Other articles


  1. My kubernetes setup

    Updated 2017/01/03: Modified setup to use Xen instead of Qemu for master+nodes, upgrade Kubernetes to 1.5.1, use CoreOS beta instead of alpha.

    This is a description of my local kubernetes setup. If you want to set up kubernetes yourself chances are you should follow the proper guide. This is intended to be the reference that I was desparate for when I set out doing this a few months ago.

    I wanted to run my own kubernetes deployment to run applications and experiment. I didn't just want to try out kubernetes, I wanted to run it 24/7. From the looks of it the easiest way to do this is using Google Compute Engine or AWS. The problem with both of these is to run 24/7 you end up spending quite a lot of money every month just to keep a basic install running.

    After considering a bunch of options (including running a Raspberry Pi Cluster) I came to the conclusion that my best setup would be to run a single physical server that hosted bunch of virtual machines.

    I picked Xen as my hypervisor, Ubuntu as my "dom0" (more on this later) and CoreOS as my kubernetes host. Here's my set up.

    Hardware

    • Dell T20 Server
    • Intel i5-4590
    • 16 GB RAM
    • 120 GB SSD

    Software

    Hypervisor: Xen Hypervisor / Ubuntu 16.04. I found myself thoroughly confused by all this talk of "dom0" but the gist of this is: You install Ubuntu 16.04 on your server, you then install (via apt-get) Xen which installs itself as the main OS with your original Ubuntu install as a virtual machine. This virtual machine is called "dom0" and is what you use to manage all your other virtual machines.

    (Another source of confusion - Xen is not XenServer, which is a commercial product you can safely ignore).

    Kubernetes OS: CoreOS Alpha Channel. Right now Stable does not include the kubelet (which we need) so I'm using Alpha. This is what I picked as it tries to support Kubernetes right out of the box.

    Installing Xen

    On a fresh Ubuntu 16.04, install Xen, libvirt and virtinst. Replace it as the deafult boot point and restart. virtinst gives us a CLI we will use to launch virtual machines later. genisoimage we need for mkisofs.

    sudo apt-get install xen-hypervisor-amd64 virtinst genisoimage
    sudo sed -i 's/GRUB_DEFAULT=.*\+/GRUB_DEFAULT="Xen 4.1-amd64"/' /etc/default/grub
    sudo update-grub
    sudo reboot
    

    What comes back up should be the original Ubuntu install running as a virtual machine on the Xen hypervisor. Because it's the original install we don't know for sure that anything actually changed. We can check with xl:

    root@xen:~# xl list
    Name                                        ID   Mem VCPUs      State   Time(s)
    Domain-0                                     0 19989     4     r-----      75.3
    

    Looks good!

    Installing Kubernetes

    Kubernetes comes with these nifty scripts that basically set up your whole cluster for you. The problem I found with this is I wanted to manage (and understand) the pieces of software myself. I didn't want a mysterious bash script that promised to take care of it all for me.

    Instead I've created my own set of mysterious scripts, that are slightly less generated and templated that may be useful to some as examples. This is how to use them.

    We're going to use as little as possible of my stuff - the following git repo is 4 CoreOS cloud-config files. These define basic configuration (network setup, applications to run). There's also a piece of config to generate our SSL certificate for the cluster.

    So, grab my config from Github and grab the latest CoreOS Alpha:

    sudo su
    mkdir -p /var/lib/libvirt/images/
    cd /var/lib/libvirt/images/
    git clone -b blog_post_v2 https://github.com/andrewmichaelsmith/xen-coreos-kube.git coreos
    cd coreos
    wget https://beta.release.core-os.net/amd64-usr/current/coreos_production_xen_image.bin.bz2 -O - | bzcat > coreos_production_xen_image.bin
    

    Now create a disk and config disk for master, node1, node2, node3:

    #To pad out extra space in the image.
    #TODO: Undoubtedly a better way than this
    dd if=/dev/zero of=tempfile bs=1G count=2
    cat tempfile >> coreos_production_xen_image.bin
    
    cp coreos_production_xen_image.bin master1.bin
    cp coreos_production_xen_image.bin node1.bin
    cp coreos_production_xen_image.bin node2.bin
    cp coreos_production_xen_image.bin node3.bin
    

    You may need to generate an SSH key if you haven't already:

    ssh-keygen -t rsa -b 4096 -C "$USER@$HOSTNAME"
    

    We then put our SSH key in to the cloud-configs for our nodes:

    KEY=$(cat ~/.ssh/id_rsa.pub)
    sed "s#SSH_KEY#$KEY#g" < master1/openstack/latest/user_data.tmpl > master1/openstack/latest/user_data
    sed "s#SSH_KEY#$KEY#g" < node1/openstack/latest/user_data.tmpl > node1/openstack/latest/user_data
    sed "s#SSH_KEY#$KEY#g" < node2/openstack/latest/user_data.tmpl > node2/openstack/latest/user_data
    sed "s#SSH_KEY#$KEY#g" < node3/openstack/latest/user_data.tmpl > node3/openstack/latest/user_data
    

    We also need to generate our certificates:

    cd certs
    openssl genrsa -out ca-key.pem 2048
    openssl req -x509 -new -nodes -key ca-key.pem -days 10000 -out ca.pem -subj "/CN=kube-ca"
    openssl genrsa -out apiserver-key.pem 2048
    openssl req -new -key apiserver-key.pem -out apiserver.csr -subj "/CN=kube-apiserver" -config openssl.cnf
    openssl x509 -req -in apiserver.csr -CA ca.pem -CAkey ca-key.pem -CAcreateserial -out apiserver.pem -days 365 -extensions v3_req -extfile openssl.cnf
    cd ..
    

