Kubernetes backup options

[mkretzer]

Hello,

we are currently searching for a way to use Kubernetes on our VMware cluster with persistent storage class without loosing the benefits of Veeam. We tried:

- NFS: does not work with every application (because of locking issues) but is simple to backup
- ISCSI-Provisioner: quite unstable as it seems
- vSphere Storage for Kubernetes: Very stable and nearly perfect but it uses independent volumes in the current release

We also looked at other block storage solutions for K8S:

- Rancher Longhorn (still unstable)
- Ceph
- Cinder

Since all these solutions distribute the disks over multiple hosts by design backup via Veeam is quite impossible because the different nodes have different backup points and inconistencies will happen... We also do not really need the redundancy as our SAN which hosts our VM storage has multiple levels of mirroring and our VMs have a very high uptime...

How do other companies resolve this issue? What does Veeam recommend?

Or are backups in "cloud native" applications not possible without going back to backups solely on the appkication level?

Markus

[anthonyspiteri79]

Hey there Markus.

This is an interesting area and one that we have people researching the viability and validity of protecting containers and Kubernetes. As you have pointed out, you need to think differently to traditional backup mentality. The answer might lie in the application level and being aware of the underlying components.

Something others may be able to comment on.

[dellock6]

Can I ask what data need to be protected? Because containers themselves should be stateless, thus without any need to be backed up. If you are looking for the configuration, it's stored into the etcd database, which is a single machine.

[mkretzer]

We are currently designing an archive system which also needs some statefull data, for example a search index. The archive data is using NFS but the index needs to be on something else because of locking issues with NFS. Every index pod needs its own disk which is not shared between the pods. We need multiple pods because of scaling.
Also the external developer we are working with had the idea that at the current time everything should be in K8S including the database and NFS server. We are not sure if that is such a good idea, normally we would do such things which are fast enough on single VMs (which we do not need to skale so fast) on VMs out of the K8S cluster.

We are currently testing vSphere Storage for Kubernetes again and will backup the index data manually. According to github:
"However In vSphere 6.7 release, we can create and manage virtual storage objects independent of VM life cycle and this also has snapshot and backup support. "

The quiestion is just if Veeam will support that at some point.