Skip to content
This repository has been archived by the owner on Mar 15, 2021. It is now read-only.

Deploy your `$GOPATH/src/k8s.io/kubernetes` inside a set of VM's

License

Notifications You must be signed in to change notification settings

ereslibre/kubernetes-cluster-vagrant

Repository files navigation

kubernetes-cluster-vagrant

What

This project intends to make it very easy to push your changes to the Kubernetes source code into a cluster.

It uses Vagrant to create the different clusters, and allows you to have different cluster profiles.

The idea behind this project is to create a base box that downloads all the required dependencies. This way, it's possible to completely remove network latency (or work completely offline if required) from your workflow.

The idea is to also automate the whole process completely, allowing for the automation to be disabled in case you as a Kubernetes developer want to perform some steps manually.

You can push your changes on the Kubernetes source code to the cluster when you are creating the cluster, or when it's already running.

Both single master and multi master deployments are supported. In the multi master deployment case, a load balancer will be created in a separate machine.

It's also possible to run several clusters at the same time, as long as they are different profiles and their cluster and/or machine names don't clash, as well as the IP addresses assigned to them.

Requirements

Quick start

You only need to have cloned Kubernetes in your $GOPATH in your host. The path should be $GOPATH/src/k8s.io/kubernetes. Let's create the Kubernetes artifacts first:

~/p/kubernetes-cluster-vagrant (master) > make artifacts

<snip>

INFO: Elapsed time: 68.716s, Critical Path: 38.72s
INFO: 60 processes: 60 processwrapper-sandbox.
INFO: Build completed successfully, 61 total actions

Now let's create a 1 master and 1 worker cluster:

~/p/kubernetes-cluster-vagrant (master) > PROFILE=bootstrap/1-master-1-worker make

<snip>

>>> kubeconfig written to /home/ereslibre/.kube/config

Now you can execute from your host directly:

~/p/kubernetes-cluster-vagrant (master) > kubectl get nodes
NAME      STATUS    ROLES     AGE       VERSION
master    Ready     master    79s       v1.14.0-alpha.0.974+f89aa180476882
worker    Ready     <none>    41s       v1.14.0-alpha.0.974+f89aa180476882

Now, let's clean the cluster machines and the container used for building the Kubernetes artifacts:

~/p/kubernetes-cluster-vagrant (master) > PROFILE=bootstrap/1-master-1-worker make clean

Building Kubernetes artifacts

This project can help you to build Kubernetes artifacts with a single command, it's handy if you don't want to leave this directory. The tasks created by this project will launch bazel in a container in your host. This container will be kept running so we leave the bazel instance running inside the container.

If you need to modify autogenerated files bazel won't do this for you, so in this case you still need to go to your $GOPATH/src/k8s.io/kubernetes and run a regular make command to regenerate the autogenerated files.

Within this project directory you can run several targets in order for artifacts to be recreated:

  • make debs

    • Will regenerate the debs based targets:
      • cri-tools
      • kubeadm
      • kubectl
      • kubelet
      • kubernetes-cni
  • make images

    • Will regenerate the container images targets:
      • kube-apiserver
      • kube-controller-manager
      • kube-proxy
      • kube-scheduler
  • make artifacts

    • Will run:
      • make debs
      • make images

After you have generated the artifacts you can read how to create a cluster using these artifacts or how to deploy them on an already running cluster:

Base box

The base box will contain some general images that will be used later when creating the cluster. These images are defined in container_images.json. At the moment they are:

  • coredns
  • etcd
  • flannel
  • pause

Also, the following images inside $GOPATH/src/k8s.io/kubernetes/bazel-bin/build (created manually or by running make images within this project) are also copied and loaded in the base box:

  • kube-apiserver
  • kube-controller-manager
  • kube-proxy
  • kube-scheduler

Aside from that, the following packages inside $GOPATH/src/k8s.io/kubernetes/bazel-bin/build/debs (created manually or by running make debs within this project) are copied and installed in the base box too:

  • cri-tools
  • kubeadm
  • kubectl
  • kubelet
  • kubernetes-cni

Building the base box requires internet connection for installing some packages and the first set of containers (the ones defined in container_images.json).

Once the base box is built, it is saved as a Vagrant box named kubernetes-vagrant.

