Limits

What no apply can finish

• Replacing the cluster (cluster_name, bootstrap_cluster_creator_admin_permissions): apply -target=module.socle.module.foundations first, then the full apply (root README).
• Destroying with the API unreachable: if the runner left cluster_endpoint_public_access_cidrs, run tofu state rm module.socle.module.socle first.
• A late control-plane log delivery: with cluster_log_types set, a destroy waits five minutes between the cluster and its log group; EKS’s delivery is best effort, and a later one recreates the group (Uninstall).
• The GuardDuty detector: the account owner’s.
• The Route 53 zone: with gateway_certificate, the public zone must already exist; the module never creates it.

Provider limits

• No IPv6 cluster: Cilium’s ENI IPv6 mode is beta.
• No Auto Mode, no Fargate: neither runs Cilium.
• EKS picks the service CIDR (172.20.0.0/16 or 10.100.0.0/16); no variable for it.
• The API endpoint is public, restricted to cluster_endpoint_public_access_cidrs, which must include the runner.
• AWS upgrades the cluster at the end of standard support, whatever the pipeline did.
• An existing VPC (create_vpc = false) gets no VPC endpoints and no flow logs.
• The NLBs’ TLS policy is AWS’s default; the socle sets none.
• Elastic IPs: one per AZ per cluster, 5 per region by default (Prerequisites).

What the socle does not offer here yet

• Capacity for workloads: no Karpenter (#53); everything runs on the two bootstrap Spot nodes.
• Backups outside the cluster’s account and region (velero).
• The Upgrade Insights pre-check and cloud-resource observability: proposed, not built.