3.4 Configure Storage¶

Each Talos node requires at least one disk for the operating system. Worker nodes typically have an additional data disk for persistent volumes (used by Rook-Ceph, Longhorn, or similar CSI drivers).

Volume Layout¶

Volume	Purpose	Lifecycle
OS disk	Talos OS, STATE and EPHEMERAL partitions	Tied to node lifecycle
Data disk (optional)	Persistent volumes for workloads	Can persist independently

Sizing Guidelines¶

Role	OS Disk	Data Disk	Notes
Control plane	50-100 GB	Optional (0-100 GB)	etcd data lives on OS disk
Worker	50-100 GB	100-500 GB	Size depends on workload storage needs

AWSBare MetalProxmox VMs

The storage module (terraform/modules/aws/ebs) creates encrypted EBS gp3 volumes for Talos nodes. These are attached to EC2 instances by the compute module. This is deployed as part of terraform apply via terraform/cluster/aws/main.tf.

Step 1: Configure Storage Variables¶

Open terraform/cluster/envs/aws.tfvars and set the volume sizes:

terraform/cluster/envs/aws.tfvars

root_volume_size          = 50   # OS root volume per node (gp3)
control_plane_volume_size = 0    # Set to 0 to skip CP data volumes
worker_volume_size        = 100  # Data volume per worker (gp3)

The variable defaults in terraform/cluster/aws/variables.tf are:

terraform/cluster/aws/variables.tf

variable "root_volume_size" {
  description = "OS root EBS volume size in GB for all nodes (gp3)"
  type        = number
  default     = 20
}

variable "control_plane_volume_size" {
  description = "EBS data volume size in GB for control plane nodes (gp3). Set to 0 to skip."
  type        = number
  default     = 100
}

variable "worker_volume_size" {
  description = "EBS data volume size in GB for worker nodes (gp3)"
  type        = number
  default     = 200
}

Note

Setting control_plane_volume_size = 0 skips creating separate data volumes for control plane nodes. The demo environment uses this to save cost:

terraform/cluster/envs/aws.tfvars

root_volume_size          = 50
control_plane_volume_size = 0    # No separate CP data volume
worker_volume_size        = 100

Volume Layout Per Node¶

Each node has up to two EBS volumes:

Volume	Device	Created In	Lifecycle
Root	`/dev/xvda`	`compute` module (inline with EC2)	Deleted with instance (`delete_on_termination = true`)
Data	`/dev/xvdf`	`ebs` module (separate resource)	Persists independently of instance

Step 2: Understand the Module¶

The module is at terraform/modules/aws/ebs/.

Worker Volumes¶

One data volume per worker node, placed in the same AZ:

terraform/modules/aws/ebs/main.tf

resource "aws_ebs_volume" "worker_nodes" {
  count             = var.worker_node_count
  availability_zone = var.availability_zones[count.index % length(var.availability_zones)]
  size              = var.worker_volume_size
  type              = "gp3"
  iops              = 3000
  throughput        = 125
  encrypted         = true

  tags = merge(local.common_tags, {
    Name      = "talos-worker-${count.index + 1}"
    NodeType  = "worker"
    NodeIndex = count.index + 1
  })
}

Key points:

availability_zone -- round-robin across configured AZs using modulo (count.index % length)
type = "gp3" -- latest generation general purpose SSD
iops = 3000 / throughput = 125 -- gp3 baseline performance
encrypted = true -- EBS encryption at rest using the default AWS-managed KMS key

Control Plane Volumes (Conditional)¶

CP volumes are only created when control_plane_volume_size > 0. This logic is handled in the root module:

terraform/cluster/aws/main.tf

module "storage" {
  source = "../../modules/aws/ebs"

  control_plane_count       = var.control_plane_volume_size > 0 ? var.control_plane_count : 0
  worker_node_count         = var.worker_count
  control_plane_volume_size = var.control_plane_volume_size > 0 ? var.control_plane_volume_size : 50
  worker_volume_size        = var.worker_volume_size
  availability_zones        = var.availability_zones
  environment               = var.environment
  tags                      = local.tags
}

When control_plane_volume_size = 0, the control_plane_count passed to the module is 0, so no CP volumes are created.

