Amazon EKS troubleshooting

This chapter covers some common errors that you may see while using Amazon EKS and how to work around them. If you need to troubleshoot specific Amazon EKS areas, see the separate Troubleshooting IAM, Troubleshooting issues in Amazon EKS Connector, and Troubleshooting issues in ADOT Amazon EKS add-on topics.

Insufficient capacity

If you receive the following error while attempting to create an Amazon EKS cluster, then one of the Availability Zones you specified doesn't have sufficient capacity to support a cluster.

Cannot create cluster 'example-cluster' because region-1d, the targeted Availability Zone, does not currently have sufficient capacity to support the cluster. Retry and choose from these Availability Zones: region-1a, region-1b, region-1c

Retry creating your cluster with subnets in your cluster VPC that are hosted in the Availability Zones returned by this error message.

Nodes fail to join cluster

There are a few common reasons that prevent nodes from joining the cluster:

To identify and troubleshoot common causes that prevent worker nodes from joining a cluster, you can use the AWSSupport-TroubleshootEKSWorkerNode runbook. For more information, see AWSSupport-TroubleshootEKSWorkerNode in the Amazon Systems Manager Automation runbook reference.

Unauthorized or access denied (kubectl)

If you receive one of the following errors while running kubectl commands, then your kubectl is not configured properly for Amazon EKS or the IAM principal credentials that you're using don't map to a Kubernetes RBAC user with sufficient permissions in your Amazon EKS cluster.

This could be because the cluster was created with credentials for one IAM principal and kubectl is using credentials for a different IAM principal.

When an Amazon EKS cluster is created, the IAM principal that creates the cluster is added to the Kubernetes RBAC authorization table as the administrator (with system:masters permissions). Initially, only that principal can make calls to the Kubernetes API server using kubectl . For more information, see Enabling IAM principal access to your cluster. If you use the console to create the cluster, make sure that the same IAM credentials are in the Amazon SDK credential chain when you are running kubectl commands on your cluster.

If you install and configure the Amazon CLI, you can configure the IAM credentials you use. For more information, see Configuring the Amazon CLI in the Amazon Command Line Interface User Guide.

If you assumed a role to create the Amazon EKS cluster, you must ensure that kubectl is configured to assume the same role. Use the following command to update your kubeconfig file to use an IAM role. For more information, see Creating or updating a kubeconfig file for an Amazon EKS cluster.

aws eks update-kubeconfig \
    --region region-code \
    --name my-cluster \
    --role-arn arn:aws-cn:iam::111122223333:role/role_name

To map an IAM principal to a Kubernetes RBAC user, see Enabling IAM principal access to your cluster.

aws-iam-authenticator Not found

If you receive the error "aws-iam-authenticator": executable file not found in $PATH, then your kubectl is not configured for Amazon EKS. For more information, see Installing aws-iam-authenticator.

hostname doesn't match

Your system's Python version must be 2.7.9 or later. Otherwise, you receive hostname doesn't match errors with Amazon CLI calls to Amazon EKS. For more information, see What are "hostname doesn't match" errors? in the Python Requests FAQ.

getsockopt: no route to host

Docker runs in the 172.17.0.0/16 CIDR range in Amazon EKS clusters. We recommend that your cluster's VPC subnets do not overlap this range. Otherwise, you will receive the following error:

Error: : error upgrading connection: error dialing backend: dial tcp 172.17.<nn>.<nn>:10250: getsockopt: no route to host

Instances failed to join the kubernetes cluster

If you receive the error Instances failed to join the kubernetes cluster in the Amazon Web Services Management Console, ensure that either the cluster's private endpoint access is enabled, or that you have correctly configured CIDR blocks for public endpoint access. For more information, see Amazon EKS cluster endpoint access control.

Managed node group error codes

If your managed node group encounters a hardware health issue, Amazon EKS returns an error code to help you to diagnose the issue. These health checks don't detect software issues because they are based on Amazon EC2 health checks. The following list describes the error codes.