The idea behind the base box is that it contains everything needed to bring up a cluster in an offline environment, so if you are temporarily working on slow networks it won't affect your workflow.

When you create a cluster, you can override whatever you want (e.g. kubeadm and kubelet packages), that will be installed on top of the ones included by the base box. This allows you to create clusters faster, only installing the packages you modified and checking the results.

From time to time you'll need to rebuild the base box, but it's an action that you won't perform nearly as often as creating machines for testing your changes on the Kubernetes source code.

Building a cluster

Building a cluster is trivial, all you need is to point to a specific profile. Take into account that this process is completely offline and doesn't need to download any information from the internet.

You can just run make with the PROFILE envvar set to a profile name that exists inside the profiles folder, or to a full path containing the profile you want.

~/p/kubernetes-cluster-vagrant (master) > time PROFILE=bootstrap/1-master-1-worker make
make -C base-box
make[1]: Entering directory '/home/ereslibre/projects/kubernetes-cluster-vagrant/base-box'
>>> Base box (kubernetes-vagrant) already exists, skipping build
make[1]: Leaving directory '/home/ereslibre/projects/kubernetes-cluster-vagrant/base-box'
vagrant up
Bringing machine 'kubernetes_master' up with 'virtualbox' provider...
Bringing machine 'kubernetes_worker' up with 'virtualbox' provider...

<snip>

>>> kubeconfig written to /home/ereslibre/.kube/config
9.30user 4.82system 2:14.49elapsed 10%CPU (0avgtext+0avgdata 64236maxresident)k
0inputs+584outputs (0major+690067minor)pagefaults 0swaps

Now you can execute from your host directly:

~/p/kubernetes-cluster-vagrant (master) > kubectl get nodes
NAME      STATUS    ROLES     AGE       VERSION
master    Ready     master    74s       v1.14.0-alpha.0.756+51453a31317118
worker    Ready     <none>    16s       v1.14.0-alpha.0.756+51453a31317118

This will use all the versions of packages and components present in the base box, what might not be what you are looking for, since your local changes to Kubernetes weren't pushed, because on the make command we didn't specify PACKAGES, IMAGES or MANIFESTS environment variables, so the ones from the base box were used.

Everything that you want to deploy to the cluster can be controlled by environment variables. The currently supported ones are:

  • PACKAGES
    • cri-tools
    • kubeadm
    • kubectl
    • kubelet
    • kubernetes-cni
  • IMAGES
    • kube-apiserver
    • kube-controller-manager
    • kube-proxy
    • kube-scheduler
  • MANIFESTS
    • flannel

It's possible to start the cluster and make it use your recently built component versions. You just need to provide what PACKAGES, IMAGES or MANIFESTS you want to be deployed when the cluster is created. For example:

~/p/kubernetes-cluster-vagrant (master) > PROFILE=bootstrap/1-master-1-worker PACKAGES=kubeadm,kubelet make

In this case, it doesn't matter what versions of kubeadm and kubelet existed on the base box, the ones that you built on the host will be transferred to the different cluster machines and installed as soon as the machines are up.

In the case that you are using a profile that is bootstrapping the cluster automatically, this bootstrap will happen after the new packages have overriden the old ones.

Deploying changes to the cluster while it's running

vagrant provision can be used to deploy different things to the cluster. As with a regular vagrant up (or make), you can also use PACKAGES, IMAGES and MANIFESTS environment variables at will, to control what to deploy.

