Building a software RAID1 mirror with mdadm and replacing a failed disk

Goal

Create a two-disk RAID1 (mirror) array with mdadm, make it survive a reboot, and replace a disk after a failure.

Prerequisites

Root; mdadm installed. Two block devices or partitions of at least the intended array size, already identified by serial/WWN as the correct targets (do not skip that check — mdadm writes to whatever device names are given).

Steps

1. Create the array: mdadm --create /dev/md0 --level=1 --raid-devices=2 /dev/sdb1 /dev/sdc1. --create/-C builds a new array; --level=1 selects mirroring.
2. Watch the initial sync: cat /proc/mdstat shows the array, its state and, during the initial resync, a progress percentage — RAID1 is usable immediately but not fully redundant until this completes.
3. Inspect full detail: mdadm --detail /dev/md0 shows array state (clean, degraded, ...), the RAID level, and each member device with its role and state (active sync, spare, faulty).
4. Format and mount /dev/md0 as an ordinary block device (see disk partitioning/formatting steps separately); add it to /etc/fstab by UUID.
5. Persist the array definition: mdadm --detail --scan >> /etc/mdadm/mdadm.conf (Debian/Ubuntu path; RHEL uses /etc/mdadm.conf; check for an existing ARRAY line first so it is not duplicated). Without it the array usually still auto-assembles at boot, but often under a different name such as /dev/md127 — one more reason to mount by UUID.
6. Rebuild the initramfs so the array is recognized at early boot if the root or a boot-critical filesystem depends on it: update-initramfs -u (Debian/Ubuntu) or dracut -f (RHEL/Fedora).
7. To replace a failed disk: identify it in mdadm --detail /dev/md0 (state faulty or missing entirely), remove it explicitly if still present (mdadm /dev/md0 --fail /dev/sdc1 --remove /dev/sdc1), physically replace the disk, partition the replacement to match (for GPT: sgdisk -R /dev/sdc /dev/sdb copies the table from the healthy disk, then sgdisk -G /dev/sdc gives the copy new unique GUIDs; if the array holds /boot, also reinstall the bootloader on it), then add it back: mdadm /dev/md0 --add /dev/sdc1. Resync progress is visible again in /proc/mdstat.

Expected result

cat /proc/mdstat shows [UU] (both members up) for a healthy RAID1; after a disk replacement it briefly shows [U_] or [_U] during resync and returns to [UU].

Limits and test basis

Verified against mdadm(8) (--create, --detail, mdadm.conf) and md(4) (/proc/mdstat). Undo: mdadm --stop /dev/md0 then mdadm --zero-superblock /dev/sdX1 on each former member removes the RAID metadata; also remove the ARRAY line from mdadm.conf and the fstab entry. This destroys the array's ability to reassemble and should only be done after data is migrated or no longer needed.