Homelab setup

• ZFS : File system
• Wireguard : VPN tunnel protocol
• Headscale : Tailscale control server (replaces wg-easy + dnsmasq)
• Caddy : reverse proxy
• Samba : shared folder
• Immich : photo management
• Joplin : Markdown note app
• Uptime Kuma : monitoring

Structure

Main compose file includes domain-specific stacks:

• compose/infra/compose.yml : headscale + caddy
• compose/immich/compose.yml : immich services
• compose/apps/compose.yml : joplin + uptime-kuma + flash-backend
• compose/samba/compose.yml : samba

NOTES

• Raspberry Pi that listens for ZFS with mbuffer and corresponding systemd service:

#!/bin/bash

TARGET_DATASET="magenta-backup/data"
MBUFFER_PORT=9090
MBUFFER_MEM="1G"
MBUFFER_BLOCK="128k"

sudo mbuffer -q -Q -m "$MBUFFER_MEM" -s "$MBUFFER_BLOCK" -I "$MBUFFER_PORT" | \
sudo zfs receive -uF "$TARGET_DATASET"

[Unit]
Description=ZFS Receive via mbuffer
After=network-online.target
Wants=network-online.target

[Service]
Type=simple
ExecStart=/home/pierre/zfs-receive-mbuffer.sh
Restart=on-failure

[Install]
WantedBy=multi-user.target

• ZFS mbuffer receive should also be in a cron job since when receiving is done, the process is terminated
• Backup destination datasets on Raspberry Pi should stay read-only for normal usage, otherwise next incremental receive can fail with "destination has been modified"
• Set a timezone if you want your cron job to run when expected
• zfs-backup.sh in a cron job at 3am (since Immich database backup is at 2am)
• To get backup alerts in Uptime Kuma:
  • Create a Push monitor in Uptime Kuma
  • Copy the generated push URL (https://<kuma>/api/push/<token>)
  • Set UPTIME_KUMA_PUSH_URL in script/zfs-backup.sh
  • Script sends status=up on success and status=down on failure automatically
  • Since the backup runs once per day at 3am, set the expected push interval to about 24h and give it a grace window like 26h or 30h