BBootMASS
bootmassconsulting · golden images

Boot metal as a service

Your workload, built once as a golden image on ZFS and delivered to bare metal running on first boot: from a USB stick, over the network, or as a file you can upload anywhere. An update is a new image promoted beside the old one. The way back is one reboot.

BootMASS is a consulting company. Its tool turns bare metal into a ZFS-on-root machine from a signed image, keeps it able to roll back, and installs over the network: a rack armed with one command, no packet from the internet. It is not sold on its own; you get it by working with BootMASS, and it ships inside every Zettarack appliance.

From a USB stickA UEFI installer with Secure Boot and no network. Single disk, mirror, raidz1 or raidz2, encryption optional.
Over the networkArm a machine by its network card and it installs on its next netboot. Machines you didn't arm boot their own disk, untouched.
Upgrades that look after themselvesEach release is a signed image in a new boot environment. It boots once and goes back by itself if its health check does not pass.
the approachgolden images

The image is the artifact

The usual way builds a machine, configures it into a workload, and keeps configuring it for life. BootMASS starts from the other end: build your artifacts, include them in the image, configure them to run, and cast the whole thing as one golden image. Nothing is applied, managed or upgraded on the machine afterwards.

This is immutable infrastructure, done on ZFS: no pet, no configuration tool, no package manager on the machine. It works with every build system and needs none. Your code stays in git, your CI builds and tests it, and the golden image is the step after that: the whole running system as one artifact.

The argument in slides, today's way and this one side by side: What if ZFS wasn't a plug-in?

1 · BuildA Linux root filesystem from the distribution's own packages, with a kernel, ZFS and your artifacts in it. apt on Debian today; dnf and pacman later.
2 · Castbootmass cast makes it a golden image, one ZFS stream with its digest; bootmass release signs it. A snapshot: it clones and replicates at will.
3 · Point and shootBoot the stick, or arm machines by their cards, one or a rack. Each comes up as your workload: the exact image that passed your tests.
deploymentsknown good

From deploying a mess to promoting a known good

A deploy stops being a procedure and becomes a promotion. The image that passed is received into a new boot environment beside the running one and switched to whole; blue/green is choosing which is active, and going back is choosing the one before. No apt or dnf upgrade, no drift, no update that fails halfway: rollback, reboot, done.

What it replacesConfiguration management and in-place upgrades: the Ansible run against a live machine, the upgrade that fails halfway, the two machines that are no longer the same.
What it does notYour build and your tests. Any CI that produces artifacts feeds this; the image is one more step at the end of the pipeline, not a new one.
What it changesDelivery. The same image goes to a rack, a single box, a stick or another site and comes up running, so staging and production differ only in where they are.
the claimunbreakable

Unbreakable, and what that means

The claim is about the system, not the data: you cannot brick a machine built this way with an upgrade, and the way back from anything wrong in the system itself is one reboot. Your data has snapshots and backups, a different thing.

Three ZFS layers: the signed image, read-only; a boot environment cloned from it per release; the machine's own state, kept across them. A release is received whole or not at all; the running environment is snapshotted first; the new boot menu is tried once. Five minutes into the first boot, bootmass doctor checks the machine, and if it does not pass, the environment before is booted into and that release is never installed automatically again. A check that could not run counts as failed.

A release that does not bootA kernel that panics, a menu that cannot load, a module the kernel refuses: the boot menu goes back to the environment before, by itself.
A release that boots sickA daemon that will not start, a pool that will not import, a firewall that locked the door: the doctor finds it and the machine goes back on its own.
A power cut mid-upgradeThe new environment is switched to only when whole. Power lost at any point leaves the machine on the release it was running.
A regretA release that passed and is still not wanted: bootmass be activate names the one before. The last five stay on the disk.
DriftNothing is installed, patched or edited on the machine. Two machines on one release are the same bytes; a machine marked as modified says so.
Not a dead boxA disk or machine that dies is a backup's job: the appliances replicate to a second box and prove the restore, and bootmass puts the system back on other hardware in minutes.
against the restmaas · infrastructure as code

Not a provisioning stack

Metal-as-a-service and infrastructure-as-code tools deploy a target, usually from a vendor's image or installer, and then rely on management to turn it into a workload and keep it that way. Each does its job well; together they are a stack to run before the first workload does.

MAAS · Foreman · CobblerDiscover, commission and install bare metal from a provisioning server, then hand the machine to cloud-init, Puppet or Ansible to configure.
Tinkerbell · Metal3Bare-metal provisioning driven from Kubernetes: a cluster has to be running before the first machine can be installed.
Terraform · Ansible · PackerDeclare and provision, configure in place over SSH, build images and stop there: another tool puts them on a machine.

BootMASS is one binary with no service and no vendor image. It assembles the distribution from its own packages, seals it with your workload as a golden image, and delivers it running on first boot; an update is a new image beside the old, undone by choosing the one before.

zfsdone properly

ZFS from the first block

The machines are yours to look after, and ZFS makes that small: pooled disks, a checksum on every block, snapshots, clones, encryption and replication in one layer. Elsewhere it is a storage plugin on a system that boots from something else. Here the root is ZFS, the machine's settings are a dataset in the pool, encrypted with it when you choose, and boot environments are chosen by a menu that reads the pool directly, with no GRUB.

consultinghire@bootmass.com

What BootMASS does for you

Work à la carte, by the engagement. Each of these is something I have built and run.

Golden image creationYour workload built into a golden image: the step after git. A Kubernetes node, a database, a LAMP stack, a desktop, whatever you ship.
Blue/green deliveryDeployments as promotions between boot environments, with the health check that keeps or returns a release, fitted to your window.
ZFS supportPools designed for the workload, on root and for data; tuning, encryption, replication, recovery.
TrainingYour team builds, casts, serves, upgrades and rolls back, on your hardware, until it is theirs.
ToolingThe bootmass tool and the image pipeline, under contract and fitted to your estate. Not sold on its own.
Bare-metal deliveryInstallers on a stick, netboot with machines armed by their cards, racks installed from the server beside them.
email
$ dig +short txt hire.bootmass.com | tr -d '"' | base64 -d | rev
or, from a terminal
sister companyzettarack.com

The appliance company, Zettarack

Zettarack, the sister company, makes whole machines for your own hardware, each with one job, built on ZFS from the disk up and installed with the bootmass tool. The method above is proven on them today: vmx, the hypervisor, and zxp, the storage server, both in testing.

website designed by North Star