With ZFS on Linux, it often happens that zpool is created using disk identifiers such as
/dev/sda. While this is fine for most scenarios, the recommended practice is to use the more guaranteed disk identifiers such as the ones found in
/dev/disk/by-id. This blog post describes 3 methods how to change the disk identifiers in such zpool after it has been created. All this without migrating data, adding disks or having physical access to the machine.
- Export and import – offline, any zpool
- Offline and replace – online, any zpool, buggy on Linux
- Detach and attach – online, mirror zpool only
Export and import
This is a well-known procedure which works if you can afford downtime. The zpool has to be unmounted for this to work.
Start by exporting the zpool (rpool0 in this example):
# zpool export rpool0
Import it right after with specifying the correct path where to look for the devices:
# zpool import -d /dev/disk/by-id/ rpool0
This might import the disks by their world wide name which is not a problem since these are guaranteed to be persistent for any given disk on Linux, no matter where you put it (different drive bay or a completely different machine).
Offline and replace
This is the official method for replacing disks as described in Oracle docs. The command
zpool labelclear refuses to do anything because it sees the disk as part of an active zpool. For some reason, none of the usual Linux tools like wipefs, parted, gdisk, fdisk could manage to properly clear ZFS metadata from the disk, so the only option is zeroing out the disk manually which takes a long time and unnecessarily wears out SSDs.
Detach and attach
This works for online renaming, for example, when your operating system’s root is on the zpool. Sadly, this does not work for RaidZ pools.
Here is the zpool mirror used in the example:
pool: rpool0 state: ONLINE scan: scrub repaired 0 in 0h1m with 0 errors on Sun Apr 14 00:25:39 2019 config: NAME STATE READ WRITE CKSUM rpool0 ONLINE 0 0 0 mirror-0 ONLINE 0 0 0 sda ONLINE 0 0 0 sdb ONLINE 0 0 0 errors: No known data errors
Start by removing the first disk in the mirror:
# zpool detach rpool0 sda
Wipe zpool information from the disk:
# zpool labelclear -f /dev/sda
If it fails, try:
# wipefs -a /dev/sda
Determine its name in /dev/disk/by-id:
# ls -al /dev/disk/by-id | grep sda
Attach it to the zpool mirror with the correct name:
# zpool attach rpool0 sdb /dev/disk/by-id/ata-disk-model-serial
zpool status and when the resilvering process completes, do the same for the other disk in the pool:
# zpool detach rpool0 sdb # zpool labelclear -f /dev/sda # wipefs -a /dev/sda # ls -al /dev/disk/by-id | grep sdb # zpool attach rpool0 sda /dev/disk/by-id/ata-disk-model-serial
Wait for the resilvering process to finish. Now you have a properly set up zpool mirror.