    And then put the certificates we generated in to the master node:

    #Total hack, so it's indented correctly when we move it in to .yml
    sed -i 's/^/        /' certs/*.pem
    sed -i $'/CA.PEM/ {r certs/ca.pem\n d}' master1/openstack/latest/user_data
    sed -i $'/APISERVER.PEM/ {r certs/apiserver.pem\n d}' master1/openstack/latest/user_data
    sed -i $'/APISERVER-KEY.PEM/ {r certs/apiserver-key.pem\n d}' master1/openstack/latest/user_data
    

    Configs done, we can validate to double check:

    curl 'https://validate.core-os.net/validate' -X PUT --data-binary '@master1/openstack/latest/user_data' | python -mjson.tool
    curl 'https://validate.core-os.net/validate' -X PUT --data-binary '@node1/openstack/latest/user_data' | python -mjson.tool
    curl 'https://validate.core-os.net/validate' -X PUT --data-binary '@node2/openstack/latest/user_data' | python -mjson.tool
    curl 'https://validate.core-os.net/validate' -X PUT --data-binary '@node3/openstack/latest/user_data' | python -mjson.tool
    

    If that passed ("null" from the server), first create an iso to get the config file in to our xen vm:

    mkisofs -R -V config-2 -o master1-config.iso master1/
    mkisofs -R -V config-2 -o node1-config.iso node1/
    mkisofs -R -V config-2 -o node2-config.iso node2/
    mkisofs -R -V config-2 -o node3-config.iso node3/
    

    Then create the CoreOS virtual machines using those disks and cloud-configs:

    xl create master1.cfg
    xl create node1.cfg
    xl create node2.cfg
    xl create node3.cfg
    

    This will start 4 virtual machines running CoreOS and our cloud configs.

    What happens:

    • Download flannel image
    • Kubelet starts and downloads hyperkube
    • Containers started for api server, controller manager, scheduler on master
    • Container for kube-proxy starts on on nodes

    If you need to you can attach to the console and monitor a node booting up:

    xl console master1
    

    You can also ssh on to the master and check journalctl:

    ssh core@192.168.1.254 
    journalctl -f
    

    So.. did it work? Let's try using kubectl (which we need to install locally first):

    curl -O https://storage.googleapis.com/kubernetes-release/release/v1.2.3/bin/linux/amd64/kubectl
    chmod +x kubectl
    mv kubectl /usr/local/bin/kubectl
    

    Let's see:

    root@xen# kubectl -s http://192.168.122.254:8080 get nodes
    NAME              STATUS    AGE
    192.168.122.2     Ready     1m
    192.168.122.254   Ready     1m
    192.168.122.3     Ready     1m
    192.168.122.4     Ready     1m
    

    One last thing, if we try and list the pods (running processes) we won't get anything. We need to create the "kube-system" namespace. Which can be easily done:

    curl -H "Content-Type: application/json" -XPOST -d'{"apiVersion":"v1","kind":"Namespace","metadata":{"name":"kube-system"}}' "http://192.168.122.254:8080/api/v1/namespaces"
    

    Now:

    NAME                                      READY     STATUS             RESTARTS   AGE
    kube-apiserver-192.168.122.254            1/1       Running            0          3m
    kube-controller-manager-192.168.122.254   1/1       Running            1          4m
    kube-proxy-192.168.122.2                  1/1       Running            1          4m
    kube-proxy-192.168.122.254                1/1       Running            0          3m
    kube-proxy-192.168.122.3                  1/1       Running            0          3m
    kube-proxy-192.168.122.4                  1/1       Running            0          3m
    kube-scheduler-192.168.122.254            1/1       Running            0          3m
    

    Woohoo!

    Conclusion

    So what have we actually done? We've turned an Ubuntu server in to a Xen Hypervisor. On that hypervisor we've created 4 virtual machines all running CoreOS. From the CoreOS config from my git repo we've set up 1 CoreOS install running the master kubernetes components, 3 others are running the node components.

    There's many ways we can get Kubernetes running on CoreOS. The particular way we have set it up as is follows.

    • flannel service - This handles our networking. It allows a container on one node to speak to a container on another node.
    • etcd service - This is where kubernetes persists state.
    • docker service - Docker is how this kubernetes setup launches images.
    • kubelet service - This is the only kubernetes component installed as a system service. We use the kubelet to join our kubernetes cluster and launch other kubernetes applications.

    As well as system services we've also installed the following as services managed by kubernetes, we do this by placing kubernetes config in /etc/kubernetes/manifests/. The kubelet service monitors this directory and launches applications based on what it finds.

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

    That's all! We've not got a fully functioning kubernetes cluster. Time to play with it.

    read more

    There are comments.

social