This document shows how to configure GKE on VMware to use bundled load balancing with the MetalLB load balancer.
In GKE on VMware, MetalLB runs in layer-2 mode.
Example of a MetalLB configuration
Here is an example of a configuration for clusters running the MetalLB load balancer:
The preceding diagram shows a MetalLB deployment. MetalLB runs directly on the cluster nodes. In this example, the admin cluster and user cluster are on two separate VLANs, and each cluster is in a separate subnet:
The following example of an admin cluster configuration file shows the configuration seen in the preceding diagram of:
MetalLB load balancer
VIPs on MetalLB for Kubernetes API server and add-ons of the admin cluster
network: hostConfig: ... ipMode: type: "static" ipBlockFilePath: "config-folder/admin-cluster-ipblock.yaml" ... loadBalancer: kind: "MetalLB" ... vips: controlPlaneVIP: "172.16.20.100" addonsVIP: "172.16.20.101"
The following example of an IP block file shows the designation of IP addresses for the nodes in the admin cluster. This also includes the address for the control-plane node for the user cluster and an IP address to use during cluster upgrade.
blocks: - netmask: "255.255.255.0" gateway: "220.127.116.11" ips: - ip: 172.16.20.50 hostname: admin-vm-1 - ip: 172.16.20.51 hostname: admin-vm-2 - ip: 172.16.20.52 hostname: admin-vm-3 - ip: 172.16.20.53 hostname: admin-vm-4 - ip: 172.16.20.54 hostname: admin-vm-5
The following example of a user cluster configuration file shows the configuration of:
Address pools for the MetalLB controller to choose from and assign to Services of type
LoadBalancer. The ingress VIP is in one of these pools.
VIP designated for the Kubernetes API server of the user cluster, and the ingress VIP you have chosen to configure for the ingress proxy. The Kubernetes API server VIP is on the admin cluster subnet because the control plane for a user cluster runs on a node in the admin cluster.
A node pool enabled to use MetalLB. MetalLB will be deployed on the nodes in the user cluster that belong to that node pool.
network: hostConfig: ... ipMode: type: "static" ipBlockFilePath: "config-folder/user-cluster-ipblock.yaml" ... loadBalancer: kind: MetalLB metalLB: addressPools: - name: "address-pool-1" addresses: - "172.16.40.100" - "172.16.40.101-172.16.40.112 avoidBuggyIPs: true ... vips: controlPlaneVIP: "172.16.20.102" ingressVIP: "172.16.40.102" ... nodePools: - name: "node-pool-1" cpus: 4 memoryMB: 8192 replicas: 3 enableLoadBalancer: true
The configuration in the preceding example specifies a set of addresses available for Services. When an application developer creates a Service of type
LoadBalancer in the user cluster, the MetalLB controller will choose an IP address from this pool.
The following example of an IP block file shows the designation of IP addresses for the nodes in the user cluster. This includes an IP address to use during cluster upgrade.
blocks: - netmask: "255.255.255.0" gateway: "18.104.22.168" ips: - ip: 172.16.40.21 hostname: user-vm-1 - ip: 172.16.40.22 hostname: user-vm-2 - ip: 172.16.40.23 hostname: user-vm-3 - ip: 172.16.40.24 hostname: user-vm-4 - ip: 172.16.40.25 hostname: user-vm-5
Set up MetalLB
Open firewall ports
MetalLB uses the
Go memberlist library
to do leader election. The
memberlist library uses TCP port 7946 and UDP port
7946 to exchange information. Make sure those ports are accessible for incoming and
outgoing traffic on all load-balancer nodes.
Enable MetalLB for a new admin cluster
admin cluster configuration
loadBalancer: kind: "MetalLB"
Fill in the rest of your admin cluster configuration file, and create your admin cluster as described in Create an admin cluster.
Specify address pools
The MetalLB controller does IP address management for Services. So when an application developer creates a Service of type LoadBalancer in a user cluster, they don't have to manually specify an IP address for the Service. Instead, the MetalLB controller chooses an IP address from an address pool that you specify at cluster creation time.
Think about how many Services of type LoadBalancer are likely to be active in
your user cluster at any given time. Then in the
section of your
user cluster configuration file, specify enough IP addresses to accommodate
If your application developers have no need to create Services of type LoadBalancer, then you don't have to specify any addresses other than the ingress VIP.
Addresses must be in CIDR format or range format. If you want to specify an individual address, use a /32 CIDR. For example:
addresses: - "192.0.2.0/26" - "192.0.2.64-192.0.2.72" - "192.0.2.75/32
If you need to adjust the addresses in a pool after the cluster is created, you
gkectl update cluster. For more information, see
Enable MetalLB for a new user cluster
In your user cluster configuration file:
- Specify one or more address pools for Services. The ingress VIP must be in one of these pools.
truefor at least one node pool in your cluster.