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: eks:node-manager
rules:
- apiGroups:
  - ''
  resources:
  - pods
  verbs:
  - get
  - list
  - watch
  - delete
- apiGroups:
  - ''
  resources:
  - nodes
  verbs:
  - get
  - list
  - watch
  - patch
- apiGroups:
  - ''
  resources:
  - pods/eviction
  verbs:
  - create               
            

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: eks:node-manager
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
  name: eks:node-manager
subjects:
- apiGroup: rbac.authorization.k8s.io
  kind: User
  name: eks:node-manager

kubectl describe clusterrole eks:node-manager

kubectl describe clusterrolebinding eks:node-manager

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: eks:node-manager
rules:
- apiGroups:
  - ''
  resources:
  - pods
  verbs:
  - get
  - list
  - watch
  - delete
- apiGroups:
  - ''
  resources:
  - nodes
  verbs:
  - get
  - list
  - watch
  - patch
- apiGroups:
  - ''
  resources:
  - pods/eviction
  verbs:
  - create
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: eks:node-manager
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
  name: eks:node-manager
subjects:
- apiGroup: rbac.authorization.k8s.io
  kind: User
  name: eks:node-manager

kubectl apply -f eks-node-manager-role.yaml

Not authorized for images

One potential cause of a Not authorized for images error message is using a private Amazon EKS Windows AMI to launch Windows managed node groups. After releasing new Windows AMIs, Amazon makes Windows AMIs that are older than three months private within 10 days. If your managed node group is using a private Amazon EKS Windows AMI, consider updating your Windows managed node group. For more information, see Patches, security updates, and AMI IDs in the Amazon EC2 User Guide for Windows Instances.

CNI log collection tool

The Amazon VPC CNI plugin for Kubernetes has its own troubleshooting script that is available on nodes at /opt/cni/bin/aws-cni-support.sh. You can use the script to collect diagnostic logs for support cases and general troubleshooting.

Use the following command to run the script on your node:

sudo bash /opt/cni/bin/aws-cni-support.sh

curl -O https://raw.githubusercontent.com/awslabs/amazon-eks-ami/master/log-collector-script/linux/eks-log-collector.sh
sudo bash eks-log-collector.sh

The script collects the following diagnostic information. The CNI version that you have deployed can be earlier than the script version.

      This is version 0.6.1. New versions can be found at https://github.com/awslabs/amazon-eks-ami

Trying to collect common operating system logs... 
Trying to collect kernel logs... 
Trying to collect mount points and volume information... 
Trying to collect SELinux status... 
Trying to collect iptables information... 
Trying to collect installed packages... 
Trying to collect active system services... 
Trying to collect Docker daemon information... 
Trying to collect kubelet information... 
Trying to collect L-IPAMD information... 
Trying to collect sysctls information... 
Trying to collect networking information... 
Trying to collect CNI configuration information... 
Trying to collect running Docker containers and gather container data... 
Trying to collect Docker daemon logs... 
Trying to archive gathered information... 

	Done... your bundled logs are located in /var/log/eks_i-0717c9d54b6cfaa19_2020-03-24_0103-UTC_0.6.1.tar.gz

The diagnostic information is collected and stored at:

/var/log/eks_i-0717c9d54b6cfaa19_2020-03-24_0103-UTC_0.6.1.tar.gz

Container runtime network not ready

You may receive a Container runtime network not ready error and authorization errors similar to the following:

4191 kubelet.go:2130] Container runtime network not ready: NetworkReady=false reason:NetworkPluginNotReady message:docker: network plugin is not ready: cni config uninitialized
4191 reflector.go:205] k8s.io/kubernetes/pkg/kubelet/kubelet.go:452: Failed to list *v1.Service: Unauthorized
4191 kubelet_node_status.go:106] Unable to register node "ip-10-40-175-122.ec2.internal" with API server: Unauthorized
4191 reflector.go:205] k8s.io/kubernetes/pkg/kubelet/kubelet.go:452: Failed to list *v1.Service: Unauthorized

The errors are most likely because the Amazon IAM Authenticator (aws-auth) configuration map isn't applied to the cluster. The configuration map provides the system:bootstrappers and system:nodes Kubernetes RBAC permissions for nodes to register to the cluster. To apply the configuration map to your cluster, see Apply the aws-authConfigMap to your cluster.