~/p/kubernetes-cluster-vagrant (master) > PROFILE=bootstrap/1-master-1-worker PACKAGES=kubeadm,kubelet IMAGES=kube-apiserver,kube-scheduler vagrant provision
==> kubernetes_master: Running provisioner: file...
==> kubernetes_master: Running provisioner: file...
==> kubernetes_master: Running provisioner: file...
==> kubernetes_master: Running provisioner: file...
==> kubernetes_master: Running provisioner: shell...
    kubernetes_master: Running: inline script
    kubernetes_master: (Reading database ... 60170 files and directories currently installed.)
    kubernetes_master: Preparing to unpack .../vagrant/kubernetes/kubeadm.deb ...
    kubernetes_master: Unpacking kubeadm (1.14.0~alpha.0.756+51453a31317118) over (1.14.0~alpha.0.756+51453a31317118) ...
    kubernetes_master: Setting up kubeadm (1.14.0~alpha.0.756+51453a31317118) ...
    kubernetes_master: (Reading database ... 60170 files and directories currently installed.)
    kubernetes_master: Preparing to unpack .../vagrant/kubernetes/kubelet.deb ...
    kubernetes_master: Unpacking kubelet (1.14.0~alpha.0.756+51453a31317118) over (1.14.0~alpha.0.756+51453a31317118) ...
    kubernetes_master: Setting up kubelet (1.14.0~alpha.0.756+51453a31317118) ...
    kubernetes_master: Loaded image: k8s.gcr.io/kube-apiserver:v1.14.0-alpha.0.756_51453a31317118
    kubernetes_master: Loaded image: k8s.gcr.io/kube-scheduler:v1.14.0-alpha.0.756_51453a31317118
==> kubernetes_worker: Running provisioner: file...
==> kubernetes_worker: Running provisioner: file...
==> kubernetes_worker: Running provisioner: file...
==> kubernetes_worker: Running provisioner: file...
==> kubernetes_worker: Running provisioner: shell...
    kubernetes_worker: Running: inline script
    kubernetes_worker: (Reading database ... 60170 files and directories currently installed.)
    kubernetes_worker: Preparing to unpack .../vagrant/kubernetes/kubeadm.deb ...
    kubernetes_worker: Unpacking kubeadm (1.14.0~alpha.0.756+51453a31317118) over (1.14.0~alpha.0.756+51453a31317118) ...
    kubernetes_worker: Setting up kubeadm (1.14.0~alpha.0.756+51453a31317118) ...
    kubernetes_worker: (Reading database ... 60170 files and directories currently installed.)
    kubernetes_worker: Preparing to unpack .../vagrant/kubernetes/kubelet.deb ...
    kubernetes_worker: Unpacking kubelet (1.14.0~alpha.0.756+51453a31317118) over (1.14.0~alpha.0.756+51453a31317118) ...
    kubernetes_worker: Setting up kubelet (1.14.0~alpha.0.756+51453a31317118) ...
    kubernetes_worker: Loaded image: k8s.gcr.io/kube-apiserver:v1.14.0-alpha.0.756_51453a31317118
    kubernetes_worker: Loaded image: k8s.gcr.io/kube-scheduler:v1.14.0-alpha.0.756_51453a31317118

HA deployments (multi master)

Multi master deployments are supported, and are as simple as setting the correct profile. Example:

~/p/kubernetes-cluster-vagrant (master) > time PROFILE=bootstrap/3-masters-1-worker make

<snip>

>>> kubeconfig written to /home/ereslibre/.kube/config
22.03user 12.93system 4:27.33elapsed 13%CPU (0avgtext+0avgdata 80360maxresident)k
0inputs+1376outputs (0major+1914771minor)pagefaults 0swaps

From your host, now you can run:

