Install

Kubernetes with Ondat

Ondat integrates transparently with Kubernetes and different distributions such as OpenShift, Rancher, EKS, AKS, GKE, etc. The user can provide standard PVC definitions and Ondat will dynamically provision matching volumes. Ondat presents volumes to containers with standard POSIX mount targets. This enables the Kubelet to mount Ondat volumes using standard linux device files. Checkout device presentation for more details.

Kubernetes and Ondat communicate with each other to perform actions such as creation, deletion or mounting of volumes. The CSI (Container Storage Interface) driver is the standard method of communication. Using CSI, Kubernetes and Ondat communicate over a Unix domain socket.

CSI (Container Storage Interface) Note

CSI is the standard method of communication that enables storage drivers for Kubernetes to release on their own schedule. The CSI standard allows storage vendors to upgrade, update, and enhance their drivers without the need to update Kubernetes source code, or follow Kubernetes release cycles.

CSI was released GA from Kubernetes 1.13. Ondat v2 only supports the use of CSI as a storage driver. In addition, the Ondat Cluster Operator handles the configuration of the CSI driver and its complexity by detecting the version of the Kubernetes installed.

Check out the status of the CSI release cycle in relation with Kubernetes in the CSI project page.

CSI communication is fully supported by Ondat if the cluster is deployed with any supported Linux Distribution.

Kubernetes Upgrades on Managed Services

Managed services that support in place upgrades are fully supported. However, upgrading Kubernetes using green/blue deployments is not supported. This is because nodes are replaced rather than being upgraded. Any data stored on the nodes is lost when new nodes replace the previous ones.

Docker

Some managed Kubernetes platforms such as Azure AKS, enable the ‘Live-Restore’ Docker feature, enabling containers to continue running while Docker is stopped or upgraded. This feature can cause nodes to hang while shutting-down or rebooting, as rather than going through an orderly shutdown, Ondat (and other processes) are killed before the disks are synced and unmounted. Devices in this inaccessible state will log a warning similar to:

Transport endpoint not connected

To prevent this behaviour, we advise disabling this feature by setting

{
    ...
    "live-restore": false
}

in /etc/docker/daemon.json.

Here’s an example Ansible snippet that might be used to achieve this

...
- name: configure /etc/docker/daemon.json
  lineinfile:
      path: /etc/docker/daemon.json
      regexp: '^.*"live-restore": true,$'
      line: '  "live-restore": false,'
      backrefs: yes
  notify: restart docker
...

Note: Use at own risk; you may need to adapt the example to work in your environment