The authenticator doesn't recognize a Role ARN if it includes a path other than /, such as the following example:

arn:aws-cn:iam::111122223333:role/development/apps/prod-iam-role-NodeInstanceRole-621LVEXAMPLE

When specifying a Role ARN in the configuration map that includes a path other than /, you must drop the path. The previous ARN should be specified as the following:

arn:aws-cn:iam::111122223333:role/prod-iam-role-NodeInstanceRole-621LVEXAMPLE

TLS handshake timeout

When a node is unable to establish a connection to the public API server endpoint, you may see an error similar to the following error.

server.go:233] failed to run Kubelet: could not init cloud provider "aws": error finding instance i-1111f2222f333e44c: "error listing Amazon instances: \"RequestError: send request failed\\ncaused by: Post  net/http: TLS handshake timeout\""

The kubelet process will continually respawn and test the API server endpoint. The error can also occur temporarily during any procedure that performs a rolling update of the cluster in the control plane, such as a configuration change or version update.

To resolve the issue, check the route table and security groups to ensure that traffic from the nodes can reach the public endpoint.

InvalidClientTokenId

If you're using IAM roles for service accounts for a pod or DaemonSet deployed to a cluster in a China Amazon Web Services Region, and haven't set the AWS_DEFAULT_REGION environment variable in the spec, the pod or DaemonSet may receive the following error:

An error occurred (InvalidClientTokenId) when calling the GetCallerIdentity operation: The security token included in the request is invalid

To resolve the issue, you need to add the AWS_DEFAULT_REGION environment variable to your pod or DaemonSet spec, as shown in the following example pod spec.

apiVersion: v1
kind: Pod
metadata:
  name: envar-demo
  labels:
    purpose: demonstrate-envars
spec:
  containers:
  - name: envar-demo-container
    image: gcr.io/google-samples/node-hello:1.0
    env:
    - name: AWS_DEFAULT_REGION
      value: "region-code"

VPC admission webhook certificate expiration

If the certificate used to sign the VPC admission webhook expires, the status for new Windows pod deployments stays at ContainerCreating.

To resolve the issue if you have legacy Windows support on your data plane, see Renewing the VPC admission webhook certificate. If your cluster and platform version are later than a version listed in the Windows support prerequisites, then we recommend that you remove legacy Windows support on your data plane and enable it for your control plane. Once you do, you don't need to manage the webhook certificate. For more information, see Enabling Windows support for your Amazon EKS cluster.

Node groups must match Kubernetes version before updating control plane

Before you update a control plane to a new Kubernetes version, the minor version of the managed and Fargate nodes in your cluster must be the same as the version of your control plane's current version. The EKS update-cluster-version API rejects requests until you update all EKS managed nodes to the current cluster version. EKS provides APIs to update managed nodes. For information on updating managed node group Kubernetes versions, see Updating a managed node group. To update the version of a Fargate node, delete the pod that's represented by the node and redeploy the pod after you update your control plane. For more information, see Updating an Amazon EKS cluster Kubernetes version.

When launching many nodes, there are Too Many Requests errors

If you launch many nodes simultaneously, you may see an error message in the Amazon EC2 user data execution logs that says Too Many Requests. This can occur because the control plane is being overloaded with describeCluster calls. The overloading results in throttling, nodes failing to run the bootstrap script, and nodes failing to join the cluster altogether.

Make sure that --apiserver-endpoint, --b64-cluster-ca, and --dns-cluster-ip arguments are being passed to the worker node bootstrap script. When including these arguments, there's no need for the bootstrap script to make a describeCluster call, which helps prevent the control plane from being overloaded. For more information, see Provide user data to pass arguments to the bootstrap.sh file included with an Amazon EKS optimized Linux/Bottlerocket AMI.

