Introduction

Longhorn is a CNCF project providing distributed block storage for Kubernetes. It aggregates local storage from your cluster nodes and presents it as persistent, replicated volumes for stateful applications. Key features include snapshots, backups (to S3/NFS), and an intuitive UI.

Best Practices & Configuration

Disk Allocation

  • Dedicated Disks: It is recommended to use a dedicated disk. If you do, you can lower the minimal available storage percentage setting to 10%.
  • Root Disks: If using the root disk, set the minimal available storage percentage to 25% and the overprovisioning percentage to 100% to minimize the chance of DiskPressure.

PostgreSQL (CloudNativePG) Integration

  • For cloudnative-pg deployments, strict-local is the recommended option for data-locality. More details can be found here.
  • You should define a StorageClass specifically tailored for PostgreSQL purposes.

Example PostgreSQL StorageClass

kind: StorageClass
apiVersion: storage.k8s.io/v1
metadata:
  name: hyper-converged
provisioner: driver.longhorn.io
allowVolumeExpansion: true
parameters:
  numberOfReplicas: "1"
  dataLocality: "strict-local"
  staleReplicaTimeout: "120" # 2 hours in minutes
  fsType: "ext4"
  fromBackup: ""
  dataEngine: "v1"

Snapshots vs. Backups

It's crucial to understand the difference between snapshots and backups:

  • Snapshot: A point-in-time, read-only image of a volume's data.

    • It is stored locally alongside the volume's replicas inside your cluster.
    • Snapshots are fast to create and useful for quick rollbacks.
    • Note: Removing the original PVC will also remove the snapshot by default unless you change the ReclaimPolicy of your storage class to Retain.

  • Backup: A copy of a snapshot's data.

    • It is stored externally in a remote location (your Backup Target, like S3 or NFS).
    • Backups are meant for disaster recovery and are safe even if your entire cluster fails.

References