[ node: kulczyk-lab ] [ status: ONLINE ]
HOMELAB
Where I break things on purpose. Self-hosted services, container builds, and the notes I wish I'd had the first time.
aaron@lab:~$ whoami aaron — sysadmin by trade, tinkerer by habit aaron@lab:~$ lab --status [ OK ] home-dash dashboard + planner [ OK ] nfc-bin-storage bin inventory [ .. ] pantry-dash building [ .. ] home-security planning (ESP32-C6 / nRF52) aaron@lab:~$ cat motto.txt if you have to do it twice, script it. aaron@lab:~$ █
# rack.view
The services I self-host. Each runs in its own Debian container as a Node.js app.
- home-dash Wall-tablet link hub, weather, to-dos, calendar, shopping lists RUNNING
- nfc-bin-storage Tap an NFC tag on a bin, see what's inside. Express + SQLite. RUNNING
- pantry-dash NFC-tagged pantry, fridge, and freezer inventory, wired into home-dash BUILDING
- home-security Local-only DIY alarm: contact, motion, and glass-break sensors, plus smart locks PLANNED
# stack.conf
// at home
- Debian 13 containers
- Node.js (via nvm) + Express
- SQLite / flat-file JSON state
- NFC tags + phones as scanners
- ESP32-C6 / nRF52 (coming soon)
// on the job
- RHEL clusters on Ceph storage
- VMware vSphere (vMotion, HA)
- Windows Server · AD · GPO · WSUS
- WDS · PXE · UEFI HTTP boot
- Tenable · Trellix · Forcepoint · syslog
// glue
- PowerShell (+ WinForms GUIs)
- Bash
- JavaScript / Node.js
- SQL
- Lots of documentation
# lab.notes
Short write-ups from things I've built. Open one to read it.
0x01 UEFI HTTP boot vs. classic PXE
Classic PXE pulls a small boot loader over TFTP, which is slow and fussy across subnets. UEFI HTTP boot has the firmware fetch the boot image straight over HTTP(S): faster, easier to route, and served by any web server.
- DHCP still drives it. HTTP boot clients identify as
HTTPClientin option 60, and you hand back a full URL as the boot file. - Keep the clients on their own VLAN and scope so imaging traffic stays isolated.
- On an air-gapped network, stage the Windows ADK and WinPE add-on installers offline ahead of time.
- Capture a golden image from a reference machine, then deploy it with in-box tools. No third-party software required.
0x02 Ceph in one screen
Ceph turns a pile of disks across several servers into one fault-tolerant storage pool. The parts:
MONmonitors hold the cluster map and need an odd number for quorum (3 or 5).OSDdaemons, roughly one per disk, store the data and handle replication and recovery.MGRprovides metrics and the dashboard.- Pools use replication (usually size 3, min_size 2) or erasure coding. Placement groups spread objects across OSDs.
First habit to build: ceph -s and ceph health detail before and after every change.
0x03 Kiosk hardening checklist
- Dedicated local or domain kiosk account with no admin rights.
- Lock the shell down with Assigned Access, Shell Launcher, or a custom app as the shell.
- Apply GPOs: no Task Manager, Run box, Control Panel, or right-click where it isn't needed.
- Use AppLocker or WDAC to allowlist only the apps the kiosk needs.
- Gate admin functions behind real AD authentication and group membership, never a hard-coded PIN.
- Build the UI DPI-aware. Hard-coded pixel offsets break at 125% and 150%.
- Apply STIG baselines, BitLocker, and patching, same as any other endpoint.
0x04 Running Node apps in a container without Docker
- Install Node with
nvmso each container gets the exact version it needs. - Keep code and data apart: app in
/opt/<app>, state in/var/lib/<app>. Upgrades become unzip-and-restart, and your data survives. - Service managers don't load your shell profile. Point them at the absolute path of the nvm node binary.
- Give every app its own port and write the ports down before two apps collide.
- Ship a one-command installer. Future-you will thank you.