HTTP 401 unauthorized error response on Kubernetes API server requests

You see these errors if a pod's service account token has expired on a 1.21 or later cluster.

Your Amazon EKS version 1.21 or later cluster's Kubernetes API server rejects requests with tokens older than 90 days. In previous Kubernetes versions, tokens did not have an expiration. This means that clients that rely on these tokens must refresh them within an hour. To prevent the Kubernetes API server from rejecting your request due to an invalid token, the Kubernetes client SDK version used by your workload must be the same, or later than the following versions:

You can identify all existing pods in your cluster that are using stale tokens. For more information, see Kubernetes service accounts.

Amazon EKS platform version is more than two versions behind the current platform version

This can happen when Amazon EKS isn't able to automatically update your cluster's platform version. Though there are many causes for this, some of the common causes follow. If any of these problems apply to your cluster, it may still function, it's platform version just won't be updated by Amazon EKS.

Problem

The cluster IAM role was deleted – This role was specified when the cluster was created. You can see which role was specified with the following command. Replace my-cluster with the name of your cluster.

aws eks describe-cluster --name my-cluster --query cluster.roleArn --output text | cut -d / -f 2

The example output is as follows.

eksClusterRole

Solution

Create a new cluster IAM role with the same name.

Problem

A subnet specified during cluster creation was deleted – The subnets to use with the cluster were specified during cluster creation. You can see which subnets were specified with the following command. Replace my-cluster with the name of your cluster.

aws eks describe-cluster --name my-cluster --query cluster.resourcesVpcConfig.subnetIds

The example output is as follows.

[
"subnet-EXAMPLE1",
"subnet-EXAMPLE2"
]

Solution

Confirm whether the subnet IDs exist in your account.

vpc_id=$(aws eks describe-cluster --name my-cluster --query cluster.resourcesVpcConfig.vpcId --output text)
aws ec2 describe-subnets --filters "Name=vpc-id,Values=$vpc_id" --query "Subnets[*].SubnetId"

The example output is as follows.

[
"subnet-EXAMPLE3",
"subnet-EXAMPLE4"
]

If the subnet IDs returned in the output don't match the subnet IDs that were specified when the cluster was created, then if you want Amazon EKS to update the cluster, you might need to create a new cluster. This is because if you specified more than two subnets when you created your cluster, Amazon EKS randomly selects subnets that you specified to create new elastic network interfaces in. These network interfaces enable the control plane to communicate with your nodes. Amazon EKS won't update the cluster if the subnet it selects doesn't exist. You have no control over which of the subnets that you specified at cluster creation that Amazon EKS chooses to create a new network interface in. You can't create new subnets for your cluster to use after cluster creation.

When you initiate a Kubernetes version update for your cluster, the update can fail for the same reason.

Problem

A security group specified during cluster creation was deleted – If you specified security groups during cluster creation, you can see their IDs with the following command. Replace my-cluster with the name of your cluster.

aws eks describe-cluster --name my-cluster --query cluster.resourcesVpcConfig.securityGroupIds

The example output is as follows.

[
    "sg-EXAMPLE1"
]

If [] is returned, then no security groups were specified when the cluster was created and a missing security group isn't the problem. If security groups are returned, then confirm that the security groups exist in your account.

Solution

Confirm whether these security groups exist in your account.

vpc_id=$(aws eks describe-cluster --name my-cluster --query cluster.resourcesVpcConfig.vpcId --output text)
aws ec2 describe-security-groups --filters "Name=vpc-id,Values=$vpc_id" --query "SecurityGroups[*].GroupId"

The example output is as follows.

[
"sg-EXAMPLE2"
]

If the security group IDs returned in the output don't match the security group IDs that were specified when the cluster was created, then if you want Amazon EKS to update the cluster, you need to create a new cluster. Amazon EKS won't update a cluster if the security group IDs specified at cluster creation don't exist. You can't specify different security groups after cluster creation.

When you initiate a Kubernetes version update for your cluster, the update can fail for the same reason.