kubernetes-vmware-autoscaler

kubernetes-vmware-autoscaler

Kubernetes autoscaler for vsphere/esxi including a custom resource controller to create managed node without code

Supported releases

• 1.26.11
  • This version is supported kubernetes v1.26 and support k3s, rke2, external kubernetes distribution
• 1.27.8
  • This version is supported kubernetes v1.27 and support k3s, rke2, external kubernetes distribution
• 1.28.1
  • This version is supported kubernetes v1.28 and support k3s, rke2, external kubernetes distribution
• 1.29.0
  • This version is supported kubernetes v1.29 and support k3s, rke2, external kubernetes distribution

How it works

This tool will drive vSphere to deploy VM at the demand. The cluster autoscaler deployment use vanilla cluster-autoscaler or my enhanced version of cluster-autoscaler.

This version use grpc to communicate with the cloud provider hosted outside the pod. A docker image is available here cluster-autoscaler

A sample of the cluster-autoscaler deployment is available at examples/cluster-autoscaler.yaml. You must fill value between <>

Before you must create a kubernetes cluster on vSphere

You can do it from scrash or you can use script from project autoscaled-masterkube-vmware to create a kubernetes cluster in single control plane or in HA mode with 3 control planes.

Commandline arguments

Build

The build process use make file. The simplest way to build is make container

New features

Use k3s, rke2 or external as kubernetes distribution method

Instead using kubeadm as kubernetes distribution method, it is possible to use k3s, rke2 or external

external allow to use custom shell script to join cluster

Samples provided here

• kubeadm
• rke2
• k3s
• external

Use the vanilla autoscaler with extern gRPC cloud provider

You can also use the vanilla autoscaler with the externalgrpc cloud provider

Samples of the cluster-autoscaler deployment with vanilla autoscaler. You must fill value between <>

• kubeadm
• rke2
• k3s
• external

Use external kubernetes distribution

When you use a custom method to create your cluster, you must provide a shell script to vmware-autoscaler to join the cluster. The script use a yaml config created by vmware-autscaler at the given path.

config: /etc/default/vmware-autoscaler-config.yaml

provider-id: vsphere://42373f8d-b72d-21c0-4299-a667a18c9fce
max-pods: 110
node-name: vmware-dev-rke2-woker-01
server: 192.168.1.120:9345
token: K1060b887525bbfa7472036caa8a3c36b550fbf05e6f8e3dbdd970739cbd7373537
disable-cloud-controller: false

If you declare to use an external etcd service

datastore-endpoint: https://1.2.3.4:2379
datastore-cafile: /etc/ssl/etcd/ca.pem
datastore-certfile: /etc/ssl/etcd/etcd.pem
datastore-keyfile: /etc/ssl/etcd/etcd-key.pem

You can also provide extras config onto this file

{
  "external": {
    "join-command": "/usr/local/bin/join-cluster.sh"
    "config-path": "/etc/default/vmware-autoscaler-config.yaml"
    "extra-config": {
        "mydata": {
          "extra": "ball"
        },
        "...": "..."
    }
  }
}

Your script is responsible to set the correct kubelet flags such as max-pods=110, provider-id=vsphere://42373f8d-b72d-21c0-4299-a667a18c9fce, cloud-provider=external, ...

Annotations requirements

If you expected to use vmware-autoscaler on already deployed kubernetes cluster, you must add some node annotations to existing node

Also don't forget to create an image usable by vmware-autoscaler to scale up the cluster create-image.sh

Sample master node

    cluster-autoscaler.kubernetes.io/scale-down-disabled: "true"
    cluster.autoscaler.nodegroup/autoprovision: "false"
    cluster.autoscaler.nodegroup/instance-id: 42373f8d-b72d-21c0-4299-a667a18c9fce
    cluster.autoscaler.nodegroup/managed: "false" 
    cluster.autoscaler.nodegroup/name: vmware-dev-rke2
    cluster.autoscaler.nodegroup/node-index: "0"

Sample first worker node

    cluster-autoscaler.kubernetes.io/scale-down-disabled: "true"
    cluster.autoscaler.nodegroup/autoprovision: "false"
    cluster.autoscaler.nodegroup/instance-id: 42370879-d4f7-eab0-a1c2-918a97ac6856
    cluster.autoscaler.nodegroup/managed: "false"
    cluster.autoscaler.nodegroup/name: vmware-dev-rke2
    cluster.autoscaler.nodegroup/node-index: "1"

