Deploying AWS Private Cluster with Hypershift

aws-private-clusters.md

Deploying AWS Private Cluster with Hypershift

Create a hypershift-operator IAM user in the management account

NOTE: An IAM Role can also be used but this is the simpliest method to document

Create the policy document

# cat << EOF >> policy.json
{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
                "ec2:CreateVpcEndpointServiceConfiguration",
                "ec2:DescribeVpcEndpointServiceConfigurations",
                "ec2:DeleteVpcEndpointServiceConfigurations",
                "elasticloadbalancing:DescribeLoadBalancers"
            ],
            "Resource": "*"
        }
    ]
}
EOF

Create the policy

# aws iam create-policy --policy-name=hypershift-operator-policy --policy-document=file://policy.json
{
    "Policy": {
        "PolicyName": "hypershift-operator-policy",
        "PolicyId": "ANPA3554CE3GL4MJ3DE7A",
        "Arn": "arn:aws:iam::820196288204:policy/hypershift-operator-policy",
        "Path": "/",
        "DefaultVersionId": "v1",
        "AttachmentCount": 0,
        "PermissionsBoundaryUsageCount": 0,
        "IsAttachable": true,
        "CreateDate": "2021-11-30T16:24:56+00:00",
        "UpdateDate": "2021-11-30T16:24:56+00:00"
    }
}

Create the hypershift-operator user

# aws iam create-user --user-name=hypershift-operator
{
    "User": {
        "Path": "/",
        "UserName": "hypershift-operator",
        "UserId": "AIDA3554CE3GEAHJKGDUP",
        "Arn": "arn:aws:iam::820196288204:user/hypershift-operator",
        "CreateDate": "2021-11-30T16:26:37+00:00"
    }
}

Attach user policy (use policy-arn from create-policy output)

aws iam attach-user-policy --user-name=hypershift-operator --policy-arn=arn:aws:iam::820196288204:policy/hypershift-operator-policy

Create access key

# aws iam create-access-key --user-name=hypershift-operator
{
    "AccessKey": {
        "UserName": "hypershift-operator",
        "AccessKeyId": "<secret>",
        "Status": "Active",
        "SecretAccessKey": "<secret>",
        "CreateDate": "2021-11-30T16:31:19+00:00"
    }
}

Create the credentials file (copy AccessKeyId and SecretAccessKey from create-access-key output)

# cat << EOF >> aws-private-creds
[default]
aws_access_key_id = <secret>
aws_secret_access_key = <secret>
EOF

Install hypershift-operator

# ./bin/hypershift install \
  --private-platform=AWS \
  --aws-private-creds=aws-private-creds \
  --aws-private-region=us-west-1 \
  --oidc-storage-provider-s3-bucket-name=sjenning-dev \
  --oidc-storage-provider-s3-credentials=$HOME/hypershift/aws-creds \
  --oidc-storage-provider-s3-region=us-west-1

Create the HostedCluster

Create the manifests

# ./bin/hypershift create cluster aws \
  --aws-creds=$HOME/hypershift/aws-creds \
  --pull-secret=$HOME/hypershift/pull-secret \
  --ssh-key=$HOME/.ssh/yubikey5.pub \
  --region=us-west-1 \
  --infra-id=sjenning-dev \
  --base-domain=hypershift.sjennings.me \
  --render > cluster.yaml

Edit the HostedCluster with endpointAccess: Private

# vi cluster.yaml
...
apiVersion: hypershift.openshift.io/v1alpha1
kind: HostedCluster
metadata:
  name: example
  namespace: clusters
spec:
...
  platform:
    aws:
      endpointAccess: Private <--

Create the HostedCluster resources

# oc create -f cluster.yaml

Observe HostedCluster rollout

Watch AWSEndpointServices

AWSEndpointService CRs are created when the following services are created

• kube-apiserver-private in the HCP namespace
• router-$hcpNamespace in the openshift-ingress namespace

# oc get awsendpointservice
NAME                      AGE
kube-apiserver-private    64m
router-clusters-example   64m

The AWSEndpointServices are acted upon by two different controllers, one in the hypershift-operator (HO) which creates the Endpoint Services in the management account, and one in the control-plane-operator (CPO) which creates the Endpoints and $hcpName.hypershift.local DNS zone that contains CNAME records for api.$hcpName.hypershift.local and *.apps.$hcpName.hypershift.local whose value is the DNS name of the corresponding Endpoint.

The AWSEndpointServices will go through several states:

• LB not active
  • AWS is still activiating the NLB for the service type LoadBalancer)
  • takes 2-3m
• EndpointService is created by HO
  • status.endpointServiceName is set and EndpointServiceAvailable condition is True
  • immediate
• Endpoint and DNS records are created CPO
  • status.{endpointID,dnsZoneID,dnsName} are set and EndpointAvailable condition is True
  • takes ~30s

Access the private guest cluster

Start a bastion

# ./bin/hypershift create bastion aws --aws-creds=$HOME/hypershift/aws-creds --infra-id=sjenning-dev --region=us-west-1 --ssh-key-file=$HOME/.ssh/yubikey5.pub
2021-11-30T09:54:36-06:00       INFO    Created security group  {"name": "sjenning-dev-bastion-sg", "id": "sg-0e07fa54fdd79c8f7"}
2021-11-30T09:54:37-06:00       INFO    Created key pair        {"id": "key-0471c1519618e1cb6", "name": "sjenning-dev-bastion"}
2021-11-30T09:54:39-06:00       INFO    Created ec2 instance    {"id": "i-0eb5316d0ecd7d1f6", "name": "sjenning-dev-bastion"}
2021-11-30T09:55:24-06:00       INFO    Successfully created bastion    {"id": "i-0eb5316d0ecd7d1f6", "publicIP": "54.177.195.110"}

Get private IP of nodes in the NodePool

# aws ec2 describe-instances --filter="Name=tag:kubernetes.io/cluster/sjenning-dev,Values=owned" | jq '.Reservations[] | .Instances[] | select(.PublicDnsName=="") | .PrivateIpAddress'
"10.0.133.98"
"10.0.136.236"

Dump kubeconfig for the HostedCluster to copy to the node

# ./bin/hypershift create kubeconfig
apiVersion: v1
clusters:
- cluster:
    certificate-authority-data: <redacted>
    server: https://api.example.hypershift.local:6443
  name: clusters-example
...

SSH into one of the nodes via the bastion (bastion IP from the create bastion output)

# ssh -o ProxyCommand="ssh ec2-user@54.177.195.110 -W %h:%p" core@10.0.133.98

Copy the kubeconfig contents to a file on the node

# cat << EOF >> kubeconfig
<paste kubeconfig contents>
EOF
# export KUBECONFIG=$PWD/kubeconfig

Observe guest cluster status

# oc get clusteroperators
...
# oc get clusterversion
...