Software RAID, on both levels
RAID is not about speed. It is about the hour after a disk dies — the hour in which a host without it stands still and everything on it is as old as the last backup.

Nothing is overwritten unasked
A disk that carries a filesystem, a partition table, a mount or another array is refused, with the reason given. Only what is expressly forced goes through, and forcing is a field of its own rather than a flag hidden in a dialogue.
An array is worth nothing until something lies on it, so creating one can put a filesystem on it and register it as a storage in the same step.
It has to come back after a restart
An array that exists only in the running kernel is one lost boot away from being gone: the members are individual disks again, and what stood on them is not to be found. So every change writes the array configuration anew and rebuilds the initramfs.
On a system array, the bootloader belongs on every disk. One that sits on only one of them turns the surviving disk into a machine that cannot start — which is the failure this whole feature exists to prevent.
Two sources, because neither is enough
What the array is made of and how it is doing comes from one place; how far a rebuild has come comes from another. The progress is the number somebody watches after replacing a disk, and it lives only in the second source — so both are read.
A missing member is reported as a missing member, not as a percentage that happens to be below a hundred.
An array inside a virtual machine
A guest can have one too, and the interesting way to build it is with the VM switched off. Its disks are files, and this host has everything needed to work on them directly.
That buys two things the in-guest way cannot offer: no agent, no network and no account are needed — a machine somebody built by hand or an appliance image works — and the array is complete straight away, because the copy is made first and the old disk is taken in immediately. Built inside a running guest, the array starts degraded and the old disk can only join after a restart.
The price, said plainly: the last step runs the guest's own tools in a chroot, as root, on the host. There is no way around it — only the guest's initramfs tools know how to build an initramfs it can boot. So this is meant for images the operator brought themselves, which in a hypervisor is the normal case.
Every topic on its own page
Fourteen pages, each written from what the program actually does — including where it stops.













Try it on your own hardware
A month, complete, with no registration and no data asked for. The installer writes a whole server — a spare machine or a VM is the place for it.