Skip to content

cost-estimate.md

Latest commit

 

History

History
115 lines (89 loc) · 4.99 KB

cost-estimate.md

File metadata and controls

115 lines (89 loc) · 4.99 KB
Raw

Hourly Cost Estimate — RKE2 and EKS Labs

All prices are on-demand us-west-2 list prices (Apr 2026). Both labs share a common VPC/VPN layer; cluster stacks are mutually exclusive (only one runs at a time, per the vCPU-quota constraint noted in CLAUDE.md).

Data-transfer and usage-driven charges (NAT data processing, VPC-endpoint data processing, NLB LCUs, CloudWatch Logs ingest, S3 PUTs, inter-AZ traffic, EBS snapshots) are excluded — they're generally small in a lab but vary.

Shared / always-on (Common)

Resource Qty Unit $/hr $/hr
OpenVPN — t3.small (openvpn/devvpn/variables.tf:41) 1 0.0208 0.0208
OpenVPN root 20 GB gp3 (openvpn/devvpn/variables.tf:47) 20 GB 0.08/GB-mo 0.0022
OpenVPN Elastic IP (public IPv4 is metered) 1 0.005 0.0050
NAT Gateway — single_nat_gateway = true (VPC/dev/variables.tf:86) 1 0.045 0.0450
Interface VPC endpoints × 3 AZs (6 services — see below) 18 ENIs 0.010 0.1800
Common subtotal ~$0.253 /hr

Interface endpoint services (VPC/dev/variables.tf:75): ecr.api, ecr.dkr, ssm, ssmmessages, ec2messages, sts. Gateway endpoints (s3, dynamodb) are free. Each interface endpoint places one ENI per AZ (3 AZs) per VPC/modules/vpc/mainf.tf:116.

Biggest single line item in the common layer. Because the lab also runs a NAT gateway, these endpoints are largely redundant for egress. Setting enable_vpc_endpoints = false in VPC/dev/terraform.tfvars drops the common baseline from $0.25/hr to **$0.07/hr** with no functional loss for this lab. The endpoints only pay off when trying to eliminate NAT (Lambda-in-VPC, air-gapped workloads).

If 8 interface endpoints are actually deployed

The Terraform in this repo declares 6 interface services (+2 free gateway = 8 total objects). If the live VPC actually has 8 interface endpoints (e.g. logs, kms, ec2, elasticloadbalancing added out-of-band), recompute:

Interface endpoints $/hr (× 3 AZs) $/mo
6 (repo default) 0.18 131
8 0.24 175

RKE2 lab (on top of common)

Resource Qty Unit $/hr $/hr
Nodes — t3a.large (3 servers + 3 agents, RKE-cluster/dev-cluster/ec2/main.tf:8,12) 6 0.0752 0.4512
EBS gp3 root (~50 GB Ubuntu 22.04 default × 6) 300 GB 0.08/GB-mo 0.0329
Traefik NLB (provisioned by deployments/rke-apps/1-infrastructure/) ~1 ~0.025 0.0250
S3 etcd backups + 1 Secrets Manager token (dev-rke2-token) — — ~0.001
RKE2 subtotal ~$0.510 /hr
RKE2 lab total (common + RKE2) ≈ $0.76 /hr (~$554/mo)

Node count is fixed, so the RKE2 total has no meaningful variability.

EKS lab (on top of common)

Resource Qty Unit $/hr $/hr
EKS control plane (EKS-cluster/eks-cluster/2-cluster/main.tf:30) 1 0.10 0.1000
Bootstrap managed node group — t3.medium × 3 (2-cluster/variables.tf:82,88) 3 0.0416 0.1248
EBS gp3 50 GB × 3 (3-karpenter/node-pools.tf:39) 150 GB 0.08/GB-mo 0.0164
Traefik NLB (deployments/eks-apps/1-infrastructure/) ~1 ~0.025 0.0250
KMS CMK for EKS secrets (1-iam/kms.tf) 1 $1/mo 0.0014
EKS subtotal (no Karpenter nodes running) ~$0.267 /hr
EKS lab idle total (common + EKS) ≈ $0.52 /hr (~$380/mo)

Karpenter variable cost

Both NodePools are capped at 16 vCPU / 64 GiB (EKS-cluster/eks-cluster/3-karpenter/node-pools.tf:56–156) across m5a.large / m5a.xlarge.

Scenario Additional $/hr
1× m5a.large on-demand +0.086
1× m5a.large spot (~70% off) +0.026
NodePool at cap, all on-demand (e.g. 2× m5a.xlarge) +0.344
NodePool at cap, all spot +0.103

Under a real workload the EKS lab lands ≈ $0.55–$0.90 /hr, up to ≈ $0.86 /hr worst case at the pool cap on-demand.

Summary

Lab Idle baseline Realistic with workload
RKE2 ~$0.76 /hr same (fixed node count)
EKS ~$0.52 /hr ~$0.55 – $0.90 /hr

Cost-reduction quick wins

  1. Disable interface VPC endpoints — saves $0.18/hr ($131/mo). NAT gateway already provides egress to all the target services. Set enable_vpc_endpoints = false in VPC/dev/terraform.tfvars.
  2. Tear down the lab when not in use. Baseline is ~$0.25/hr even with no cluster; with a cluster up it's $0.5–0.8/hr. A day's usage is $6–20; a full month idle is ~$380–560.
  3. Prefer the EKS lab for ad-hoc work — its bootstrap node group is only 3× t3.medium vs. RKE2's 6× t3a.large, and Karpenter scales to zero when workloads are gone.
  4. Favor spot in Karpenter NodePools — the spot-interruptible NodePool already exists; prefer it for non-critical workloads (~70% off on-demand).
  5. NAT gateway is already cost-optimized (single_nat_gateway = true). No action needed.