Skip to content
This repository has been archived by the owner on Mar 8, 2022. It is now read-only.

Latest commit

 

History

History
304 lines (241 loc) · 26.4 KB

README.md

File metadata and controls

304 lines (241 loc) · 26.4 KB
Raw

k8s-vagrant-multi-node

Build Status

This project was based on work from coolsvap/kubeadm-vagrant by @coolsvap, now it is mostly independent.

A demo of the start and destroy of a cluster can be found here: README.md Demo section.

Prerequisites

  • make
  • kubectl - Optional when KUBECTL_AUTO_CONF is set to false (default: true).
  • grep
  • cut
  • rsync
  • Source for randomness (only used to generate a kubeadm token, when no custom KUBETOKEN is given):
    • /dev/urandom
    • openssl command - Fallback for when /dev/urandom is not available.
  • Vagrant (>= 2.2.0)
    • Tested with 2.2.2 (if you should experience issues, please upgrade to at least this Vagrant version or higher)
    • Plugins
      • vagrant-reload REQUIRED For BOX_OS=fedora (set by default) and when using the vagrant-reload* targets, the vagrant-reload plugin is needed. An automatic attempt to install the plugin is made. To install manually run one of the following commands:
        • make vagrant-plugins or
        • vagrant plugin install vagrant-reload
  • Vagrant Provider (one of the following two is needed)
    • libvirt (vagrant plugin install vagrant-libvirt)
      • Tested with libvirtd version 5.10.0.
      • Libvirt support is still a bit experimental and can be unstable (e.g., VMs not getting IPs).
        • Troubleshooting: If your VM creation is hanging at Waiting for domain to get an IP address..., using virsh run virsh force reset VM_NAME (VM_NAME can be obtained using virsh list command) or in virt-manager Force Reset on the VM.
    • Virtualbox (WARNING VirtualBox seems to hang the Makefile randomly for some people, libvirt is recommended)
      • Tested with 6.0.0 (if you should experience issues, please upgrade to at least this version or higher)
      • VBoxManage binary in PATH.

NOTE kubectl is only needed when the kubectl auto configuration is enabled (default is enabled), to disable it set the variable KUBECTL_AUTO_CONF to false. For more information, see the Variables section.

Hardware Requirements

  • Master
    • CPU: 2 Cores (MASTER_CPUS)
    • Memory: 2GB (MASTER_MEMORY_SIZE_GB)
  • 1x Node:
    • CPU: 1 Core (it is recommended to use at least 2 Cores; NODE_CPUS)
    • Memory: 2GB (it is recommended to use more than 2GB; NODE_MEMORY_SIZE_GB)

These resources can be changed by setting the according variables for the make up command, see Variables section.

Quickstart

To start with the defaults, 1x master and 2x workers, run the following:

$ make up -j 3

The -j3 will cause three VMs to be started in parallel to speed up the cluster creation.

NOTE Your kubectl is automatically configured to use a context for the created cluster, after the master VM is started. The context is named after the directory the Makefile is in.

$ kubectl config current-context
k8s-vagrant-multi-node
$ kubectl get componentstatus
NAME                 STATUS    MESSAGE              ERROR
scheduler            Healthy   ok
controller-manager   Healthy   ok
etcd-0               Healthy   {"health": "true"}
$ kubectl get nodes
NAME      STATUS    ROLES     AGE       VERSION
master    Ready     master    4m        v1.17.3
node1     Ready     <none>    4m        v1.17.3
node2     Ready     <none>    4m        v1.17.3

Different OS / Vagrantfiles

There are multiple sets of Vagrantfiles available (see vagrantfiles/) which can be used to use a different OS for the Kubernetes environment.

List of currently available Vagrantfile sets:

Name Container Runtime OS Version Special Notes
centos7 Docker/Moby CentOS 7 N/A
centos8 Docker/Moby CentOS 8 KUBE_NETWORK=calico is forced, due to issues under CentOS 8 regarding iptables.
centos Docker/Moby CentOS 7 Use centos7 in favor of this, as this "target" might be changed to centos8 in a future release.
fedora Docker/Moby Fedora 30 N/A
ubuntu Docker/Moby Ubuntu 18.04 KUBE_NETWORK=canal is forced, due to issues under Ubuntu. Additionally Google DNS Servers are used as the nameservers.

To use a different set than the default fedora one's, add BOX_OS=__NAME__ (where __NAME__ is, e.g., fedora).

Usage

Starting the environment

To start up the Vagrant Kubernetes multi node environment with the default of two worker nodes + a master (not parallel) run:

$ make up

NOTE Your kubectl is automatically configured to use a context for the created cluster, after the master VM is started. The context is named after the directory the Makefile is in.

Faster (parallel) environment start

To start up 4 VMs in parallel run (-j flag does not control how many (worker) VMs are started, the NODE_COUNT variable is used for that):

$ NODE_COUNT=3 make up -j4

The flag -j CORES/THREADS allows yout to set how many VMs (Makefile targets) will be run at the same time. You can also use -j $(nproc) to start as many VMs as cores/threads you have in your machine. So to start up all VMs (master and three nodes) in parallel, you would add one to the chosen NODE_COUNT.

