GCS Deployment scripts

This repository contains playbooks to deploy a GCS cluster on Kubernetes. It also contains a Vagrantfile to setup a local GCS cluster using vagrant-libvirt.

IMP: Clone this repository with submodules

Available playbooks

deploy-k8s.yml

This playbook deploys a kubernetes cluster on the configured nodes, and creates a copy of the kube-config to allow kubectl from the Ansible host. Kubernetes is configured to use Flannel as the network plugin.

TODO: Describe the minimum required ansible inventory format. Till then ./kubespray/inventory/sample/hosts.ini

deploy-gcs.yml

This playbook deploys GCS on a Kubernetes cluster. All of GCS, except for the StorageClass is deployed into its own namespace. The playbook deploys the following,

• etcd-operator
• etcd-cluster with 3 nodes
• glusterd2-cluster with a pod on each kube node and configured to use the deployed etcd-cluster
• glusterd2-client service providing a single rest access point to GD2
• csi-driver (provisioner, nodeplugin, attacher) configured to use glusterd2-client service to reach GD2

Uses the inventory defined for deploy-k8s.yml.

At the moment this is a playbook, but it really shuold become a role. TODO: Convert to role TODO: Add more configurability

add-devices.yml

This playbook adds block devices to the glusterd2-cluster to allow smart volume creation to work. At the moment it works only for the Vagrant provisioned environment.

TODO: Make more universal TODO: Convert to role

vagrant-playbook.yml

This playbook combines all the above to provision the local cluster brought up by Vagrant

Helper scripts

prepare.sh

This scripts prepares the local environement to run the deployment.

How to use

Requirements

• ansible
• python-virtualenv

Optional

• vagrant, vagrant-libvirt - Only required if you want to bring up the vagrant powered local cluster
• kubectl - Only required if you plan to manage the kube cluster from the ansible host

External kube cluster

TODO: Write up how to use the deploy-gcs.yml playbook against an external K8S cluster.

Local cluster using Vagrant

The provided Vagrantfile brings up a 3-node kubernetes cluster, with 3 1000GB virtual disks attached to each node.

Note that the virtual disks are created in the default storage pool of libvirt, though they are all sparse images. The default storage pool is set to /var/lib/libvirt/images, so if used the cluster is used heavily for GCS testing, it is possible to run out of space on your root partition.

• Run the prepare.sh script to perform perliminary checks and prepare the virtualenv.

$ ./prepare.sh

• Activate the virtualenv

$ source gcs-venv/bin/activate
(gcs-venv) $

• Bring up the cluster. This will take a long time.

(gcs-venv) $ vagrant up

Once begin using the cluster once its up by either,

• SSHing into one of the Kube nodes and running the kubectl commands in there

(gcs-venv) $ vagrant ssh kube1

Resetting the cluster

If the Vagrant vms are restarted (say because of host reboot), the kubernete cluster cannot come back up. In such a case, reset the cluster and re-deploy kube and GCS.

(gcs-venv) $ ansible-playbook --become vagrant-reset.yml
.
.
.
(gcs-venv) $ ansible-playbook --skip-tags predeploy --become vagrant-playbook.yml
.
.
.

This will create a brand new cluster. If this fails, destroy the vagrant environment and restart fresh.

(gcs-venv) $ vagrant destroy -f
.
.
.
(gcs-venv) $ vagrant up

Deploying an app with GCS backed volume

An example config is provided for creating a PersistentVolumeClaim against GCS and using the claim in an app, in examples/app-using-gcs-volume.yml

Deploy it with,

(gcs-venv) $ ./kubectl create -f examples/app-using-gcs-volume.yml