Install an OpenShift 4 cluster on Windows nodes
Big picture
Install an OpenShift 4 cluster on AWS with Calico on Windows nodes.
Value
Run Windows workloads on OpenShift 4 with Calico.
How to
Before you begin
Ensure that your environment meets the Calico system requirements.
Ensure that you have configured an AWS account appropriate for OpenShift 4, and have set up your AWS credentials. Note that the OpenShift installer supports a subset of AWS regions.
Ensure that you have a RedHat account. A RedHat account is required to obtain the pull secret necessary to provision an OpenShift cluster.
Ensure that you have installed the OpenShift installer v4.4 or later and OpenShift command line interface from cloud.redhat.com.
Ensure that you have generated a local SSH private key and have added it to your ssh-agent
Limitations
Due to an upstream issue, Windows pods can only be run in specific namespaces if you disable SCC. To do this, label the namespace with openshift.io/run-level: "1"
.
Create a configuration file for the OpenShift installer
First, create a staging directory for the installation. This directory will contain the configuration file, along with cluster state files, that OpenShift installer will create:
mkdir openshift-tigera-install && cd openshift-tigera-install
Now run OpenShift installer to create a default configuration file:
openshift-install create install-config
note
Refer to the OpenShift installer documentation for more information about the installer and any configuration changes required for your platform.
Once the installer has finished, your staging directory will contain the configuration file install-config.yaml
.
Update the configuration file to use Calico
Override the OpenShift networking to use Calico and update the AWS instance types to meet the system requirements:
sed -i 's/\(OpenShiftSDN\|OVNKubernetes\)/Calico/' install-config.yaml
Generate the install manifests
Now generate the Kubernetes manifests using your configuration file:
openshift-install create manifests
Download the Calico manifests for OpenShift and add them to the generated manifests directory:
mkdir calico
wget -qO- https://github.com/projectcalico/calico/releases/download/v3.26.4/ocp.tgz | tar xvz --strip-components=1 -C calico
cp calico/* manifests/
Configure VXLAN
Edit the Installation custom resource manifest manifests/01-cr-installation.yaml
so that it configures an OpenShift Calico cluster with VXLAN enabled and BGP disabled:
apiVersion: operator.tigera.io/v1
kind: Installation
metadata:
name: default
spec:
variant: Calico
calicoNetwork:
bgp: Disabled
ipPools:
- blockSize: 26
cidr: 10.128.0.0/14
encapsulation: VXLAN
natOutgoing: Enabled
nodeSelector: all()
Create the cluster
Start the cluster creation with the following command and wait for it to complete.
openshift-install create cluster
Once the above command is complete, you can verify Calico is installed by verifying the components are available with the following command.
oc get tigerastatus
note
To get more information, add -o yaml
to the above command.
Next, install calicoctl and ensure strict affinity is true:
kubectl patch ipamconfigurations default --type merge --patch='{"spec": {"strictAffinity": true}}'
Add Windows nodes to the cluster
Download the latest Windows Node Installer (WNI) binary wni
that matches your OpenShift minor version.
note
For OpenShift 4.6, use the latest wni for OpenShift 4.5. A wni binary for OpenShift 4.6 is not published yet.
Next, determine the AMI id corresponding to Windows Server 1903 (build 18317) or greater. wni
defaults to using Windows Server 2019 (build 10.0.17763) which does not include WinDSR support. One way to do this is by searching for AMI’s matching the string Windows_Server-1903-English-Core-ContainersLatest
in the Amazon EC2 console
Next, run wni
to add a Windows node to your cluster. Replace AMI_ID, AWS_CREDENTIALS_PATH, AWS_KEY_NAME and AWS_PRIVATE_KEY_PATH with your values:
chmod u+x wni
./wni aws create \
--image-id AMI_ID \
--kubeconfig openshift-tigera-install/auth/kubeconfig \
--credentials AWS_CREDENTIALS_PATH \
--credential-account default \
--instance-type m5a.large \
--ssh-key AWS_KEY_NAME \
--private-key AWS_PRIVATE_KEY_PATH
An example of running the above steps:
$ chmod u+x wni
$ ./wni aws create \
> --kubeconfig openshift-tigera-install/auth/kubeconfig \
> --credentials ~/.aws/credentials \
> --credential-account default \
> --instance-type m5a.large \
> --ssh-key test-key \
> --private-key /home/user/.ssh/test-key.pem
2020/10/05 12:52:51 kubeconfig source: /home/user/openshift-tigera-install/auth/kubeconfig
2020/10/05 12:52:59 Added rule with port 5986 to the security groups of your local IP
2020/10/05 12:52:59 Added rule with port 22 to the security groups of your local IP
2020/10/05 12:52:59 Added rule with port 3389 to the security groups of your local IP
2020/10/05 12:52:59 Using existing Security Group: sg-06d1de22807d5dc48
2020/10/05 12:57:30 External IP: 52.35.12.231
2020/10/05 12:57:30 Internal IP: 10.0.90.193
Get the administrator password
The wni
binary writes the instance details to the file windows-node-installer.json
. An example of the file:
{"InstanceIDs":["i-02e13d4cc76c13c83"],"SecurityGroupIDs":["sg-0a777565d64e1d2ef"]}
Use the instance ID from the file and the path of the private key used to create the instance to get the Administrator user’s password:
aws ec2 get-password-data --instance-id <instance id> --priv-launch-key <aws private key path>
Install Calico for Windows
Remote into the Windows node, open a Powershell window, and prepare the directory for Kubernetes files.
mkdir c:\k
Copy the Kubernetes kubeconfig file (default location: openshift-tigera-install/auth/kubeconfig), to the file c:\k\config.