Show status of VMs

$ make status
master                    not created (virtualbox)
node1                     not created (virtualbox)
node2                     not created (virtualbox)

Shutting down the environment

To destroy the Vagrant environment run:

$ make clean
$ make clean-data

Copy local Docker image into VMs

The make load-image target can be used to copy a docker image from your local docker daemon to all the VMs in your cluster. The IMG variable can be expressed in a few ways, for example:

$ make load-image IMG=your_name/your_image_name:your_tag
$ make load-image IMG=your_name/your_image_name
$ make load-image IMG=my-private-registry.com/your_name/your_image_name:your_tag

You can also specify a new image name and tag to use after the image has been copied to the VM's by setting the TAG variable. This will not change the image/tag in your local docker daemon, it will only affect the image in the VM's.

$ make load-image IMG=repo/image:tag TAG=new_repo/new_image:new_tag

Data inside VM

See the data/VM_NAME/ directories, where VM_NAME is for example master.

Show make targets

$ make help
Usage: make [TARGET ...]

clean-data                     Remove data (shared folders) and disks of all VMs (master and nodes).
clean-force                    Remove all drives which should normally have been removed by the normal clean-master or clean-node-% targets.
clean                          Destroy master and node VMs, delete data and the kubectl context.
clean-master                   Remove the master VM and the kubectl context.
clean-node-%                   Remove a node VM, where `%` is the number of the node.
clean-nodes                    Remove all node VMs.
help                           Show this help menu.
kubectl                        Configure kubeconfig context for the cluster using `kubectl config` (automatically done by `up` target).
kubectl-delete                 Delete the created CLUSTER_NAME context from the kubeconfig (uses kubectl).
load-image                     Load local/pulled Docker image into master and all node VMs.
load-image-master              Load local/pulled image into master VM.
load-image-node-%              Load local/pulled image into node VM, where `%` is the number of the node.
load-image-nodes               Load local/pulled Docker image into all node VMs.
preflight                      Run checks and gather variables, used for the the `up` target.
pull                           Add and download, or update the box image for the chosen provider on the host.
ssh-config-master              Generate SSH config just for the master.
ssh-config-node-%              Generate SSH config just for the one node number given.
ssh-config-nodes               Generate SSH config just for the nodes.
ssh-config                     Generate SSH config for master and nodes.
ssh-master                     SSH into the master VM.
ssh-node-%                     SSH into a node VM, where `%` is the number of the node.
start-master                   Start up master VM (automatically done by `up` target).
start-node-%                   Start node VM, where `%` is the number of the node.
start-nodes                    Create and start all node VMs by utilizing the `node-X` target (automatically done by `up` target).
status-master                  Show status of the master VM.
status-nodes                   Show status of all node VMs.
status-node-%                  Show status of a node VM, where `%` is the number of the node.
status                         Show status of master and all node VMs.
stop-master                    Stop/Halt the master VM.
stop-nodes                     Stop/Halt all node VMs.
stop-node-%                    Stop/Halt a node VM, where `%` is the number of the node.
stop                           Stop/Halt master and all nodes VMs.
tests                          Run shunit2 tests (`expect` command is required).
token                          Generate a kubeadm join token, if needed (token file is `DIRECTORY_OF_MAKEFILE/.vagrant/KUBETOKEN`).
up                             Start Kubernetes Vagrant multi-node cluster. Creates, starts and bootsup the master and node VMs.
vagrant-plugins                Checks that vagrant-reload plugin is installed, if not try to install it
vagrant-reload-master          Run vagrant reload for master VM.
vagrant-reload-nodes           Run `vagrant reload` for all node VMs.
vagrant-reload-node-%          Run `vagrant reload` for specific node  VM.
vagrant-reload                 Run vagrant reload on master and nodes.
versions                       Print the "imporant" tools versions out for easier debugging.

