TrueNAS VM Build And Pool Design

This page covers the mechanical part of the move: make room, back up the old naspool, build the VM, pass the disks through properly, and recreate storage as a TrueNAS-managed tank.

The commands below stay close to the original host notes on purpose.

Why Mirror Plus Hot Spare

The source layout uses three 8 TB WD RED drives. For that shape, the original choice was not more raw capacity. It was faster recovery and less manual intervention.

Topology	Drives	Usable Space	Read Speed	Resilver Time (8TB)	Failure Tolerance
Mirror + Hot Spare	2 active + 1 spare	about 8 TB	about 2x single drive	about 4-6 hours	1 drive plus automatic recovery
RAIDZ1 (3 drives)	3 active	about 16 TB	about 2x single drive	about 12-24 hours	1 drive, manual replace
RAIDZ1 (2 drives)	2 active	about 8 TB	about 1x single drive	about 12-24 hours	1 drive, manual replace

The trade is simple: less usable space, shorter vulnerability window.

Physical Drive Map

Port	Drive	Capacity	Linux Device	Role
M.2_2	NVMe	2 TB	`/dev/nvme0n1`	`rpool` - Proxmox OS and container storage
SATA 1	HDD	1 TB	`/dev/sda`	`backup` mirror leg 1
SATA 2	HDD	1 TB	`/dev/sdb`	`backup` mirror leg 2
SATA 3	WD RED NAS HDD	8 TB	`/dev/sdc`	`tank` mirror leg 1
SATA 4	WD RED NAS HDD	8 TB	`/dev/sdd`	`tank` mirror leg 2
SATA 5	WD RED NAS HDD	8 TB	`/dev/sde`	`tank` hot spare

Always pass the WD RED disks through with /dev/disk/by-id/ paths, not /dev/sdX. Device letters are not stable across reboots.

Phase 0: Resource Reallocation

TrueNAS SCALE needs real RAM. The original host made room by trimming the GPU containers and capping host ARC.

Audit current allocation

ssh root@192.168.50.20
 
# Show actual vs allocated RAM per container
for ct in $(pct list | awk 'NR>1 {print $1}'); do
  name=$(pct list | awk -v id="$ct" '$1==id {print $3}')
  alloc=$(pct config $ct | grep memory | awk '{print $2}')
  echo "CT $ct ($name): ${alloc}MB allocated"
done
 
# Host memory overview
free -h
 
# ZFS ARC usage (this is reclaimable memory)
grep -E "^size" /proc/spl/kstat/zfs/arcstats | awk '{printf "ZFS ARC: %.1f GB\n", $3/1024/1024/1024}'

Reduce GPU container RAM allocations

# Reduce Ollama from 30GB -> 16GB
# (Models load into 24GB VRAM, not system RAM. 16GB is plenty for the inference server process)
pct set 100 -memory 16384
 
# Reduce llama-cpp from 30GB -> 16GB
# (Same reasoning - llama-cpp offloads all layers to GPU VRAM)
pct set 102 -memory 16384
 
# Verify changes
pct config 100 | grep memory
pct config 102 | grep memory

Limit host ZFS ARC

# Set host ZFS ARC max to 4GB (sufficient for rpool + backup)
echo "options zfs zfs_arc_max=4294967296" > /etc/modprobe.d/zfs.conf
 
# Apply without reboot
echo 4294967296 > /sys/module/zfs/parameters/zfs_arc_max
 
# Verify
cat /sys/module/zfs/parameters/zfs_arc_max

Phase 1: Physical Install And Pre-Migration

Install the third WD RED and verify detection

# Verify all 3 WD RED drives are visible
lsblk -o NAME,SIZE,MODEL,SERIAL
 
# Expected output should show 3x 8TB drives:
# sdc   7.3T  WDC WD80EFAX-...  <SERIAL1>
# sdd   7.3T  WDC WD80EFAX-...  <SERIAL2>
# sde   7.3T  WDC WD80EFAX-...  <SERIAL3>