Sample autoscaled worker node

    cluster-autoscaler.kubernetes.io/scale-down-disabled: "false"
    cluster.autoscaler.nodegroup/autoprovision: "true"
    cluster.autoscaler.nodegroup/instance-id: 3d25c629-3f1d-46b3-be9f-b95db2a64859
    cluster.autoscaler.nodegroup/managed: "false"
    cluster.autoscaler.nodegroup/name: vmware-dev-rke2
    cluster.autoscaler.nodegroup/node-index: "2"

Node labels

These labels will be added

Network

Now it's possible to disable dhcp-default routes and custom route

VMWare CPI compliant

Version 1.24.6 and 1.25.2 and above are vsphere cloud provider by building provider-id conform to syntax vsphere://<VM UUID>

CRD controller

This new release include a CRD controller allowing to create kubernetes node without use of govc or code. Just by apply a configuration file, you have the ability to create nodes on the fly.

As exemple you can take a look on artifacts/examples/example.yaml on execute the following command to create a new node

kubectl apply -f artifacts/examples/example.yaml

If you want delete the node just delete the CRD with the call

kubectl delete -f artifacts/examples/example.yaml

You have the ability also to create a control plane as instead a worker

kubectl apply -f artifacts/examples/controlplane.yaml

The resource is cluster scope so you don't need a namespace. The name of the resource is not the name of the managed node.

The minimal resource declaration

apiVersion: "nodemanager.aldunelabs.com/v1alpha1"
kind: "ManagedNode"
metadata:
  name: "vmware-ca-k8s-managed-01"
spec:
  nodegroup: vmware-ca-k8s
  vcpus: 2
  memorySizeInMb: 2048
  diskSizeInMb: 10240

The full qualified resource including networks declaration to override the default controller network management and adding some node labels & annotations. If you specify the managed node as controller, you can also allows the controlplane to support deployment as a worker node

apiVersion: "nodemanager.aldunelabs.com/v1alpha1"
kind: "ManagedNode"
metadata:
  name: "vmware-ca-k8s-managed-01"
spec:
  nodegroup: vmware-ca-k8s
  controlPlane: false
  allowDeployment: false
  vcpus: 2
  memorySizeInMb: 2048
  diskSizeInMb: 10240
  labels:
  - demo-label.acme.com=demo
  - sample-label.acme.com=sample
  annotations:
  - demo-annotation.acme.com=demo
  - sample-annotation.acme.com=sample
  networks:
    -
      network: "VM Network"
      address: 10.0.0.80
      netmask: 255.255.255.0
      gateway: 10.0.0.1
      use-dhcp-routes: false
      routes:
        - to: x.x.0.0/16
          via: 10.0.0.253
          metric: 100
        - to: y.y.y.y/8
          via: 10.0.0.253
          metric: 500
    -
      network: "VM Private"
      address: 192.168.1.80
      netmask: 255.255.255.0
      use-dhcp-routes: false

Declare additional routes and disable default DHCP routes

The release 1.24 and above allows to add additionnal route per interface, it also allows to disable default route declared by DHCP server.

