Tutorial: Security groups for pods

To deploy security groups for pods

If you're using security groups for Fargate pods only, and don't have any Amazon EC2 nodes in your cluster, skip to Deploy an example application.

1. Check your current Amazon VPC CNI plugin for Kubernetes version with the following command:

   kubectl describe daemonset aws-node --namespace kube-system | grep amazon-k8s-cni: | cut -d : -f 3
   

   The example output is as follows.

   v1.7.6

   If your Amazon VPC CNI plugin for Kubernetes version is earlier than 1.7.7, then update the plugin to version 1.7.7 or later. For more information, see Working with the Amazon VPC CNI plugin for Kubernetes Amazon EKS add-on.

2. Add the AmazonEKSVPCResourceController managed IAM policy to the cluster role that is associated with your Amazon EKS cluster. The policy allows the role to manage network interfaces, their private IP addresses, and their attachment and detachment to and from network instances.

   1. Retrieve the name of your cluster IAM role and store it in a variable. Replace my-cluster with the name of your cluster.

      cluster_role=$(aws eks describe-cluster --name my-cluster --query cluster.roleArn --output text | cut -d / -f 2)

   2. Attach the policy to the role.

      aws iam attach-role-policy --policy-arn arn:aws-cn:iam::aws:policy/AmazonEKSVPCResourceController --role-name $cluster_role
      

3. Enable the Amazon VPC CNI add-on to manage network interfaces for pods by setting the ENABLE_POD_ENI variable to true in the aws-node DaemonSet. Once this setting is set to true, for each node in the cluster the add-on adds a label with the value vpc.amazonaws.com/has-trunk-attached=true. The VPC resource controller creates and attaches one special network interface called a trunk network interface with the description aws-k8s-trunk-eni.

   kubectl set env daemonset aws-node -n kube-system ENABLE_POD_ENI=true
   

   Note

   The trunk network interface is included in the maximum number of network interfaces supported by the instance type. For a list of the maximum number of network interfaces supported by each instance type, see IP addresses per network interface per instance type in the Amazon EC2 User Guide for Linux Instances. If your node already has the maximum number of standard network interfaces attached to it then the VPC resource controller will reserve a space. You will have to scale down your running pods enough for the controller to detach and delete a standard network interface, create the trunk network interface, and attach it to the instance.

   You can see which of your nodes have aws-k8s-trunk-eni set to true with the following command. If No resources found is returned, then wait several seconds and try again. The previous step requires restarting the Amazon VPC CNI plugin for Kubernetes pods, which takes several seconds.

   kubectl get nodes -o wide -l vpc.amazonaws.com/has-trunk-attached=true
   

   Once the trunk network interface is created, pods are assigned secondary IP addresses from the trunk or standard network interfaces. The trunk interface is automatically deleted if the node is deleted.

   When you deploy a security group for a pod in a later step, the VPC resource controller creates a special network interface called a branch network interface with a description of aws-k8s-branch-eni and associates the security groups to it. Branch network interfaces are created in addition to the standard and trunk network interfaces attached to the node. If you are using liveness or readiness probes, then you also need to disable TCP early demux, so that the kubelet can connect to pods on branch network interfaces using TCP. To disable TCP early demux, run the following command:

   kubectl patch daemonset aws-node -n kube-system \
     -p '{"spec": {"template": {"spec": {"initContainers": [{"env":[{"name":"DISABLE_TCP_EARLY_DEMUX","value":"true"}],"name":"aws-vpc-cni-init"}]}}}}'
   

   Note

   If you're using 1.11.0 or later of the Amazon VPC CNI plugin for Kubernetes add-on and set POD_SECURITY_GROUP_ENFORCING_MODE=standard, as described in the next step, then you don't need to run the previous command.

4. If your cluster uses NodeLocal DNSCache, or you want to use Calico network policy with your pods that have their own security groups, or you have Kubernetes services of type NodePort and LoadBalancer using instance targets with an externalTrafficPolicy set to Local for pods that you want to assign security groups to, then you must be using version 1.11.0 or later of the Amazon VPC CNI plugin for Kubernetes add-on, and you must enable the following setting:

   kubectl set env daemonset aws-node -n kube-system POD_SECURITY_GROUP_ENFORCING_MODE=standard
   

   Important

   • Pod security group rules aren't applied to traffic between pods or between pods and services, such as kubelet or nodeLocalDNS, that are on the same node. Pods using different security groups on the same node can't communicate because they are configured in different subnets, and routing is disabled between these subnets.

   • Outbound traffic from pods to addresses outside of the VPC is network address translated to the IP address of the instance's primary network interface (unless you've also set AWS_VPC_K8S_CNI_EXTERNALSNAT=true). For this traffic, the rules in the security groups for the primary network interface are used, rather than the rules in the pod's security groups.

