What should replace vSphere CSI when we move our Kubernetes VMs?

November 19, 2026 5 answers
Asked by _x61!tarnelo.3

We run a vanilla Kubernetes cluster in VMs and are evaluating Proxmox and HPE VM Essentials as alternatives to VMware. We expect to use SAN storage. What persistent storage options are worth considering, and what are the tradeoffs?

Answers

_t39+zeldravo.6

Start with the SAN vendor's CSI driver. If it supports your array, Kubernetes version and chosen VM platform, it may be the closest replacement for vSphere CSI without adding another distributed storage system.

Check provisioning, snapshots, volume attachment and failover in a test cluster. Also ask about LUN and CSI operation limits if you expect a lot of volumes. Running replicated storage software on top of an already replicated SAN can add cost and complexity without giving you much extra.

_x61!tarnelo.3:

That sounds like a useful first check. The array and driver compatibility will matter as much as the hypervisor choice.

Reply to this answer
q.371~brindlepocket

Ceph is an option if you want block volumes through RBD or shared files through CephFS. Rook can manage Ceph through Kubernetes, while an existing external Ceph cluster is another route. If Proxmox ends up using Ceph, look into whether Kubernetes can use that storage directly rather than building a second cluster on top.

The catch is operational overhead. Ceph needs careful capacity and failure planning, and its network and disk demands can be significant. For demanding workloads, 10 Gbps networking and SSDs are worth considering, but size and test for your actual workload rather than treating those as universal requirements.

Romi:

There is disagreement about Rook. Some prefer the Kubernetes integration, while others would rather manage Ceph separately because the added layer can make troubleshooting more complex.

Reply to this answer
Trelvani.9x

Longhorn and Piraeus are both worth a look if you want storage managed around the Kubernetes cluster rather than the SAN. Compare their replication, recovery and operational requirements against what your underlying storage already provides.

Run failure and restore tests before choosing either one. A setup that looks simple in a small environment can have different resource and maintenance demands in production.

Reply to this answer
_W26+torvani.9

Not every workload needs the same storage design. A database with its own replication and backups may not need another replication layer beneath it. Object storage can also work for some assets, though plenty of applications still need a normal filesystem. NFS is another option for shared files, but its availability and backup design still need attention.

Reply to this answer
Mavdrix!2c

There are a couple of bigger changes to consider too. Kubernetes could run on bare metal with KubeVirt for workloads that still need VMs, if your hardware and operational requirements fit. Portworx is another storage option, though it comes with licensing costs. Neither is necessary just to replace the CSI driver.

Reply to this answer

Add your answer

Answers are reviewed before publication.

Related Discussions