As example of use generated by autoscaled-masterkube-vmware scripts

{
    "use-external-etcd": false,
    "src-etcd-ssl-dir": "/etc/etcd/ssl",
    "dst-etcd-ssl-dir": "/etc/kubernetes/pki/etcd",
    "kubernetes-pki-srcdir": "/etc/kubernetes/pki",
    "kubernetes-pki-dstdir": "/etc/kubernetes/pki",
    "distribution": "rke2",
    "network": "unix",
    "listen": "/var/run/cluster-autoscaler/vmware.sock",
    "cert-private-key": "/etc/ssl/client-cert/tls.key",
    "cert-public-key": "/etc/ssl/client-cert/tls.crt",
    "cert-ca": "/etc/ssl/client-cert/ca.crt",
    "secret": "vmware",
    "minNode": 0,
    "maxNode": 9,
    "maxNode-per-cycle": 2,
    "node-name-prefix": "autoscaled",
    "managed-name-prefix": "managed",
    "controlplane-name-prefix": "master",
    "nodePrice": 0,
    "podPrice": 0,
    "image": "jammy-kubernetes-cni-flannel-v1.27.8-containerd-amd64",
    "optionals": {
        "pricing": false,
        "getAvailableMachineTypes": false,
        "newNodeGroup": false,
        "templateNodeInfo": false,
        "createNodeGroup": false,
        "deleteNodeGroup": false,
    },
    "kubeadm": {
        "address": "192.168.1.120:6443",
        "token": "h1g55p.hm4rg52ymloax182",
        "ca": "sha256:c7a86a7a9a03a628b59207f4f3b3e038ebd03260f3ad5ba28f364d513b01f542",
        "extras-args": [
            "--ignore-preflight-errors=All"
        ],
    },
    "k3s": {
        "address": "192.168.1.120:6443",
        "token": "h1g55p.hm4rg52ymloax182",
        "datastore-endpoint": "https://1.2.3.4:2379",
        "extras-commands": []
    },
    "external": {
        "address": "192.168.1.120:6443",
        "token": "h1g55p.hm4rg52ymloax182",
        "datastore-endpoint": "https://1.2.3.4:2379",
        "join-command": "/usr/local/bin/join-cluster.sh",
        "config-path": "/etc/default/vmware-autoscaler-config.yaml",
        "extra-config": {
            "...": "..."
        }
    },
    "default-machine": "large",
    "machines": {
        "tiny": {
            "memsize": 2048,
            "vcpus": 2,
            "disksize": 10240
        },
        "small": {
            "memsize": 4096,
            "vcpus": 2,
            "disksize": 20480
        },
        "medium": {
            "memsize": 4096,
            "vcpus": 4,
            "disksize": 20480
        },
        "large": {
            "memsize": 8192,
            "vcpus": 4,
            "disksize": 51200
        },
        "xlarge": {
            "memsize": 16384,
            "vcpus": 4,
            "disksize": 102400
        },
        "2xlarge": {
            "memsize": 16384,
            "vcpus": 8,
            "disksize": 102400
        },
        "4xlarge": {
            "memsize": 32768,
            "vcpus": 8,
            "disksize": 102400
        },
    },
    "node-labels": [
        "topology.kubernetes.io/region=home",
        "topology.kubernetes.io/zone=office",
        "topology.csi.vmware.com/k8s-region=home",
        "topology.csi.vmware.com/k8s-zone=office",
    ],
    "cloud-init": {
        "package_update": false,
        "package_upgrade": false,
        "runcmd": [
            "echo 1 > /sys/block/sda/device/rescan",
            "growpart /dev/sda 1",
            "resize2fs /dev/sda1",
            "echo '192.168.1.120 vmware-ca-k8s-masterkube vmware-ca-k8s-masterkube.acme.com' >> /etc/hosts",
        ],
    },
    "ssh-infos": {
        "user": "kubernetes",
        "ssh-private-key": "/root/.ssh/id_rsa"
    },
    "autoscaling-options": {
        "scaleDownUtilizationThreshold": 0.5,
        "scaleDownGpuUtilizationThreshold": 0.5,
        "scaleDownUnneededTime": "1m",
        "scaleDownUnreadyTime": "1m",
    },
    "vmware": {
        "vmware-ca-k8s": {
            "url": "https://administrator@acme.com:mySecret@vsphere.acme.com/sdk",
            "uid": "administrator@vsphere.acme.com",
            "password": "mySecret",
            "insecure": true,
            "dc": "DC01",
            "datastore": "datastore1",
            "resource-pool": "ACME/Resources/FR",
            "vmFolder": "HOME",
            "timeout": 300,
            "template-name": "jammy-kubernetes-cni-flannel-v1.26.0-containerd-amd64",
            "template": false,
            "linked": false,
            "customization": "",
            "network": {
                "domain": "acme.com",
                "dns": {
                    "search": [
                        "acme.com"
                    ],
                    "nameserver": [
                        "10.0.0.1"
                    ]
                },
                "interfaces": [
                    {
                        "primary": false,
                        "exists": true,
                        "network": "VM Network",
                        "adapter": "vmxnet3",
                        "mac-address": "generate",
                        "nic": "eth0",
                        "dhcp": true,
                        "use-dhcp-routes": true,
                        "routes": [
                            {
                                "to": "172.30.0.0/16",
                                "via": "10.0.0.5",
                                "metric": 500,
                            },
                        ],
                    },
                    {
                        "primary": true,
                        "exists": true,
                        "network": "VM Private",
                        "adapter": "vmxnet3",
                        "mac-address": "generate",
                        "nic": "eth1",
                        "dhcp": true,
                        "use-dhcp-routes": false,
                        "address": "192.168.1.124",
                        "gateway": "10.0.0.1",
                        "netmask": "255.255.255.0",
                        "routes": []
                    }
                ]
            }
        }
    }
}

Unmaintened releases

All release before 1.26.11 are not maintened