Fill in the rest of your user cluster configuration file, and create your user cluster as described in Create a user cluster.
Manual assignment of Service addresses
If you do not want the MetalLB controller to automatically assign IP
addresses from a particular pool to Services, set the
manualAssign field of
the pool to
true. Then a developer can create a Service of type
and manually specify one of the addresses from the pool. For example:
loadBalancer: metalLB: addressPools: - name: "my-address-pool-2" addresses: - "192.0.2.73-192.0.2.80" manualAssign: true
Avoiding buggy IP addresses
If you set the
avoidBuggyIPs field of an address pool to
true, the MetalLB
controller will not use addresses from the pool that end in .0 or .255. This
avoids the problem of buggy consumer devices mistakenly dropping traffic sent
to those special IP addresses. For example:
loadBalancer: metalLB: addressPools: - name: "my-address-pool-1" - "192.0.2.0/24" avoidBuggyIPs: true
Create a Service of type LoadBalancer
Here are two manifests: one for a Deployment and one for a Service:
apiVersion: apps/v1 kind: Deployment metadata: name: my-deployment spec: selector: matchLabels: greeting: hello replicas: 3 template: metadata: labels: greeting: hello spec: containers: - name: hello image: gcr.io/google-samples/hello-app:2.0 --- apiVersion: v1 kind: Service metadata: name: my-service spec: type: LoadBalancer selector: greeting: hello ports: - name: metal-lb-example-port protocol: TCP port: 60000 targetPort: 8080
Notice that the Service manifest does not specify an external IP address. The MetalLB controller will choose an external IP address from the address pool you specified in the user cluster configuration file.
Save the manifests in a file named
my-dep-svc.yaml. Then create the Deployment
and Service objects:
kubectl --kubeconfig USER_CLUSTER_KUBECONFIG apply -f my-dep-svc.yaml
View the Service:
kubectl --kubeconfig USER_CLUSTER_KUBECONIFG get service my-service --output wide
The output shows the external IP address that was automatically assigned to the Service. For example:
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE SELECTOR my-service LoadBalancer 10.96.2.166 192.0.2.2 60000:31914/TCP 28s
Verify that the assigned external IP address was taken from the address pool you specified in your user cluster configuration file. For example, 192.0.2.2 is in this address pool:
metalLB: addressPools: - name: "address-pool-1" addresses: - "192.0.2.0/24" - "198.51.100.1-198.51.100.3"
Call the Service:
The output displays a
Hello, world! message:
Hello, world! Version: 2.0.0
After you create your cluster, you can update the MetalLB address pools and the
enableLoadBalancer field in your node pools. Make the desired changes in the
user cluster configuration file, and then call
gkectl update cluster:
gkectl update cluster --kubeconfig ADMIN_CLUSTER_KUBECONIFG --config USER_CLUSTER_CONFIG
MetalLB Pods and ConfigMap
The MetalLB controller runs as a Deployment, and the MetalLB speaker runs as a
DaemonSet on nodes in pools that have
enableLoadBalancer set to
MetalLB controller manages the IP addresses assigned to Services. The MetalLB
speaker does leader election and announces Service VIPs.
View all MetalLB Pods:
kubectl --kubeconfig USER_CLUSTER_KUBECONIFG get pods --namespace kube-system --selector app=metallb
You can use the logs from the MetalLB Pods for troubleshooting.
MetalLB configuration is stored in a ConfigMap in a format known by MetalLB.
Do not change the ConfigMap directly. Instead, use
gkectl update cluster as
described previously. To view the ConfigMap for troubleshooting:
kubectl --kubeconfig USER_CLUSTER_KUBECONIFG get configmap metallb-config --namespace kube-system
Benefits of using MetalLB
MetalLB runs directly on your cluster nodes, so it doesn't require extra VMs.
The MetalLB controller does IP address management for Services, so you don't have to manually choose an IP address for each Service.
Active instances of MetalLB for different Services can run on different nodes.
You can share an IP address among different Services.
MetalLB compared to F5 BIG-IP and Seesaw
VIPs must be in the same subnet as the cluster nodes. This is also a requirement for Seesaw, but not for F5 BIG-IP.
There are no metrics for traffic.
There is no hitless failover; existing connections are reset during failover.
External traffic to the Pods of a particular Service passes through a single node running the MetalLB speaker. This means that the client IP address is usually not visible to containers running in the Pod.