Quickstart Guide

To quickly set up a testing cluster using MetalK8s, you need 3 machines running CentOS 7.4 to which you have SSH access (these can be VMs). Each machine acting as a Kubernetes node (all of them, in this example) also need to have at least one disk available to provision storage volumes.


Give some sizing examples

Defining an Inventory

To tell the Ansible-based deployment system on which machines MetalK8s should be installed, a so-called inventory needs to be provided. This inventory contains a file listing all the hosts comprising the cluster, as well as some configuration.

First, create a directory, e.g. inventory/quickstart-cluster, in which the inventory will be stored. For our setup, we need to create two files. One listing all the hosts, aptly called hosts:

node-01 ansible_host= ansible_user=centos
node-02 ansible_host= ansible_user=centos
node-03 ansible_host= ansible_user=centos





Make sure to change IP-addresses, usernames etc. according to your infrastructure.

In a second file, called kube-node.yml in a group_vars subdirectory of our inventory, we declare how to setup storage (in the default configuration) on hosts in the kube-node group, i.e. hosts on which Pods will be scheduled:

      drives: ['/dev/vdb']

In the above, we assume every kube-node host has a disk available as /dev/vdb which can be used to set up Kubernetes PersistentVolumes. For more information about storage, see Storage Architecture.

Entering the MetalK8s Shell

To easily install a supported version of Ansible and its dependencies, as well as some Kubernetes tools (kubectl and helm), we provide a make target which installs these in a local environment. To enter this environment, run make shell (this takes a couple of seconds on first run):

$ make shell
Creating virtualenv...
Installing Python dependencies...
Downloading kubectl...
Downloading Helm...
Launching MetalK8s shell environment. Run 'exit' to quit.
(metal-k8s) $

Now we’re all set to deploy a cluster:

(metal-k8s) $ ansible-playbook -i inventory/quickstart-cluster -b playbooks/deploy.yml

Grab a coffee and wait for deployment to end.

Inspecting the cluster

Once deployment finished, a file containing credentials to access the cluster is created: inventory/quickstart-cluster/artifacts/admin.conf. We can export this location in the shell such that the kubectl and helm tools know how to contact the cluster kube-master nodes, and authenticate properly:

(metal-k8s) $ export KUBECONFIG=`pwd`/inventory/quickstart-cluster/artifacts/admin.conf

Now, assuming port 6443 on the first kube-master node is reachable from your system, we can e.g. list the nodes:

(metal-k8s) $ kubectl get nodes
NAME        STATUS    ROLES            AGE       VERSION
node-01     Ready     master,node      1m        v1.9.5+coreos.0
node-02     Ready     master,node      1m        v1.9.5+coreos.0
node-03     Ready     master,node      1m        v1.9.5+coreos.0

or list all pods:

(metal-k8s) $ kubectl get pods --all-namespaces
NAMESPACE      NAME                                                   READY     STATUS      RESTARTS   AGE
kube-ingress   nginx-ingress-controller-9d8jh                         1/1       Running     0          1m
kube-ingress   nginx-ingress-controller-d7vvg                         1/1       Running     0          1m
kube-ingress   nginx-ingress-controller-m8jpq                         1/1       Running     0          1m
kube-ingress   nginx-ingress-default-backend-6664bc64c9-xsws5         1/1       Running     0          1m
kube-ops       alertmanager-kube-prometheus-0                         2/2       Running     0          2m
kube-ops       alertmanager-kube-prometheus-1                         2/2       Running     0          2m
kube-ops       es-client-7cf569f5d8-2z974                             1/1       Running     0          2m
kube-ops       es-client-7cf569f5d8-qq4h2                             1/1       Running     0          2m
kube-ops       es-data-cd5446fff-pkmhn                                1/1       Running     0          2m
kube-ops       es-data-cd5446fff-zzd2h                                1/1       Running     0          2m
kube-ops       es-exporter-elasticsearch-exporter-7df5bcf58b-k9fdd    1/1       Running     3          1m

Similarly, we can list all deployed Helm applications:

(metal-k8s) $ helm list
NAME                    REVISION        UPDATED                         STATUS          CHART                           NAMESPACE
es-exporter             3               Wed Apr 25 23:10:13 2018        DEPLOYED        elasticsearch-exporter-0.1.2    kube-ops
fluentd                 3               Wed Apr 25 23:09:59 2018        DEPLOYED        fluentd-elasticsearch-0.1.4     kube-ops
heapster                3               Wed Apr 25 23:09:37 2018        DEPLOYED        heapster-0.2.7                  kube-system
kibana                  3               Wed Apr 25 23:10:06 2018        DEPLOYED        kibana-0.2.2                    kube-ops
kube-prometheus         3               Wed Apr 25 23:09:22 2018        DEPLOYED        kube-prometheus-0.0.33          kube-ops
nginx-ingress           3               Wed Apr 25 23:09:09 2018        DEPLOYED        nginx-ingress-0.11.1            kube-ingress
prometheus-operator     3               Wed Apr 25 23:09:14 2018        DEPLOYED        prometheus-operator-0.0.15      kube-ops

Cluster Services

Various services to operate and monitor your MetalK8s cluster are provided. To access these, first create a secure tunnel into your cluster by running kubectl proxy. Then, while the tunnel is up and running, the following tools are available:

Service Role Link Notes
Kubernetes dashboard A general purpose, web-based UI for Kubernetes clusters http://localhost:8001/api/v1/namespaces/kube-system/services/https:kubernetes-dashboard:/proxy/  
Grafana Monitoring dashboards for cluster services http://localhost:8001/api/v1/namespaces/kube-ops/services/kube-prometheus-grafana:http/proxy/  
Cerebro An administration and monitoring console for Elasticsearch clusters http://localhost:8001/api/v1/namespaces/kube-ops/services/cerebro:http/proxy/ When accessing Cerebro, connect it to http://elasticsearch:9200 to operate the MetalK8s Elasticsearch cluster.
Kibana A search console for logs indexed in Elasticsearch http://localhost:8001/api/v1/namespaces/kube-ops/services/http:kibana:/proxy/ When accessing Kibana for the first time, set up an index pattern for the logstash-* index, using the @timestamp field as Time Filter field name.

See Cluster Services for more information about these services and their configuration.