~/p/kubernetes-cluster-vagrant (master) > kubectl get nodes
NAME      STATUS    ROLES     AGE       VERSION
master1   Ready     master    3m4s      v1.14.0-alpha.0.1058+73b9effcb6dab5
master2   Ready     master    2m17s     v1.14.0-alpha.0.1058+73b9effcb6dab5
master3   Ready     master    68s       v1.14.0-alpha.0.1058+73b9effcb6dab5
worker    Ready     <none>    11s       v1.14.0-alpha.0.1058+73b9effcb6dab5
~/p/kubernetes-cluster-vagrant (master) > kubectl get pods --all-namespaces -o wide
NAMESPACE     NAME                              READY     STATUS    RESTARTS   AGE       IP              NODE      NOMINATED NODE   READINESS GATES
kube-system   coredns-86c58d9df4-bfdq6          1/1       Running   0          2m58s     10.244.0.3      master1   <none>           <none>
kube-system   coredns-86c58d9df4-vsfjf          1/1       Running   0          2m58s     10.244.0.2      master1   <none>           <none>
kube-system   etcd-master1                      1/1       Running   0          2m3s      172.28.128.21   master1   <none>           <none>
kube-system   etcd-master2                      1/1       Running   3          2m27s     172.28.128.22   master2   <none>           <none>
kube-system   etcd-master3                      1/1       Running   0          78s       172.28.128.23   master3   <none>           <none>
kube-system   kube-apiserver-master1            1/1       Running   0          29s       172.28.128.21   master1   <none>           <none>
kube-system   kube-apiserver-master2            1/1       Running   2          2m27s     172.28.128.22   master2   <none>           <none>
kube-system   kube-apiserver-master3            1/1       Running   0          78s       172.28.128.23   master3   <none>           <none>
kube-system   kube-controller-manager-master1   1/1       Running   1          2m5s      172.28.128.21   master1   <none>           <none>
kube-system   kube-controller-manager-master2   1/1       Running   0          2m27s     172.28.128.22   master2   <none>           <none>
kube-system   kube-controller-manager-master3   1/1       Running   0          78s       172.28.128.23   master3   <none>           <none>
kube-system   kube-flannel-ds-amd64-2jgn4       1/1       Running   2          22s       172.28.128.24   worker    <none>           <none>
kube-system   kube-flannel-ds-amd64-77nsm       1/1       Running   0          2m27s     172.28.128.22   master2   <none>           <none>
kube-system   kube-flannel-ds-amd64-wzcbx       1/1       Running   0          74s       172.28.128.23   master3   <none>           <none>
kube-system   kube-flannel-ds-amd64-zgnc5       1/1       Running   0          2m58s     172.28.128.21   master1   <none>           <none>
kube-system   kube-proxy-9vz98                  1/1       Running   0          2m27s     172.28.128.22   master2   <none>           <none>
kube-system   kube-proxy-h82kz                  1/1       Running   0          2m58s     172.28.128.21   master1   <none>           <none>
kube-system   kube-proxy-j9n4c                  1/1       Running   0          22s       172.28.128.24   worker    <none>           <none>
kube-system   kube-proxy-rc7nm                  1/1       Running   0          74s       172.28.128.23   master3   <none>           <none>
kube-system   kube-scheduler-master1            1/1       Running   1          2m5s      172.28.128.21   master1   <none>           <none>
kube-system   kube-scheduler-master2            1/1       Running   0          2m27s     172.28.128.22   master2   <none>           <none>
kube-system   kube-scheduler-master3            1/1       Running   0          78s       172.28.128.23   master3   <none>           <none>

A load balancer (haproxy) will be created, what will be the entry point for all master node apiservers. The kubeconfig file that will get generated in your $HOME/.kube/config will include the reference to this load balancer IP address.

SSH'ing into the machines

You can use vagrant ssh normally, but you need to point to the PROFILE in order to refer into what machine you want to SSH into. If you have created a cluster with, say:

~/p/kubernetes-cluster-vagrant (master) > PROFILE=bootstrap/1-master-1-worker make

Then, you can ssh into two machines, named: master or worker, as the profile shows. So, you can do:

~/p/kubernetes-cluster-vagrant (master) > PROFILE=bootstrap/1-master-1-worker vagrant ssh kubernetes_master
~/p/kubernetes-cluster-vagrant (master) > PROFILE=bootstrap/1-master-1-worker vagrant ssh kubernetes_worker

Take into account that the prefix is the name of the cluster also present in the PROFILE file.

Destroying the cluster

You can destroy the cluster by pointing to the profile using the PROFILE environment variable and calling to make clean. This will also clean up your ~/.kube folder on the host.

~/p/kubernetes-cluster-vagrant (master) > PROFILE=bootstrap/1-master-1-worker make clean
vagrant destroy -f
==> kubernetes_worker: Forcing shutdown of VM...
==> kubernetes_worker: Destroying VM and associated drives...
==> kubernetes_master: Forcing shutdown of VM...
==> kubernetes_master: Destroying VM and associated drives...

In case you also ran make images, make debs or make artifacts, make clean will also remove the docker container created to build kubernetes artifacts.

If you want to destroy the cluster, but keep the docker container for building Kubernetes artifacts faster, you can run make destroy.

Executing any bazel target

It's possible to call any bazel task within the Kubernetes build container, like:

make bazel WHAT="test //cmd/kubeadm/..."

License

kubernetes-cluster-vagrant
Copyright (C) 2018 Rafael Fernández López <ereslibre@ereslibre.es>

This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with this program.  If not, see <https://www.gnu.org/licenses/>.