Download the powershell script, install-calico-windows.ps1.
Invoke-WebRequest https://github.com/projectcalico/calico/releases/download/v3.26.4/install-calico-windows.ps1 -OutFile c:\install-calico-windows.ps1
Run the installation script, replacing the Kubernetes version with the version corresponding to your version of OpenShift.
c:\install-calico-windows.ps1 -KubeVersion <kube version> -ServiceCidr 172.30.0.0/16 -DNSServerIPs 172.30.0.10
note
Get the Kubernetes version with
oc version
and use only the major, minor, and patch version numbers. For example from a cluster that returns:$ oc version
Client Version: 4.5.3
Server Version: 4.5.14
Kubernetes Version: v1.18.3+5302882
You will use
1.18.3
:Install and start kube-proxy service. Execute following powershell script/commands.
C:\CalicoWindows\kubernetes\install-kube-services.ps1 -service kube-proxy
Start-Service -Name kube-proxy
Verify kube-proxy service is running.
Get-Service -Name kube-proxy
Configure kubelet
From the Windows node, download the Windows Machine Config Bootstrapper wmcb.exe
that matches your OpenShift minor version from Windows Machine Config Bootstrapper releases. For example, for OpenShift 4.5.x:
curl https://github.com/openshift/windows-machine-config-bootstrapper/releases/download/v4.5.2-alpha/wmcb.exe -o c:\wmcb.exe
note
For OpenShift 4.6, use the latest wmcb.exe for OpenShift 4.5. A wmcb.ex binary for OpenShift 4.6 is not published yet.
Next, we will download the worker.ign
file from the API server:
$apiServer = c:\k\kubectl --kubeconfig c:\k\config get po -n openshift-kube-apiserver -l apiserver=true --no-headers -o custom-columns=":metadata.name" | select -first 1
c:\k\kubectl --kubeconfig c:\k\config -n openshift-kube-apiserver exec $apiserver -- curl -ks https://localhost:22623/config/worker > c:\worker.ign
((Get-Content c:\worker.ign) -join "`n") + "`n" | Set-Content -NoNewline c:\worker.ign
Next, we run wmcb to configure the kubelet:
c:\wmcb.exe initialize-kubelet --ignition-file worker.ign --kubelet-path c:\k\kubelet.exe
note
The kubelet configuration installed by Windows Machine Config Bootstrapper includes --register-with-taints="os=Windows:NoSchedule"
which will require Windows pods to tolerate that taint.
Next, we make a copy of the kubeconfig because wmcb.exe
expects the kubeconfig to be the file c:\k\kubeconfig
. Then we configure kubelet to use Calico CNI:
cp c:\k\config c:\k\kubeconfig
c:\wmcb.exe configure-cni --cni-dir c:\k\cni --cni-config c:\k\cni\config\10-calico.conf
Finally, clean up the additional files created on the Windows node:
rm c:\k\kubeconfig,c:\wmcb.exe,c:\worker.ign
Exit the remote session to the Windows node and return to a shell to a Linux node.
We need to approve the CSR’s generated by the kubelet’s bootstrapping process. First, view the pending CSR’s:
oc get csr
For example:
$ oc get csr
NAME AGE SIGNERNAME REQUESTOR CONDITION
csr-55brx 4m32s kubernetes.io/kube-apiserver-client-kubelet system:admin Approved,Issued
csr-bmnfd 4m30s kubernetes.io/kubelet-serving system:node:ip-10-0-45-102.us-west-2.compute.internal Pending
csr-hwl89 5m1s kubernetes.io/kube-apiserver-client-kubelet system:serviceaccount:openshift-machine-config-operator:node-bootstrapper Pending
To approve the pending CSR’s:
oc get csr -o name | xargs oc adm certificate approve
For example:
$ oc get csr -o name | xargs oc adm certificate approve
certificatesigningrequest.certificates.k8s.io/csr-55brx approved
certificatesigningrequest.certificates.k8s.io/csr-bmnfd approved
certificatesigningrequest.certificates.k8s.io/csr-hwl89 approved
Finally, wait a minute or so and get all nodes:
$ oc get node -owide
If the Windows node registered itself successfully, it should appear in the list with a Ready status, ready to run Windows pods!
Next steps
Recommended - Security
- Secure Calico component communications
- Secure pods with Calico network policy
- If you are using Calico with Istio service mesh, get started here: Enable application layer policy