Volume Attachment¶

The actual attachment of volumes to EC2 instances happens in the compute module (terraform/modules/aws/compute/main.tf), not in the ebs module. The storage module exports volume IDs that the compute module consumes:

terraform/modules/aws/compute/main.tf

resource "aws_volume_attachment" "worker" {
  count = var.worker_count

  device_name = "/dev/xvdf"
  volume_id   = var.worker_volume_ids[count.index]
  instance_id = aws_instance.worker[count.index].id
}

Step 3: Module Outputs¶

The ebs module exports:

Output	Consumed By
`control_plane_volume_ids`	`compute` module (volume attachments at `/dev/xvdf`)
`worker_volume_ids`	`compute` module (volume attachments at `/dev/xvdf`)
`total_storage_capacity_gb`	Informational (total provisioned capacity)

Detailed volume information is available via:

# Volume IDs
terraform output -json worker_volume_ids

# Detailed volume info (size, IOPS, throughput, AZ, encryption status)
terraform output -json worker_volumes

Customisation Summary¶

What to Change	Where	Variable
Root volume size	`aws.tfvars`	`root_volume_size`
CP data volume size	`aws.tfvars`	`control_plane_volume_size` (0 = none)
Worker data volume size	`aws.tfvars`	`worker_volume_size` (minimum 50 GB)

Warning

IOPS (3000) and throughput (125 MB/s) are hardcoded in the ebs module. To change these, edit terraform/modules/aws/ebs/main.tf directly.

Bare metal storage uses the server's physical disks. Talos partitions the install disk automatically.

Disk Layout¶

When Talos installs to a disk, it creates:

Partition	Purpose	Size
EFI	Boot partition	100 MB
BIOS	Legacy boot	1 MB
BOOT	Talos boot assets	1 GB
META	Machine metadata	1 MB
STATE	Machine configuration, etcd data	100 MB
EPHEMERAL	Kubernetes pods, logs, containerd	Remaining space

Install Disk Selection¶

Specify the install disk in the Talos machine config. Common device paths:

Device	Type
`/dev/sda`	SATA/SAS SSD or HDD
`/dev/nvme0n1`	NVMe SSD
`/dev/vda`	VirtIO (unlikely on bare metal)

Check available disks on a booted server:

talosctl -n <node-ip> disks

Data Disks¶

Additional disks for persistent volumes are not managed by Talos. They are discovered by the CSI driver (Rook-Ceph) running in the cluster.

For Rook-Ceph: - Leave data disks unpartitioned and unformatted - Rook-Ceph discovers and provisions OSDs on them automatically

RAID Configuration¶

If your servers have hardware RAID controllers: - Configure RAID before booting Talos (via BIOS/UEFI or IPMI) - Present a single logical disk to Talos as the install target - For software RAID, Talos does not support mdadm — use hardware RAID or single disks

Proxmox storage is managed through Proxmox storage pools. The Terraform VM module creates virtual disks on the configured storage pool.

Storage Pool¶

All VM disks are created on the pool specified by storage_pool (default: local-lvm):

# In envs/proxmox.tfvars
storage_pool = "local-lvm"

Verify the pool exists on your Proxmox node:

pvesm status

Disk Configuration¶

The VM module creates disks with these settings:

# From terraform/modules/proxmox/vm/main.tf
disk {
  datastore_id = var.storage_pool
  interface    = "scsi0"          # OS disk
  size         = var.disk_size_gb
  iothread     = true             # Improved performance
  discard      = "on"             # TRIM/discard support
}

Worker nodes get an optional data disk on scsi1:

dynamic "disk" {
  for_each = var.data_disk_size_gb > 0 ? [1] : []
  content {
    datastore_id = coalesce(var.data_disk_storage, var.storage_pool)
    interface    = "scsi1"        # Data disk
    size         = var.data_disk_size_gb
    iothread     = true
    discard      = "on"
    file_format  = "raw"
  }
}

Proxmox Environment Sizes¶

Node Type	OS Disk (scsi0)	Data Disk (scsi1)
Control plane	50 GB	None
Worker	50 GB	50 GB

These are set in envs/proxmox.tfvars:

control_plane_disk_size_gb = 50
worker_disk_size_gb        = 50
worker_data_disk_size_gb   = 50

Storage Backend Recommendations¶

Backend	Type	Best For
`local-lvm`	LVM thin	Default, good performance
`local-zfs`	ZFS	Snapshots, compression, dedup
NFS/CIFS	Network	Shared storage (slower)

For production, ZFS on local NVMe provides the best combination of performance and data protection.

Data Disk Usage¶

The data disk (scsi1) appears as /dev/sdb inside the VM. It is not managed by Talos — leave it for the CSI driver:

Rook-Ceph: Discovers /dev/sdb and provisions OSDs automatically
Longhorn: Uses the disk for replica storage