   • For this setting to apply to existing pods, you must restart the pods or the nodes that the pods are running on.

To deploy an example pod with a security group

1. Create a Kubernetes namespace to deploy resources to. You can replace my-namespace with the name of a namespace that you want to use.

   kubectl create namespace my-namespace

2. Deploy an Amazon EKS SecurityGroupPolicy to your cluster.

   1. Copy the following contents to your device. You can replace podSelector with serviceAccountSelector if you'd rather select pods based on service account labels. You must specify one selector or the other. An empty podSelector (example: podSelector: {}) selects all pods in the namespace. You can change my-role to the name of your role. An empty serviceAccountSelector selects all service accounts in the namespace. You can replace my-security-group-policy with a name for your SecurityGroupPolicy and my-namespace with the namespace that you want to create the SecurityGroupPolicy in.

      You must replace my_pod_security_group_id with the ID of an existing security group. If you don't have an existing security group, then you must create one. For more information, see Amazon EC2 security groups for Linux instances in the Amazon EC2 User Guide for Linux Instances. You can specify 1-5 security group IDs. If you specify more than one ID, then the combination of all the rules in all the security groups are effective for the selected pods.

      cat >my-security-group-policy.yaml <<EOF
      apiVersion: vpcresources.k8s.aws/v1beta1
      kind: SecurityGroupPolicy
      metadata:
        name: my-security-group-policy
        namespace: my-namespace
      spec:
        podSelector: 
          matchLabels:
            role: my-role
        securityGroups:
          groupIds:
            - my_pod_security_group_id
      EOF
      

      Important

      The security group or groups that you specify for your pods must meet the following criteria:

      • They must exist. If they don't exist, then, when you deploy a pod that matches the selector, your pod remains stuck in the creation process. If you describe the pod, you'll see an error message similar to the following one: An error occurred (InvalidSecurityGroupID.NotFound) when calling the CreateNetworkInterface operation: The securityGroup ID 'sg-05b1d815d1EXAMPLE' does not exist.

      • They must allow inbound communication from the security group applied to your nodes (for kubelet) over any ports that you've configured probes for.

      • They must allow outbound communication over TCP and UDP ports 53 to a security group assigned to the pods (or nodes that the pods run on) running CoreDNS. The security group for your CoreDNS pods must allow inbound TCP and UDP port 53 traffic from the security group that you specify.

      • They must have necessary inbound and outbound rules to communicate with other pods that they need to communicate with.

      • They must have rules that allow the pods to communicate with the Kubernetes control plane if you're using the security group with Fargate. The easiest way to do this is to specify the cluster security group as one of the security groups.

      Security group policies only apply to newly scheduled pods. They do not affect running pods.

   2. Deploy the policy.

      kubectl apply -f my-security-group-policy.yaml
      

3. Deploy a sample application with a label that matches the my-role value for podSelector that you specified in a previous step.

   1. Copy the following contents to your device. Replace the example values with your own and then run the modified command. If you replace my-role, make sure that it's the same as the value you specified for the selector in a previous step.

      cat >sample-application.yaml <<EOF
      apiVersion: apps/v1
      kind: Deployment
      metadata:
        name: my-deployment
        namespace: my-namespace
        labels:
          app: my-app
      spec:
        replicas: 4
        selector:
          matchLabels:
            app: my-app
        template:
          metadata:
            labels:
              app: my-app
              role: my-role
          spec:
            terminationGracePeriodSeconds: 120
            containers:
            - name: nginx
              image: public.ecr.aws/nginx/nginx:1.23
              ports:
              - containerPort: 80
      ---
      apiVersion: v1
      kind: Service
      metadata:
        name: my-app
        namespace: my-namespace
        labels:
          app: my-app
      spec:
        selector:
          app: my-app
        ports:
          - protocol: TCP
            port: 80
            targetPort: 80
      EOF
      

   2. Deploy the application with the following command. When you deploy the application, the Amazon VPC CNI plugin for Kubernetes matches the role label and the security groups that you specified in the previous step are applied to the pod.

      kubectl apply -f sample-application.yaml
      

4. View the pods deployed with the sample application. For the remainder of this topic, this terminal is referred to as TerminalA.

   kubectl get pods -n my-namespace -o wide
   

   The example output is as follows.

