waterlemon0108/nexboot

Keeping 50 or 500 PCs on the same Windows build is a chore. Every game patch, driver update or bad install means walking from seat to seat.

With NexBoot the PCs have no system disk of their own. They boot over the network, and each one gets its own writable copy of a master image. You update the image once. Every PC picks it up on the next reboot, and a reboot always brings a PC back to a clean state.

On an Ubuntu 22.04 / 24.04 server with a spare disk for the ZFS pool:

sudo apt install ./ndiskless_<version>_amd64.deb
sudo ndiskless-configure --bootstrap-password '<admin password>'

Then open http://<server-ip>:8080, sign in as admin and follow the setup wizard:

1. Pick the network card that faces the clients.
2. Create a storage pool on the spare disk.
3. Import a Windows or Ubuntu image (.vmdk, .vhdx, .qcow2, .raw or an exported .zfs stream).
4. Create a group (IP range + image) and register the clients by MAC address.
5. Set the clients to boot from the network. Done.

No internet at the site? Use the offline bundle: sudo ./install.sh from nexboot-offline-ubuntu24.04-amd64.tar.gz. See Install for all options.

 client PC                 NexBoot server (one Go binary)
 ─────────                 ──────────────────────────────────────────────
 PXE ROM ──DHCP/TFTP────▶  dnsmasq (managed)  ──▶ iPXE
 iPXE ────HTTP /boot────▶  boot script for this MAC (which image, which disk)
 iPXE ────iSCSI─────────▶  LIO target ──▶ ZFS clone of the restore point
 Windows / Linux boots     ▲
 from the network disk     └─ image ─▶ config ─▶ restore point ─▶ per-PC clone

• Copy-on-write clones. Each PC boots from a ZFS clone of a restore point. Clones cost almost no space and take a second to create.
• Images, configs and restore points. One image can have several configs (for example "esports" and "office"), and each config keeps its own history of restore points.
• Super machine. Boot one PC in edit mode, install your updates, shut it down and save the result as a new restore point.
• Data plane stays local. DHCP, /boot and iSCSI go to the node that hosts the PC. Nothing is proxied through a central controller.

Overview: nodes, clients, groups and storage at a glance	Images: system disks and data disks
Restore points: apply, save as image, export	Groups: IP range, system disk and data disks
Client matrix	Storage pools per node
Create a cluster with just a VIP	Add nodes found on the network
Windows 11 client booted over the network	Ubuntu 24.04 client booted over the network

More screenshots are in the Chinese README.

Each server

Network

• No other DHCP server may answer the clients. NexBoot only answers registered MACs, and a second server would hand out wrong addresses.
• Set the clients to boot from the network card. BIOS and UEFI both work.

Don't set a static IP inside the image. Every PC boots the same image, so a fixed address would be claimed by all of them at once. NexBoot gives each MAC its own address over DHCP. Ubuntu images are switched to DHCP automatically on import.

Every server runs the same install and starts as a standalone node. Whether it joins a cluster is decided later in the web UI.

# .deb package (recommended)
sudo apt install ./ndiskless_<version>_amd64.deb
sudo ndiskless-configure --bootstrap-password '<admin password>'

# offline bundle, for sites without internet
mkdir nexboot && tar -xzf nexboot-offline-ubuntu24.04-amd64.tar.gz -C nexboot && cd nexboot
sudo ./install.sh
sudo ndiskless-configure --bootstrap-password '<admin password>'

# from source (see Development for build prerequisites)
make build-static
sudo bash scripts/install-go.sh --install-deps --binary ./bin/ndiskless-linux-amd64 \
  --bootstrap-password '<admin password>'

Check it is up:

systemctl is-active ndiskless            # active
curl -s http://127.0.0.1:8080/healthz    # {"status":"ok"}

Upgrading a .deb does not restart the service. Run sudo systemctl restart ndiskless; in a cluster, upgrade the standbys first and the active node last.

Failover is automatic and takes seconds to minutes, not zero. Clients running on a failed node blue-screen and reboot once, then come back on another node. The VIP moves in about a second.

To build a cluster:

1. On the first node: System → Parameters → High availability → Create cluster, and enter a free VIP.
2. Servers → Cluster nodes → Add node. NexBoot scans the client network for installed but unconfigured servers.
3. Create a storage pool on every node.
4. Wait for the first replication to finish (shown under High availability).

A single node that already has images and clients can be turned into a cluster in place. Nothing is reinstalled or migrated.

cmd/ndiskless/   entry point, wiring, HA role selection
internal/
  api/           HTTP routes, auth, /boot
  boot/          iPXE scripts and embedded TFTP assets
  control/       images, configs, restore points, groups, clients, drivers, alarms, backup, cluster ops
  dhcp/          dnsmasq config generation
  ha/            role state machine, epoch fencing, replication endpoints
  storage/       ZFS, iSCSI (LIO), image import and adaptation
  store/         SQLite repositories and goose migrations
web/             React web console, embedded into the binary
deploy/          systemd unit, keepalived template, health scripts
packaging/       .deb and offline bundle builders
tools/e2e/       end-to-end test matrices run against real servers
docs/images/     screenshots, demo GIF and project artwork

Use the Go version specified in go.mod and Node.js 22 (at least 22.22.2) or 24 (at least 24.15.0), with npm. Build and verification targets install frontend dependencies with npm ci, build web/dist and embed it before compiling Go. Generated files are not committed.

make dev          # dev binary at bin/ndiskless-dev
make verify       # gofmt, vet, go test, static build
make verify-web   # vitest, vite build
make verify-all   # all checks; shares the frontend build
make deb          # .deb package (on Linux)
make offline      # offline bundle (on the target Ubuntu version)

End-to-end matrices need real ZFS, LIO and keepalived, so they are not part of CI. See tools/e2e/README.md.

Issues and pull requests are welcome. Read CONTRIBUTING.md first.

AGPL-3.0. You can use, study and modify NexBoot for free, including running it in your own internet cafe or school. If you distribute a modified version, or offer a modified version to others over a network, you must publish its full source under the AGPL.

The NexBoot name and logo are not covered by the license. Bundled third-party components (iPXE, wimboot) keep their own licenses; see THIRD_PARTY_NOTICES.md.