Record stable drive identities

# Get stable by-id paths for all 3 WD RED drives
ls -la /dev/disk/by-id/ | grep -E "^l.*ata-WD" | grep -v part
 
# Save output - you'll need these exact paths for VM passthrough
# Example output:
# ata-WDC_WD80EFAX_<SERIAL1> -> ../../sdc
# ata-WDC_WD80EFAX_<SERIAL2> -> ../../sdd
# ata-WDC_WD80EFAX_<SERIAL3> -> ../../sde
 
# Also record serial numbers for reference
smartctl -i /dev/sdc | grep "Serial Number"
smartctl -i /dev/sdd | grep "Serial Number"
smartctl -i /dev/sde | grep "Serial Number"

Back up the existing `naspool`

# Create a final snapshot of everything in naspool
zfs snapshot -r naspool@pre-truenas-migration
 
# Option A: Copy to backup pool (if space permits - 1TB mirror may be too small)
zfs send -R naspool@pre-truenas-migration | zfs receive -F backup/naspool-backup
 
# Option B: Copy to external USB drive (recommended for large datasets)
# Plug in a USB drive, identify it:
lsblk
# Mount it:
mkdir -p /mnt/usb-backup
mount /dev/sdX1 /mnt/usb-backup
 
# Copy data
rsync -avP /naspool/media/ /mnt/usb-backup/naspool-media/
rsync -avP /naspool/documents/ /mnt/usb-backup/naspool-documents/
rsync -avP /naspool/vm-backups/ /mnt/usb-backup/naspool-vm-backups/
 
# Verify backup integrity
ls -laR /mnt/usb-backup/
 
# Unmount USB
umount /mnt/usb-backup

Save current storage and cron state

# Save current vzdump crontab for reference
crontab -l > /root/crontab-backup-pre-truenas.txt
 
# Save naspool properties
zfs get all naspool > /root/naspool-properties-backup.txt
zpool status naspool > /root/naspool-status-backup.txt

Phase 2: Retire `naspool` And Create The VM

Remove `naspool` from Proxmox storage config if needed

# Check if naspool is registered as Proxmox storage
pvesm status | grep naspool
 
# If listed, remove it
pvesm remove naspool
 
# Verify removal
pvesm status

Export or destroy the host-side pool

# Export (cleanly unmount) the pool
zpool export naspool
 
# If export fails due to busy datasets:
zpool export -f naspool
 
# Verify naspool is gone
zpool list
# Should show only: rpool, backup

# Alternative if you want to wipe metadata instead:
zpool destroy naspool
# Then wipe partition tables:
wipefs -a /dev/sdc
wipefs -a /dev/sdd

Download the ISO and create the VM

# Download TrueNAS SCALE ISO (check for latest version at https://www.truenas.com/download-truenas-scale/)
cd /var/lib/vz/template/iso/
 
# Download latest stable release (update URL as needed)
wget https://download.truenas.com/TrueNAS-SCALE-ElectricEel/24.10.2/TrueNAS-SCALE-24.10.2.iso
 
# Verify download
ls -la /var/lib/vz/template/iso/TrueNAS-SCALE-*.iso

# Create VM
qm create 300 \
  --name truenas \
  --memory 16384 \
  --balloon 0 \
  --cores 4 \
  --cpu host \
  --machine q35 \
  --bios ovmf \
  --efidisk0 local-zfs:1,efitype=4m,pre-enrolled-keys=0 \
  --scsihw virtio-scsi-single \
  --scsi0 local-zfs:32,discard=on,ssd=1 \
  --net0 virtio,bridge=vmbr0 \
  --cdrom local:iso/TrueNAS-SCALE-24.10.2.iso \
  --boot order=scsi0 \
  --onboot 1 \
  --startup order=1,up=120

Pass through the WD RED drives

# Pass through all 3 WD RED 8TB drives using by-id paths
# Replace <SERIAL1>, <SERIAL2>, <SERIAL3> with your actual drive serial values from Phase 1 Step 2
 
 
qm set 300 -scsi1 /dev/disk/by-id/ata-WDC_WD80EFZX-68UW8N0_VK0JEMLY
qm set 300 -scsi2 /dev/disk/by-id/ata-WDC_WD80EFZX-68UW8N0_VK1GLR1Y
qm set 300 -scsi3 /dev/disk/by-id/ata-WDC_WD80EFZX-68UW8N0_VLGTYHKY
 
