Kubernetes

How To Enable or Disable Load Balancer for Master Nodes

A default state for the Kubernetes cluster in Creodias OpenStack Magnum hosting is to have no load balancer set up in advance. You can decide to have a load balancer created together with the basic Kubernetes cluster by checking on option Enable Load Balancer for Master Nodes in window Network when creating a cluster through Horizon interface. (See Prerequisite No. 3 for the complete procedure.)

The check box to enable load balancer for master nodes has two completely different meanings when checked and not checked.

Checked state

If checked, the load balancer for master nodes will be created. If you specified two or more master nodes in previous screens, then this field must be checked.

Regardless of the number of master nodes you have specified, checking this field on yields higher chances of successfully creating the Kubernetes cluster.

Non-checked state

If you accept the default state of unchecked, no load balancer will be created. However, without any load balancer “in front” of the cluster, the cluster API is being exposed only within the Kubernetes network. You save on the existence of the load balancer but the direct connection from local machine to the cluster is lost.

One Master Node, No Load Balancer and the Problem It All Creates

To show exactly what the problem is, use

• Prerequisite No. 2 to install openstack client for the local machine, so that you can use the openstack command.

• Then use Prerequisite No. 4 to connect to the OpenStack cloud and start using openstack command from the local terminal.

Then you can try a very usual command such as

kubectl get nodes

but it will not work. If there were load balancer “in front of the cluster”, it would work, but here there isn’t so it won’t. The rest of this article will show you how to still make it work, using the fact that the master node of the cluster has its own load balancer for kube-api.

Step 1 Create a Cluster With One Master Node and No Load Balancer

Create cluster NoLoadBalancer as explained in Prerequisite No. 3. Let there be

• one master node and

• no load balancers (do not check field Enable Load Balancer for Master Nodes in subwindow Network).

• Use any key pair that you might have – it is of no concern for this article.

• Activate NGINX as Ingress controller and then

• enter the following group of labels into the Advanced subwindow for cluster creation

octavia_enable_ingress_hostname=true,auto_healing_enabled=True,auto_healing_controller=magnum-auto-healer,magnum_auto_healer_tag=v1.21.0

The result will be creation of cluster NoLoadBalancer as seen in this image:

To illustrate the problem, a very basic command such as

kubectl get pods NoLoadBalancer -o yaml

to list the pods in cluster NoLoadBalancer, will show an error message like this one:

Unable to connect to the server: dial tcp 10.0.0.54:6443: i/o timeout

Addresses starting with 10.0… are usually reserved for local networks, meaning that no access from the Internet is enabled at this time.

Step 2 Create Floating IP for Master Node

Here are the instances that serve as nodes for that cluster:

Master node is called noloadbalancer-3h2i5x5iz2u6-master-0 and click on the right side of its row and click option Associate Floating IP.

To add the IP, click on a selection of available addresses (there may be only one but in certain cases, there can be several to choose from):

This is the result:

The IP number is 64.225.135.112 – you are going to use it later on, to change config file for access to the Kubernetes cluster.

Step 3 Create config File for Kubernetes Cluster

You are now going to connect to NoLoadBalancer cluster in spite of it not having a load balancer from the very start. To that end, create a config file to connect to the cluster, with the following command:

openstack coe cluster config NoLoadBalancer --force

It will return a row such as this:

export KUBECONFIG=/Users/duskosavic/CloudferroDocs/config

Execute this command from terminal command line. A config file has also been created at that address. To show its contents, execute command

cat config

assuming you already are in the required folder (CloudferroDocs in this case).

Config file will look a lot like giberish because it contains certificates, tokens and other rows with random content, some of them hundreds of characters long. Here is one part of it:

The important row here is this network address:

server: https://10.0.0.54:6443

Step 4 Swap Existing Floating IP Address for the Network Address

Now back to Horizon inteface and execute commands Compute -> Instances to see the addresses for master node of the NoLoadBalancer cluster:

There are two addresses:

10.0.0.54, 64.225.135.112

Incidentally, the same 10.0.0.54 address is also present in the config file, ending with port address :6443.

Try now in terminal to execute kubectl command and see the result, perhaps like this one:

The access is there but the nodes and pods are still out of reach. That is because address 10.0.0.54 is internal network address for the cluster and was never supposed to work as the Internet address.

So, open the config file using nano (or other text editor of your choice). Swap 10.0.0.54 for 64.225.135.112 in line for server access. The address 64.225.135.112 is the address of the floating IP for master node and will fit in perfectly.

The line should look like this:

Save the edited file. In case of nano, those will be commands Control-x, Y and pressing Enter on the keyboard.

Step 4 Add Parameter –insecure-skip-tls-verify=true to Make kubectl Work

Try again to activate kubectl and again it will fail. To make it work, add parameter –insecure-skip-tls-verify=true:

kubectl get pods --insecure-skip-tls-verify=true

Or, try out a more meaningful command

kubectl get nodes --insecure-skip-tls-verify=true

This is the result of all these commands, in terminal window:

To continue working successfully, use normal kubectl commands and always add –insecure-skip-tls-verify=true in the end.

For production, it is strongly recommended to check on field Enable Load Balancer for Master Nodes when creating a new cluster, regardless of the number of master nodes you have specified.