AWS EC2 Quota Management

How to prevent quota-related Karpenter failures on AWS EKS (eu-north-1 / Stockholm)

Overview

AWS enforces per-region Service Quotas on EC2 instances. Karpenter must stay within these limits or node provisioning fails silently. Unlike OVH (single project quota), AWS maintains separate quotas per instance-type family for both On-Demand and Spot instances.

The Problem

10 pending pods × 8 vCPU each = 80 vCPU total

Karpenter: "I'll provision c7g.2xlarge nodes (8 vCPU each) to cover the demand"

AWS API: "Your Running On-Demand Standard instances quota is 32 vCPU. 80 > 32. Denied."

Karpenter retries, nodeclaims loop in Pending, tasks never start.

Symptoms

Symptom Likely Cause
Nodeclaims stuck in Pending On-Demand or Spot vCPU quota exhausted
InsufficientInstanceCapacity in EC2 events No capacity in AZ for requested family
NodeCreationError in Karpenter logs EC2 API rejected the RunInstances call
MaxSpotInstanceCountExceeded Spot instance request quota reached
Pods pending > 5 min despite Karpenter scaling events Instance launch failed after NodeClaim created

Quota Types

AWS distinguishes four quota categories relevant to this deployment:

Quota Name Scope Default (eu-north-1) Notes
Running On-Demand Standard (A,C,D,G,I,M,R,T,Z) vCPU per region 32 vCPU Covers c7g, c6g, m7g, r6g, t4g, etc.
Running On-Demand F instances vCPU per region 0 FPGA — not used
Running Spot Instance Requests vCPU per region 5 vCPU Applies to all Spot families combined
Amazon EKS clusters Count per region 100 Rarely a bottleneck

Critical: The default 5 vCPU Spot limit is nearly useless for genomics workloads. This must be increased before using Spot instances with Karpenter.

Quota Recommended Minimum Notes
On-Demand Standard vCPU 64–128 vCPU Matches Karpenter NodePool limits.cpu
Spot vCPU 64–256 vCPU Size to peak concurrent task demand
EKS clusters 10 Default 100 is usually fine

Instance Families in eu-north-1

Family vCPU RAM On-Demand (xlarge) Spot est. Generation
c7g 4 / 8 / 16 / 32 8 / 16 / 32 / 64 GB ~$0.073/hr ~$0.022/hr Graviton3 ⭐
c6g 4 / 8 / 16 / 32 8 / 16 / 32 / 64 GB ~$0.065/hr ~$0.020/hr Graviton2
c6i 4 / 8 / 16 / 32 8 / 16 / 32 / 64 GB ~$0.085/hr ~$0.026/hr Intel Ice Lake
c6a 4 / 8 / 16 / 32 8 / 16 / 32 / 64 GB ~$0.076/hr ~$0.023/hr AMD EPYC

General Purpose (when extra RAM needed)

Family vCPU RAM On-Demand (xlarge) Spot est.
m7g 4 / 8 / 16 / 32 16 / 32 / 64 / 128 GB ~$0.092/hr ~$0.028/hr
m6g 4 / 8 / 16 / 32 16 / 32 / 64 / 128 GB ~$0.083/hr ~$0.025/hr
m6i 4 / 8 / 16 / 32 16 / 32 / 64 / 128 GB ~$0.107/hr ~$0.032/hr

Prices are eu-north-1 on-demand Linux, approximate as of March 2026. Spot prices fluctuate. Check the AWS Spot Price History for current rates.

Three-Layered Karpenter Filtering Strategy

Layer 1: Instance Family Allowlist (NodePool)

Restrict Karpenter to instance families that are quota-approved and appropriate for the workload:

# karpenter NodePool — instance type filtering
spec:
  template:
    spec:
      requirements:
        - key: "karpenter.k8s.aws/instance-family"
          operator: In
          values: ["c7g", "c6g", "c6i", "c6a"]    # compute-optimised only
        - key: "karpenter.k8s.aws/instance-size"
          operator: In
          values: ["xlarge", "2xlarge", "4xlarge"]  # 4–16 vCPU per node
        - key: "kubernetes.io/arch"
          operator: In
          values: ["arm64", "amd64"]

Layer 2: Per-Node vCPU Cap (instance-size)

Preventing Karpenter from selecting large nodes keeps quota consumption predictable:

# Cap nodes at 4xlarge (16 vCPU) maximum
- key: "karpenter.k8s.aws/instance-size"
  operator: In
  values: ["xlarge", "2xlarge", "4xlarge"]
# c7g.8xlarge (32 vCPU) is excluded → stays within 64 vCPU quota with 4 nodes

Layer 3: NodePool Aggregate Limits

spec:
  limits:
    cpu: "64"         # Karpenter will not provision beyond 64 total vCPU
    memory: "256Gi"   # Matches typical quota ceiling

How to Request Quota Increases

  1. Open the AWS Service Quotas console in eu-north-1
  2. Search for Running On-Demand Standard instances
  3. Click Request quota increase → enter the desired vCPU count
  4. Repeat for Running Spot Instance Requests
  5. Requests are typically processed within minutes to a few hours

Quota increases in eu-north-1 are subject to AWS regional capacity. If a request is denied, try requesting a smaller increase first or contact AWS support.

Checking Current Quota Usage

# Check current On-Demand vCPU usage vs quota
aws service-quotas get-service-quota \
  --service-code ec2 \
  --quota-code L-1216C47A \
  --region eu-north-1

# Check Spot quota
aws service-quotas get-service-quota \
  --service-code ec2 \
  --quota-code L-34B43A08 \
  --region eu-north-1

# Current running On-Demand instance count
aws ec2 describe-instances \
  --region eu-north-1 \
  --filters "Name=instance-state-name,Values=running" \
  --query 'Reservations[*].Instances[*].[InstanceType,InstanceId]' \
  --output table

# Karpenter NodeClaim status
kubectl get nodeclaims -A
kubectl describe nodeclaim <name>   # shows AWS launch errors

Availability Zone Considerations (eu-north-1)

eu-north-1 has 3 AZs: eu-north-1a, eu-north-1b, eu-north-1c.

Karpenter selects AZs automatically. If InsufficientInstanceCapacity occurs in one AZ, Karpenter will retry in another. Configure your NodePool to allow all three:

requirements:
  - key: topology.kubernetes.io/zone
    operator: In
    values: ["eu-north-1a", "eu-north-1b", "eu-north-1c"]

Some instance families (e.g. newer Graviton generations) may not be available in all AZs. Check EC2 Instance Types by Region before committing to a family.