Bundled load balancing with MetalLB

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:

MetalLB load balancer configuration.
MetalLB load balancer configuration

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:

Cluster Subnet
Admin cluster 172.16.20.0/24
User cluster 172.16.40.0/24

admin-cluster.yaml

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"

admin-cluster-ipblock.yaml

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: "17.16.20.1"
  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

user-cluster.yaml

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:
      - "172.16.40.100/32"
      - "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.

user-cluster-ipblock.yaml

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: "17.16.40.1"
  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

In your admin cluster configuration file, set loadBalancer.kind to "MetalLB".

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 loadBalancer.metalLB.addressPools section of your user cluster configuration file, specify enough IP addresses to accommodate those Services.

The ingress VIP for your user cluster must be among the addresses that you specify in an address pool. This is because the ingress proxy is exposed by a Service of type LoadBalancer.

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 can use gkectl update cluster. For more information, see Update MetalLB.

Enable MetalLB for a new user cluster

In your user cluster configuration file:

  • Set loadBalancer.kind to "MetalLB".
  • Specify one or more address pools for Services. The ingress VIP must be in one of these pools.
  • Set enableLoadBalancer to true for 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 LoadBalancer 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"
      addresses:
      - "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:

curl EXTERNAL_IP_ADDRESS:60000

The output displays a Hello, world! message:

Hello, world!
Version: 2.0.0

Update MetalLB

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 true. The 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.