Skip to content

StorageClass Parameters

All MeshStor StorageClass parameters, their types, defaults, and constraints. Parameters are set in the parameters field of a Kubernetes StorageClass resource.

Provisioner

provisioner: io.meshstor.csi.mesh

Parameters

Parameter Type Default Minimum Description
replicaCount string (integer) "2" "1" Number of data replicas. Each replica is placed on a different node. Higher values increase fault tolerance and storage overhead.
stripeWidth string (integer) "1" "1" Stripe width — number of drives a single replica is striped across. Value 1 create RAID1, values >1 create RAID10 (Stripe of mirrors). Minimum node count is still replicaCount; each of those nodes must have at least stripeWidth NVMe drives available.
memberMissingTimeout string (integer, seconds) "900" "60" How long to wait before replacing a missing RAID member. Lower values recover faster but may trigger unnecessary rebuilds during transient failures.
sectorSize string (integer, bytes) "4096" — Logical block size every RAID member must report. Allowed values: "512" or "4096". Immutable after creation — an MD array's logical block size cannot change once data exists. All members of a volume share this size, so the volume is placed only on nodes whose NVMe drives report it (see warning below).

String values

All parameters are strings in the Kubernetes StorageClass API. MeshStor parses them internally. Invalid values (non-numeric, below minimum, or outside the allowed set) return an InvalidArgument error on volume creation.

Sector size must match your drives

The default sectorSize is "4096". A volume is placed only on nodes that have at least replicaCount drives whose logical block size equals sectorSize; otherwise CreateVolume fails with FailedPrecondition ("not enough nodes"). On clusters whose NVMe drives report 512-byte logical blocks, set sectorSize: "512". Check a drive's reported size with cat /sys/block/<device>/queue/logical_block_size, or the SectorSize column of kubectl get msnd. This constraint prevents a mixed-block-size array, which the kernel would otherwise leave with a stranded, non-rebuilding member.

StorageClass Fields

These standard Kubernetes StorageClass fields are relevant to MeshStor:

Field Recommended Value Notes
reclaimPolicy Delete Retain is supported but leaves orphaned partitions on drives
allowVolumeExpansion true Set to false when stripeWidth > 1 (expansion not supported for RAID10) — see Volume Expansion
volumeBindingMode WaitForFirstConsumer Immediate also works. WaitForFirstConsumer defers CreateVolume until a pod is scheduled, which avoids zombie PVs from PVCs that are never claimed. The driver doesn't consume CSI topology hints today, so the topology-aware benefit of WaitForFirstConsumer doesn't apply — but the binding-deferral benefit does.
allowedTopologies not strictly enforced today The scoring algorithm already places partitions on the best available nodes (RDMA support, free space, network latency, subnet pairing). Strict allowedTopologies enforcement is planned — until it ships, use node labels and the meshstor.io/nvme-over-*-address annotations to shape the candidate set.

Effective RAID Levels

replicaCount stripeWidth RAID Level Total Partitions Storage Overhead Fault Tolerance
1 1 RAID1 (1+placeholder) 2 1x Relocatable, no redundancy
2 1 RAID1 2 2x 1 node failure
3 1 RAID1 3 3x 2 node failures
1 2 RAID0 (2+placeholders) 4 1x Relocatable, no redundancy
2 2 RAID10 4 2x 1 node failure

Example

apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: mesh-2copy-tcp
provisioner: io.meshstor.csi.mesh
parameters:
  replicaCount: "2"
  stripeWidth: "1"
  memberMissingTimeout: "900"
  sectorSize: "4096"
reclaimPolicy: Delete
allowVolumeExpansion: true

See StorageClass Examples for ready-to-use configurations.

What's Next