   NAME                             READY   STATUS    RESTARTS   AGE     IP               NODE                                            NOMINATED NODE   READINESS GATES
   my-deployment-5df6f7687b-4fbjm   1/1     Running   0          7m51s   192.168.53.48    ip-192-168-33-28.region-code.compute.internal   <none>           <none>
   my-deployment-5df6f7687b-j9fl4   1/1     Running   0          7m51s   192.168.70.145   ip-192-168-92-33.region-code.compute.internal   <none>           <none>
   my-deployment-5df6f7687b-rjxcz   1/1     Running   0          7m51s   192.168.73.207   ip-192-168-92-33.region-code.compute.internal   <none>           <none>
   my-deployment-5df6f7687b-zmb42   1/1     Running   0          7m51s   192.168.63.27    ip-192-168-33-28.region-code.compute.internal   <none>           <none>

   Note

   • If any pods are stuck in the Waiting state, then run kubectl describe pod my-deployment-xxxxxxxxxx-xxxxx -n my-namespace. If you see Insufficient permissions: Unable to create Elastic Network Interface., confirm that you added the IAM policy to the IAM cluster role in a previous step.

   • If any pods are stuck in the Pending state, confirm that your node instance type is listed in limits.go and that the product of the maximum number of branch network interfaces supported by the instance type multiplied times the number of nodes in your node group hasn't already been met. For example, an m5.large instance supports nine branch network interfaces. If your node group has five nodes, then a maximum of 45 branch network interfaces can be created for the node group. The 46th pod that you attempt to deploy will sit in Pending state until another pod that has associated security groups is deleted.

   If you run kubectl describe pod my-deployment-xxxxxxxxxx-xxxxx -n my-namespace and see a message similar to the following message, then it can be safely ignored. This message might appear when the Amazon VPC CNI plugin for Kubernetes tries to set up host networking and fails while the network interface is being created. The plugin logs this event until the network interface is created.

   Failed to create pod sandbox: rpc error: code = Unknown desc = failed to set up sandbox container "e24268322e55c8185721f52df6493684f6c2c3bf4fd59c9c121fd4cdc894579f" network for pod "my-deployment-5df6f7687b-4fbjm": networkPlugin
   cni failed to set up pod "my-deployment-5df6f7687b-4fbjm-c89wx_my-namespace" network: add cmd: failed to assign an IP address to container

   You can't exceed the maximum number of pods that can be run on the instance type. For a list of the maximum number of pods that you can run on each instance type, see eni-max-pods.txt on GitHub. When you delete a pod that has associated security groups, or delete the node that the pod is running on, the VPC resource controller deletes the branch network interface. If you delete a cluster with pods using pods for security groups, then the controller doesn't delete the branch network interfaces, so you'll need to delete them yourself. For information about how to delete network interfaces, see Delete a network interface in the Amazon EC2 User Guide for Linux Instances.

5. In a separate terminal, shell into one of the pods. For the remainder of this topic, this terminal is referred to as TerminalB. Replace 5df6f7687b-4fbjm with the ID of one of the pods returned in your output from the previous step.

   kubectl exec -it -n my-namespace my-deployment-5df6f7687b-4fbjm -- /bin/bash

6. From the shell in TerminalB, confirm that the sample application works.

   curl my-app
   

   The example output is as follows.

   <!DOCTYPE html>
   <html>
   <head>
   <title>Welcome to nginx!</title>
   ...

   You received the output because all pods running the application are associated with the security group that you created. That group contains a rule that allows all traffic between all pods that the security group is associated to. DNS traffic is allowed outbound from that security group to the cluster security group, which is associated with your nodes. The nodes are running the CoreDNS pods, which your pods did a name lookup to.

7. From TerminalA, remove the security group rules that allow DNS communication to the cluster security group from your security group. If you didn't add the DNS rules to the cluster security group in a previous step, then replace $my_cluster_security_group_id with the ID of the security group that you created the rules in.

   aws ec2 revoke-security-group-ingress --group-id $my_cluster_security_group_id --security-group-rule-ids $my_tcp_rule_id
   aws ec2 revoke-security-group-ingress --group-id $my_cluster_security_group_id --security-group-rule-ids $my_udp_rule_id
   

8. From TerminalB, attempt to access the application again.

   curl my-app
   

   The example output is as follows.

   curl: (6) Could not resolve host: my-app

   The attempt fails because the pod is no longer able to access the CoreDNS pods, which have the cluster security group associated to them. The cluster security group no longer has the security group rules that allow DNS communication from the security group associated to your pod.

   If you attempt to access the application using the IP addresses returned for one of the pods in a previous step, you still receive a response because all ports are allowed between pods that have the security group associated to them and a name lookup isn't required.

9. Once you've finished experimenting, you can remove the sample security group policy, application, and security group that you created. Run the following commands from TerminalA.

   kubectl delete namespace my-namespace
   aws ec2 revoke-security-group-ingress --group-id $my_pod_security_group_id --security-group-rule-ids $my_inbound_self_rule_id
   wait
   sleep 45s 
   aws ec2 delete-security-group --group-id $my_pod_security_group_id