Variables

Variable Name Default Value Description
VAGRANT_DEFAULT_PROVIDER virtualbox Which Vagrant provider to use. Available are virtualbox and libvirt.
BOX_OS fedora Which set of Vagrantfiles to use to start the VMs, see Different OS / Vagrantfiles section.
BOX_IMAGE "" (empty) Override the VM box image used (only use for override purposes as the image is set based on the BOX_OS variable).
DISK_COUNT 2 Set how many additional disks will be added to the VMs.
DISK_SIZE_GB 20 GB Size of additional disks added to the VMs.
MASTER_CPUS 2 Core Amount of cores to use for the master VM.
MASTER_MEMORY_SIZE_GB 2 GB Size of memory (in GB) to be allocated for the master VM.
NODE_CPUS 1 Amount of cores to use for each node VM.
NODE_MEMORY_SIZE_GB 2 GB Size of memory (in GB) to be allocated for each node VM.
NODE_COUNT 2 How many worker nodes should be spawned.
LIBVIRT_STORAGE_POOL "default" Storage pool which libvirt should use. Libvirt only.
MASTER_IP 192.168.26.10 The Kubernetes master node IP.
NODE_IP_NW 192.168.26. The first three parts of the IPs used for the nodes.
POD_NW_CIDR 10.244.0.0/16 The Pod (container) network CIDR used for the CNI.
K8S_DASHBOARD false Install the Kubernetes dashboard addon.
K8S_DASHBOARD_VERSION v1.10.1 The Kubernetes dashboard addon version. Note it is recommended to at least version 1.10.1.
KUBE_WEB_VIEW false Install the kube-web-view from Henning Jacobs (Twitter: @try_except_). Access it using kubectl port-forward -n kube-web-view service/kube-web-view 8080:8080 and point your browser to http://127.0.0.1:8080/.
CLUSTER_NAME k8s-vagrant-multi-node The name of the directory the Makefile is in. This is not the Kubernetes cluster name, due to kubeadm init limitations.
KUBETOKEN "" (empty) The kubeadm "join" token to use. Will be generated automatically using /dev/urandom/ when empty.
KUBEADM_INIT_FLAGS "" (empty) The kubeadm init flags to use. (When KUBERNETES_VERSION is set and KUBEADM_INIT_FLAGS is empty, KUBEADM_INIT_FLAGS will automatically be set to --kubernetes-version=$KUBERNETES_VERSION).
KUBEADM_JOIN_FLAGS "" (empty) The kubeadm join flags to use.
KUBERNETES_VERSION "" (empty) The kubeadm and kubelet package and API server version to install. Must be a fully qualified version, e.g., 1.18.3, v1.18.3, but not just 1.18.
KUBERNETES_PKG_VERSION_SUFFIX "" (empty) String which will be appended to the kubeadm and kubelet package versions when installed (only used for vagrantfiles/ubuntu).
KUBE_PROXY_IPVS false Enable IPVS kernel modules to then use IPVS for the kube-proxy.
KUBE_NETWORK calico What CNI to install, if empty don't install any CNI. calico, canal, flannel and none are supported options. none will cause no CNI to be installed. (See Different OS / Vagrantfiles for OS specific overrides, none is never overriden).
KUBECTL_AUTO_CONF true If kubectl should be automatically configured to be able to talk with the cluster (if disabled, removes need for kubectl binary).
USER_SSHPUBKEY "" (empty) Your SSH public key (not private) to add to the VMs vagrant users .ssh/authorized_keys file during VM provisioning.
HTTP_PROXY "" (empty) HTTP proxy to set for package installation and the Docker daemon (for pulling images).
HTTPS_PROXY "" (empty) HTTPS proxy to set for package installation and the Docker daemon (for pulling images).
HTTP_PROXY_USERNAME "" (empty) Only used for CentOS yum and Fedora dnf package managers. HTTP and HTTPS proxy username.
HTTP_PROXY_PASSWORD "" (empty) Only used for CentOS yum and Fedora dnf package managers. HTTP and HTTPS proxy password.
NO_PROXY "" (empty) NO_PROXY / no_proxy list to set for the Docker daemon (for pulling images). It is currently not possible to set this for package installation.
INSTALL_ADDITIONAL_PACKAGES "" (empty) List of additional packages to install in the VMs (packages are space separated; the variable is directly passed to the pacakge manager install command).
VAGRANT vagrant Path to vagrant binary (only needed when vagrant is no in your PATH)
KUBECTL kubectl Path to kubectl binary (only needed when kubectl is no in your PATH)
PARALLEL_VM_START false (Only use if you know what the effects can be) If master and nodes should be started in parallel, this does not affect the nodes creation + startup. This is normally controlled by passing -j JOBS to the make command.
NETWORK_VM_MTU 1500 Set to, e.g., 1350, to have the VMs interfaces MTU be set to 1350. This can be used to prevent issues with VPNs running on the host machine (e.g., OpenVPN, Wireguard, etc).
KUBE_NETWORK_MTU 1450 Use in combination with the NETWORK_VM_MTU parameter, this should be set to the value of NETWORK_VM_MTU - 50.

Troubleshooting

When usign Virtualbox as the provider make up hangs after it is done

For unknown reasons the makefile is not exiting after it has printed the "cluster creation successful" message. The issue is being looked into it, till then just do CTRL+C to exit the make up command.

"I have a VPN running on my host machine, what should I look out for?"

TL;DR Set the following variables on your make up run as follows: NETWORK_VM_MTU=1350 and KUBE_NETWORK_MTU=1300.

Set the NETWORK_VM_MTU and KUBE_NETWORK_MTU according to the MTU of your VPN interface(s) - "overhead" (50). Using the values in the TL;DR should work for "99% percent" of common VPNs.

Demo

Please note that these terminal recordings are currently outdated.

Start Cluster

asciicast

Destroy Cluster

asciicast

Creating an Issue

Please attach the make versions output to the issue as is shown in the issue template. This makes debugging easier.