Pool will not import after power loss: zpool import -f, then -F if you must rewind txgs, then zpool scrub.

After a power cut the pool is often still on the disks but zpool status is empty: it was left dirty-exported, or the host booted without it. Do not zpool create on those disks.

See what is there

1

zpool import

Lists importable pools by name (VOL1 below) and whether they are FAULTED / UNAVAIL.

Import

1

zpool import -f VOL1

-f means “I know it might still look imported elsewhere”. That is the usual crash case: ZFS thinks the other side still holds it.

If -f is not enough because the last txgs were half-written:

1

zpool import -F VOL1

-F rewinds to an earlier consistent txg. You can lose the last seconds of writes. That is better than a pool that will not mount.

Dry-run:

1

zpool import -F -n VOL1

Afterwards: scrub

1
2

zpool scrub VOL1
zpool status VOL1

Scrub confirms the rewind did not leave checksum errors. Let it finish before you delete “just in case” snapshots.

Do not

• zpool create on the same sdX devices wipes the labels.
• -F is not the first knob: try -f first.
• If the pool is the root (rpool / Proxmox bpool), import from live media, not from the system that will not boot.

See also: replacing disks, basic commands.

OFFLINE disk: zpool replace -f <GUID> /dev/disk/by-id/… — GUID from zpool status, not /dev/sdX.

Proxmox documents the case: ZFS on Linux — change a failed device. The symptom is an OFFLINE / FAULTED vdev and a long number instead of sde.

Status

1

zpool status vol1

The dead member often shows up as a GUID:

1

1894156996840098641  OFFLINE

That happens when the kernel no longer has a /dev/sdX for that disk (you yanked it, it died, or the letter moved).

Replace

1

zpool replace -f vol1 1894156996840098641 /dev/disk/by-id/ata-NEWDISK

-f forces if ZFS still “remembers” the old disk. The replacement must be at least the same size (prefer by-id, not sde).

If the new disk is in the same slot and ZFS sees it:

1

zpool replace vol1 /dev/disk/by-id/ata-OLDDISK /dev/disk/by-id/ata-NEWDISK

or auto-detect a replacement in-place:

1

zpool replace vol1 /dev/disk/by-id/ata-OLDDISK

Wait for resilver

1
2

zpool status vol1
watch -n 5 zpool status vol1

Do not reboot, export, or start another replace on the same vdev until resilvered. On RAIDZ1 you are one disk down: a second failure in that window is data loss.

Afterwards

1
2

zpool detach vol1 1894156996840098641   # only if status still shows it as spare/old
zpool labelclear /dev/sdOLD             # if you reuse the old disk elsewhere

See also: import after a power outage.

Cheat sheet: zpool create, export/import, lz4, L2ARC, SLOG, labelclear, destroy — the 2 a.m. set.

Short reference. Longer procedures are linked at the bottom.

Pool

1
2
3
4
5
6

zpool create vol1 raidz sda sdb sdc
zpool status vol1
zpool export vol1
zpool import -f vol1
zpool destroy tank          # irreversible
zpool labelclear ada0       # wipe ZFS labels from a spare disk

raidz here is RAIDZ1 (one parity). For two, raidz2. Address disks as /dev/disk/by-id/…, not sdX, which reorder.

Mountpoint

1
2
3

zfs unmount vol1
zfs set mountpoint=/mnt/vol1 vol1
zfs mount vol1

L2ARC (cache) and SLOG (ZIL)

1
2
3
4

zpool add vol1 cache <disk-id>
zpool add vol1 log <disk-id>
zpool add vol1 log mirror <disk-id-a> <disk-id-b>
zpool remove vol1 <device>

Cache = hot reads. Log = sync writes (NFS, VMs). A “fast” USB SLOG is worse than no SLOG: if it dies mid-txg, you feel it. Mirror the log.

Compression

1
2
3

zfs set compression=lz4 newvol
zfs set compression=off newvol
zfs get compressratio newvol

lz4 is the sane default on modern OpenZFS. zstd compresses more and costs CPU. The property inherits to child datasets that do not override it.

See also

• Import after a power outage
• Replace disks
• Native encryption