Right now we have S3-backed PV management for on-premise Kubernetes. It is doubtful to be a good solution for Postgres PV. This PR focus is to introduce PV management that can be used for Postgres

Important

Critical Services (e.g. PostgreSQL) shall avoid using Longhorn and use local disks directly and implement HA in other way (e.g. application level HA like Postgres HA / Cluster setup)

Options considered

• Ceph (via Rook Ceph CSI)
  An open‑source distributed storage platform that supports block, file, and object storage.
  https://rook.io/
  -
  CEPH is complicated to manage. It is [probably] the most mature and robust solution that requires expertise to be run smoothly / robustly on PROD. Since we will try to use application-level HA (e.g. Postgres Cluster) or some other tools (e.g. directly S3), I would stick to Longhorn as it is easier to manage and seems to be robust enough (for sure enough for non-critical services like Portainer)

  • Ask O+P
  • Configureing rook to use "external" ceph is already complicated + requires lots of future collaboration with O+P

• NetApp ONTAP/Trident
  Enterprise‑grade storage integrated with Kubernetes through the NetApp Trident CSI driver.
  https://www.netapp.com/solutions/kubernetes/
  -
  Does not seem to be very popular. Focus on other options

• OpenEBS
  Container‑attached storage offering flexibility and ease‑of‑deployment for on‑prem clusters.
  https://openebs.io/
  -
  Incubating stage. Let's pick something more mature according to CNCF stages

• --> Longhorn <--
  A lightweight, distributed block storage system that leverages local disks with replication for high availability.
  https://longhorn.io/
  -
  Rook Ceph seems to be more reliable https://www.reddit.com/r/kubernetes/comments/1cbggo8/longhorn_is_unreliable/
  BUT https://www.reddit.com/r/kubernetes/comments/1j02w70/is_usedproperly_longhorn_productionready_in_2025/
  YH picks Longhorn

• StorageOS
  A software‑defined storage solution designed for container environments with enterprise‑grade features.
  https://www.storageos.com/
  -
  Does not seem to be very popular. Focus on other options

• GlusterFS
  An open‑source, scalable network filesystem that’s been used as a persistent storage option in many on‑prem deployments.
  https://gluster.org/
  -
  Not actively maintained --> Is the project still alive? gluster/glusterfs#4324

Resources

NFS:

• https://gist.github.com/fardjad/ea358f9bf844889ecad109b352dd0d5b
• https://documentation.ubuntu.com/server/how-to/networking/install-nfs/
• https://www.postgresql.org/docs/devel/creating-cluster.html#CREATING-CLUSTER-FILESYSTEM
• https://www.reddit.com/r/elasticsearch/comments/vjlys8/using_nfs_as_persistent_storage_for_elasticsearch/
• Is NFS production-ready
  • https://www.reddit.com/r/kubernetes/comments/ehqxo9/who_is_using_nfs_for_persistent_volumes/
  • https://stackoverflow.com/questions/73549120/persistent-storage-for-kubernetes-in-production
  • https://devops.stackexchange.com/questions/17004/how-reliable-is-nfs-for-using-in-kubernetes

Longhorn:

• https://www.reddit.com/r/kubernetes/comments/1cllsox/why_should_i_use_longhorn/
• https://www.reddit.com/r/kubernetes/comments/10war0o/can_someone_explain_me_the_true_benefits_of/
• https://www.reddit.com/r/kubernetes/comments/1chude7/would_you_rather_use_rookceph_or_longhorn_for_on/
• https://www.reddit.com/r/kubernetes/comments/1cbggo8/longhorn_is_unreliable/
• https://medium.com/@petolofsson/best-practices-for-longhorn-067d4ccb5fdd
• [FEATURE] Detached volumes not replicating - data loss potential - easy reproduction longhorn/longhorn#3619

Comparisons:

• https://medium.com/volterra-io/kubernetes-storage-performance-comparison-v2-2020-updated-1c0b69f0dcf4
• https://www.reddit.com/r/kubernetes/comments/18njhpk/bare_metal_storage_choices_for_a_homelab/
• Longhorn vs Rook Ceph

S3 CSI (e.g. from yandex):

• https://www.reddit.com/r/kubernetes/comments/1j2l7ec/what_are_the_valid_use_cases_for_s3_csi/