynamoAI cluster prerequisites chart or provided by the customer. | Exposes the platform to the customer network; the final handover check depends on it. | All three endpoints respond over HTTPS from the customer network. |
Known Issues
No AWS-specific known issues are currently recorded.
Pre-Deployment Checklist
Your progress is saved in your browser. Use Download to export the current state as Markdown.
AWS Pre-Deployment0/10