BBootMASS
bootmassrack builder

BootMASS custom rack builder

What if a 42U rack were an order form? Pick every server, write one file, power the rack on, and every workload is running at first boot. Not install an OS, dress it up with tools, reboot it and start the services. That is the golden image workflow, and BootMASS is built for exactly that: a fully running system, built once, replicated in place onto every server.

If that sounds like NixOS, it is similar, and not the same. There is no language to learn and no build to run on the box: one short configuration you can read in a minute, the rack file this page writes. The image is whatever you build into it, an appliance, a database, a desktop, a Kubernetes node, your own application, and it arrives fully running on the MAC and the address you gave it.

ZFS end to end, and only ZFS: the root, its boot environments and every data pool. That is what makes the rest possible: an image is a stream, a server is a clone, an update is a new clone beside the old one. There is no other filesystem to choose.

Built onceThe system and its workload, built and tested once as a golden image: one ZFS stream, signed. A Kubernetes node, a database, a desktop, whatever you ship.
Replicated in placeReceived whole onto each server's own pool and run as a clone of the read-only image. Two servers on one release are the same bytes; an update is a new clone beside it.
Running at first bootNo install phase, no configuration run, no services to bring up by hand: first boot is the workload, on the network the file gave it.
the builderruns in this page

Build a rack

Pick servers from the shapes, or design your own down to the pool, fill the rack, and take the list BootMASS deploys the rack from: every server's image, root pool, special and log devices, data pools, address and zone, in one file. BootMASS checks the whole rack first and arms all of it or none of it; each server is deployed at its next network boot, from the server beside it.

The file is the rack, so a rack is as reusable as a file. Arm the same list again and every server re-images itself at its next boot; arm a different one and the same metal is a different rack: an office at eight, a render farm at noon, a lab for the evening, every disk named erased in between and a pool named by import kept through it all. The arming is one command, so a line in cron is a schedule for a whole rack.

Nothing you type is sent anywhere or kept. A server is armed by its network card, so it is written as a line once it has its card's MAC and the by-id name of each of its disks; until then it is written commented out, as planned, with what it still needs.

settingsimages and network
Image versions
The network, decided here (empty: DHCP)
The rack
the rack42U
Design your own server: the root pool, its special and log devices, data pools beside it

root pool

special vdev: metadata and small blocks on flash (a mirror: its loss is the pool's)

log device: sync writes, on something faster than the pool and safe on power loss

data pools beside the root pool (a reinstall keeps them with one import line)

the list
rack-42u.list

    
Have BootMASS deploy it
website designed by North Star