Isolating Classic NLBs to edge worker nodes

Classic

In the following steps, you add the dedicated=edge label to worker nodes on each public or private VLAN in your cluster. This label is used to deploy your network load balancers (NLBs) to those worker nodes only. Both public and private NLBs can be deployed to edge worker nodes.

If you plan to use an existing worker pool, the pool must span all zones in your cluster and have at least two worker nodes per zone. You can label the worker pool with dedicated=edge by using the ibmcloud ks worker-pool label set command.

Before you begin

Ensure that you have the following IBM Cloud IAM roles:
- Any platform access role for the cluster
- Writer or Manager service access role for all namespaces
Log in to your account. If applicable, target the appropriate resource group. Set the context for your cluster.

Create a worker pool with the label dedicated=edge or add the label to one of your existing worker pools.

To create a worker pool you can use the worker-pool create classic command.

ibmcloud ks worker-pool create classic --name POOL_NAME --cluster CLUSTER --flavor FLAVOR --size-per-zone WORKERS_PER_ZONE --hardware ISOLATION --label dedicated=edge

To label an existing worker pool, you can use the worker-pool label set command.

ibmcloud ks worker-pool label set --cluster CLUSTER --worker-pool POOL --label dedicated=edge

Verify that the worker pool and worker nodes have the dedicated=edge label.
- To check the worker pool, run the get command.
```
ibmcloud ks worker-pool get --cluster CLUSTER_NAME_OR_ID --worker-pool WORKER_POOL_NAME_OR_ID
```
- To check individual worker nodes, review the Labels field of the output of the following command.
```
kubectl describe node <worker_node_private_IP>
```

Retrieve all existing NLBs in the cluster. In the output, note the namespace and name of each load balancer.

kubectl get services --all-namespaces | grep LoadBalancer

Example output:

kube-system           private-crc81nk5l10gfhdql4i3qg-nlb1   LoadBalancer   172.21.233.160   10.216.23.123    80:31345/TCP,443:32630/TCP   8d
kube-system           public-crc81nk5l10gfhdql4i3qg-nlb1    LoadBalancer   172.21.190.18    169.46.17.2      80:31345/TCP,443:32630/TCP   8d

Using the output from the previous step, run the following command for each NLB. This command redeploys the NLB to an edge worker node.

kubectl get service -n <namespace> <name> -o yaml | kubectl apply -f -

Example output:

service "private-crc81nk5l10gfhdql4i3qg-nlb1" configured
service "public-crc81nk5l10gfhdql4i3qg-nlb1" configured

To verify that networking workloads are restricted to edge nodes, confirm that the load balancers are scheduled onto the edge nodes and are not scheduled onto non-edge nodes.
- NLB pods
  1. Confirm that the NLB pods are deployed to edge nodes. Search for the external IP address of the load balancer service that is listed in the output of the previous step. Replace the periods (.) with hyphens (-). In the following example for the crc81nk5l10gfhdql4i3qg, the NLB has an external IP address of 169.46.17.2.
```
kubectl describe nodes -l dedicated=edge | grep "169-46-17-2"
```
    Example output:
```
ibm-system                 ibm-cloud-provider-ip-169-46-17-2-76fcb4965d-wz6dg                 5m (0%)       0 (0%)      10Mi (0%)        0 (0%)
ibm-system                 ibm-cloud-provider-ip-169-46-17-2-76fcb4965d-2z64r                 5m (0%)       0 (0%)      10Mi (0%)        0 (0%)
```
  2. Confirm that no NLB pods are deployed to non-edge nodes. Example for the public-crc81nk5l10gfhdql4i3qg-nlb1 NLB that has an external IP address of 169.46.17.2:
```
kubectl describe nodes -l dedicated!=edge | grep "169-46-17-2"
```
    - If the NLB pods are correctly deployed to edge nodes, no NLB pods are returned. Your NLBs are successfully rescheduled onto only edge worker nodes.
    - If NLB pods are returned, continue to the next step.
If NLB pods are still deployed to non-edge nodes, you can delete the pods so that they redeploy to edge nodes.

Delete only one pod at a time, and verify that the pod is rescheduled onto an edge node before you delete other pods.
1. Delete a pod. For example, if one of the public-crc81nk5l10gfhdql4i3qg-alb1 NLB pods did not schedule to an edge node:
```
kubectl delete pod ibm-cloud-provider-ip-169-46-17-2-76fcb4965d-wz6dg -n ibm-system
```
2. Verify that the pod is rescheduled onto an edge worker node. Rescheduling is automatic, but might take a few minutes. Example for the public-crc81nk5l10gfhdql4i3qg-alb1 NLB that has an external IP address of 169.46.17.2:
```
kubectl describe nodes -l dedicated=edge | grep "169-46-17-2"
```
  Example output:
```
ibm-system                 ibm-cloud-provider-ip-169-46-17-2-76fcb4965d-wz6dg                 5m (0%)       0 (0%)      10Mi (0%)        0 (0%)
ibm-system                 ibm-cloud-provider-ip-169-46-17-2-76fcb4965d-2z64r                 5m (0%)       0 (0%)      10Mi (0%)        0 (0%)
```

You labeled worker nodes in a worker pool with dedicated=edge and redeployed all the existing NLBs to the edge nodes. All subsequent NLBs that are added to the cluster are also deployed to an edge node in your edge worker pool. Next, prevent other workloads from running on edge worker nodes and block inbound traffic to NodePorts on worker nodes.