c12s-kubespray/docs/getting-started.md
2020-06-25 11:56:37 -07:00

6.6 KiB

Getting started

Building your own inventory

Ansible inventory can be stored in 3 formats: YAML, JSON, or INI-like. There is an example inventory located here.

You can use an inventory generator to create or modify an Ansible inventory. Currently, it is limited in functionality and is only used for configuring a basic Kubespray cluster inventory, but it does support creating inventory file for large clusters as well. It now supports separated ETCD and Kubernetes master roles from node role if the size exceeds a certain threshold. Run python3 contrib/inventory_builder/inventory.py help for more information.

Example inventory generator usage:

cp -r inventory/sample inventory/mycluster
declare -a IPS=(10.10.1.3 10.10.1.4 10.10.1.5)
CONFIG_FILE=inventory/mycluster/hosts.yml python3 contrib/inventory_builder/inventory.py ${IPS[@]}

Then use inventory/mycluster/hosts.yml as inventory file.

Starting custom deployment

Once you have an inventory, you may want to customize deployment data vars and start the deployment:

IMPORTANT: Edit my_inventory/groups_vars/*.yaml to override data vars:

ansible-playbook -i inventory/mycluster/hosts.yml cluster.yml -b -v \
  --private-key=~/.ssh/private_key

See more details in the ansible guide.

Adding nodes

You may want to add worker, master or etcd nodes to your existing cluster. This can be done by re-running the cluster.yml playbook, or you can target the bare minimum needed to get kubelet installed on the worker and talking to your masters. This is especially helpful when doing something like autoscaling your clusters.

  • Add the new worker node to your inventory in the appropriate group (or utilize a dynamic inventory).
  • Run the ansible-playbook command, substituting cluster.yml for scale.yml:
ansible-playbook -i inventory/mycluster/hosts.yml scale.yml -b -v \
  --private-key=~/.ssh/private_key

Remove nodes

You may want to remove master, worker, or etcd nodes from your existing cluster. This can be done by re-running the remove-node.yml playbook. First, all specified nodes will be drained, then stop some kubernetes services and delete some certificates, and finally execute the kubectl command to delete these nodes. This can be combined with the add node function. This is generally helpful when doing something like autoscaling your clusters. Of course, if a node is not working, you can remove the node and install it again.

Use --extra-vars "node=<nodename>,<nodename2>" to select the node(s) you want to delete.

ansible-playbook -i inventory/mycluster/hosts.yml remove-node.yml -b -v \
--private-key=~/.ssh/private_key \
--extra-vars "node=nodename,nodename2"

If a node is completely unreachable by ssh, add --extra-vars reset_nodes=no to skip the node reset step. If one node is unavailable, but others you wish to remove are able to connect via SSH, you could set reset_nodes=no as a host var in inventory.

Connecting to Kubernetes

By default, Kubespray configures kube-master hosts with insecure access to kube-apiserver via port 8080. A kubeconfig file is not necessary in this case, because kubectl will use http://localhost:8080 to connect. The kubeconfig files generated will point to localhost (on kube-masters) and kube-node hosts will connect either to a localhost nginx proxy or to a loadbalancer if configured. More details on this process are in the HA guide.

Kubespray permits connecting to the cluster remotely on any IP of any kube-master host on port 6443 by default. However, this requires authentication. One can get a kubeconfig from kube-master hosts (see below) or connect with a username and password.

For more information on kubeconfig and accessing a Kubernetes cluster, refer to the Kubernetes documentation.

Accessing Kubernetes Dashboard

Supported version is kubernetes-dashboard v2.0.x :

  • Login options are : token/kubeconfig by default, basic can be enabled with kube_basic_auth: true inventory variable - not recommended because this requires ABAC api-server which is not tested by kubespray team
  • Deployed by default in "kube-system" namespace, can be overridden with dashboard_namespace: kubernetes-dashboard in inventory,
  • Only serves over https

Access is described in dashboard docs. With kubespray's default deployment in kube-system namespace, instead of kuberntes-dashboard :

Accessing through Ingress is highly recommended. For proxy access, please note that proxy must listen to localhost (proxy --address="x.x.x.x" will not work)

For token authentication, guide to create Service Account is provided in dashboard sample user doc. Still take care of default namespace.

Access can also by achieved via ssh tunnel on a master :

# localhost:8081 will be sent to master-1's own localhost:8081
ssh -L8001:localhost:8001 user@master-1
sudo -i
kubectl proxy

Accessing Kubernetes API

The main client of Kubernetes is kubectl. It is installed on each kube-master host and can optionally be configured on your ansible host by setting kubectl_localhost: true and kubeconfig_localhost: true in the configuration:

  • If kubectl_localhost enabled, kubectl will download onto /usr/local/bin/ and setup with bash completion. A helper script inventory/mycluster/artifacts/kubectl.sh also created for setup with below admin.conf.
  • If kubeconfig_localhost enabled admin.conf will appear in the inventory/mycluster/artifacts/ directory after deployment.
  • The location where these files are downloaded to can be configured via the artifacts_dir variable.

You can see a list of nodes by running the following commands:

cd inventory/mycluster/artifacts
./kubectl.sh get nodes

If desired, copy admin.conf to ~/.kube/config.