# Verify VM config shows all drives
qm config 300 | grep scsi
# Expected:
# scsi0: local-zfs:vm-300-disk-0,discard=on,size=32G,ssd=1
# scsi1: /dev/disk/by-id/ata-WDC_WD80EFAX_<SERIAL1>,size=7452G
# scsi2: /dev/disk/by-id/ata-WDC_WD80EFAX_<SERIAL2>,size=7452G
# scsi3: /dev/disk/by-id/ata-WDC_WD80EFAX_<SERIAL3>,size=7452G

Phase 3: Install TrueNAS SCALE

Boot the installer and remove the ISO after setup

# Start the VM
qm start 300

Install TrueNAS onto the 32 GB virtio disk, not any of the 8 TB WD RED drives. After the install finishes, remove the attached ISO from the Proxmox side:

# From Proxmox host, remove the CD-ROM
qm set 300 --cdrom none

Set a static IP in the TrueNAS console menu, then log in through the web UI. The original notes referenced the older admin path, but fresh SCALE 24.10 installs now use truenas_admin as the default administrative account instead.¹

IP address: 192.168.50.50
subnet mask: 255.255.255.0
default gateway: 192.168.50.1
DNS: 192.168.50.10

Set hostname, domain, timezone, and SMTP details from the UI once the web console is reachable.

# SSH into TrueNAS or use the Shell widget in web UI
# Identify drive IDs
ls -la /dev/disk/by-id/ | grep -v part | grep ata
 
# Create pool: mirror of 2 drives + 1 hot spare
zpool create -f tank \
  mirror \
    /dev/disk/by-id/ata-WDC_WD80EFAX_<SERIAL1> \
    /dev/disk/by-id/ata-WDC_WD80EFAX_<SERIAL2> \
  spare \
    /dev/disk/by-id/ata-WDC_WD80EFAX_<SERIAL3>

# Enable zstd compression on the pool (matches your existing convention)
zfs set compression=zstd tank

Create datasets with tuned record sizes

# Create all datasets
zfs create -o recordsize=1M tank/media
zfs create -o recordsize=128K -o quota=500G tank/documents
zfs create -o recordsize=1M -o quota=2T tank/models
zfs create -o recordsize=1M tank/backups
zfs create -o recordsize=1M -o quota=2T tank/backups/vzdump
zfs create -o recordsize=1M -o quota=1T tank/backups/timemachine
zfs create -o recordsize=128K -o quota=500G tank/docker
zfs create -o recordsize=128K -o quota=500G tank/scratch
 
# Verify all datasets
zfs list -r tank

Use 1M record sizes for large sequential files like media, models, and backups. Keep 128K for mixed or smaller-file datasets like documents and Docker volumes.

TrueNAS documents that the legacy root login path was removed earlier in SCALE and that fresh 24.10 installs use truenas_admin instead of admin: https://www.truenas.com/docs/scale/24.10/scaletutorials/credentials/adminroles/ ↩

TrueNAS VM Build And Pool Design

TrueNAS VM Build And Pool Design

Why Mirror Plus Hot Spare

Physical Drive Map

Phase 0: Resource Reallocation

Audit current allocation

Reduce GPU container RAM allocations

Limit host ZFS ARC

Phase 1: Physical Install And Pre-Migration

Install the third WD RED and verify detection

Record stable drive identities

Back up the existing `naspool`

Save current storage and cron state

Phase 2: Retire `naspool` And Create The VM

Remove `naspool` from Proxmox storage config if needed

Export or destroy the host-side pool

Download the ISO and create the VM

Pass through the WD RED drives

Phase 3: Install TrueNAS SCALE

Boot the installer and remove the ISO after setup

Initial host settings

Phase 4: Create `tank` As Mirror Plus Hot Spare

Verify the disks inside TrueNAS

Create the pool and enable compression

Create datasets with tuned record sizes

Comments