Skip to content

MaterializeInc/k8s-eip-operator

Repository files navigation

⚠️⚠️ WARNING! ⚠️⚠️ The Materialize K8s-eip-operator will soon be archived and is no longer be under active development.

k8s-eip-operator

Manage external connections to Kubernetes pods or nodes using AWS Elastic IPs (EIPs).

This operator manages the following:

  • Creation of an EIP Custom Resource Definition in K8S.
  • Creation/destruction of EIP allocations in AWS.
    • These AWS EIPs will be tagged with the K8S EIP uid they will be assigned to (eip.materialize.cloud/eip_uid) for later identification.
  • Association/disassociation of the EIP to the Elastic Network Interface (ENI) on the private IP of the pod or node.
  • Annotation of external-dns.alpha.kubernetes.io/target on the pod, so that the external-dns operator knows to use the IP of the EIP instead of the external IP of the primary ENI on the host.

Prerequisites

  • You must be using the AWS VPC CNI plugin. This is the default for AWS EKS, so you probably don't need to change anything here.

  • You must set AWS_VPC_K8S_CNI_EXTERNALSNAT=true on the aws-node daemonset.

    kubectl set env daemonset -n kube-system aws-node AWS_VPC_K8S_CNI_EXTERNALSNAT=true
    

    See the AWS external-snat docs for more details.

  • For external-dns support, you must be using a version with headless ClusterIp support, either by waiting until this PR is merged or by using a fork with it already included.

Installation

  1. Create an AWS IAM role with the following policy:
    {
        "Version": "2012-10-17",
        "Statement": [
            {
                "Action": [
                    "ec2:AllocateAddress",
                    "ec2:ReleaseAddress",
                    "ec2:DescribeAddresses",
                    "ec2:AssociateAddress",
                    "ec2:DisassociateAddress",
                    "ec2:DescribeNetworkInterfaces",
                    "ec2:CreateTags",
                    "ec2:DeleteTags",
                    "ec2:DescribeInstances",
                    "ec2:ModifyNetworkInterfaceAttribute",
                    "servicequotas:GetServiceQuota"
                ],
                "Effect": "Allow",
                "Resource": "*"
            }
        ]
    }
  2. Create a K8S ServiceAccount.
    apiVersion: v1
    kind: ServiceAccount
    metadata:
      name: eip-operator
      annotations:
        eks.amazonaws.com/role-arn: arn:aws:iam::ACCOUNT-ID:role/IAM-SERVICE-ROLE-NAME
  3. Create a K8S ClusterRole.
    apiVersion: rbac.authorization.k8s.io/v1
    kind: ClusterRole
    metadata:
      name: eip-operator
    rules:
    - apiGroups: [""]
      resources: ["pods", "pods/status"]
      verbs: ["get", "watch", "list", "update", "patch"]
    - apiGroups: [""]
      resources: ["nodes", "nodes/status"]
      verbs: ["get"]
  4. Create a K8S ClusterRoleBinding.
    apiVersion: rbac.authorization.k8s.io/v1
    kind: ClusterRoleBinding
    metadata:
      name: eip-operator-viewer
    roleRef:
      apiGroup: rbac.authorization.k8s.io
      kind: ClusterRole
      name: eip-operator
    subjects:
    - kind: ServiceAccount
      name: eip-operator
      namespace: default
  5. Create a K8S Deployment. You must specify the CLUSTER_NAME environment variable. NAMESPACE and DEFAULT_TAGS are optional. If you do not set the NAMESPACE environment variable, the eip-operator will operate on all namespaces. Even in this global mode, the eip and pod must be in the same namespace.
    apiVersion: apps/v1
    kind: Deployment
    metadata:
      name: eip-operator
    spec:
      strategy:
        type: Recreate
      selector:
        matchLabels:
          app: eip-operator
      template:
        metadata:
          labels:
            app: eip-operator
        spec:
          containers:
          - name: eip-operator
            image: materialize/k8s-eip-operator:latest
            env:
            - name: NAMESPACE
              value: default
            - name: CLUSTER_NAME
              value: my-example-cluster
            - name: DEFAULT_TAGS
              value: '{"tag1": "value1", "tag2": "value2"}'

Usage

A. If you want your EIP to survive beyond the lifetime of the pod (ie: for static reservations when updating a statefulset):

Instantiate an Eip Kubernetes object, specifying the podName in the spec.

apiVersion: "materialize.cloud/v2"
kind: Eip
metadata:
  name: my-new-eip
spec:
  selector:
    pod:
      podName: my-pod

Add the eip.materialize.cloud/manage=true label to the pod with name matching the podName specified above.

B. If you don't care about getting a new IP if the pod gets recreated:

No need to manually create the Eip in this case, the operator can do it for you.

Add both the eip.materialize.cloud/manage=true and eip.materialize.cloud/autocreate_eip=true labels to your pod.

Do NOT manually create the Eip Kubernetes object if setting the eip.materialize.cloud/autocreate_eip=true label, or the two objects will fight over your pod.

C. If you want to attach an EIP to attach to a node directly instead of a pod, specify a node selector instead in the Eip Kubernetes resource:
apiVersion: "materialize.cloud/v2"
kind: Eip
metadata:
  name: my-new-eip
spec:
  selector:
    node:
      selector:
        some-label: some-value
        some-other-label: some-other-value

The node selector should contain a set of labels which should match a single node - if multiple nodes are matched, the EIP will be attached to one of them arbitrarily.

Add the eip.materialize.cloud/manage=true label to the node whose labels match the labels in the selector.

Cilium Support

If using Cilium in ENI mode, you can still use this operator, but you will need to disable masquerade for pods with EIPs assigned. Cilium (as of 1.12.0) does not seem to support configuring masquerade on a per-pod basis, so you will need to do one of the following:

A. Disable masquerade globally.
B. Run a privileged daemonset in the host network to inject ip rules for pods managed by the eip-operator.
apiVersion: apps/v1
kind: DaemonSet
metadata:
  name: cilium-eip-no-masquerade-agent
spec:
  selector:
    matchLabels:
      app: cilium-eip-no-masquerade-agent
  template:
    metadata:
      labels:
        app: cilium-eip-no-masquerade-agent
    spec:
      containers:
      - command:
        - ./cilium-eip-no-masquerade-agent
        env:
        - name: RUST_LOG
          value: INFO
        image: materialize/k8s-eip-operator
        name: eip-operator
        securityContext:
          privileged: true
      dnsPolicy: ClusterFirstWithHostNet
      hostNetwork: true
      restartPolicy: Always
      serviceAccount: eip-operator
      serviceAccountName: eip-operator

Testing

  • Install the KUTTL testing tool
  • Run ./bin/run-tests

OpenTelemetry Integration

We now have support for sending traces using the OpenTelemetry OTLP format. This is configured through environment variables:

OPENTELEMETRY_ENDPOINT is the endpoint to send the logs to.

OPENTELEMETRY_HEADERS is a json formatted map of key/value pairs to be included in the GRPC request headers.

OPENTELEMETRY_TOPLEVEL_FIELDS is a json formatted map of key/value pairs to be included in all trace spans emitted from the service.

OPENTELEMETRY_LEVEL_TARGETS is a tracing crate level filter. Defaults to "DEBUG".

OPENTELEMETRY_SAMPLE_RATE is a float value controlling the trace sample rate. Default is 0.05.

References