ZPool Spiekbriefje

OpenZFS combines the volume manager and the filesystem. A pool (zpool) is built from vdevs, and vdevs are built from disks; datasets (zfs) are the filesystems and volumes that live inside a pool.

⚠️ Redundancy lives at the vdev level, not the pool level. Lose one vdev and you lose the whole pool, however healthy the others are.

📦 Pool Basics

# create a pool from a single disk (no redundancy)
zpool create mypool /dev/disk/by-id/<disk>

# a two-disk mirror, 4K-aligned
zpool create -o ashift=12 mypool mirror /dev/disk/by-id/<disk1> /dev/disk/by-id/<disk2>

# raidz2 across six disks
zpool create -o ashift=12 tank raidz2 <disk1> <disk2> <disk3> <disk4> <disk5> <disk6>

# mount the pool somewhere other than /<poolname>
zpool create -m /srv/data mypool <disk>

# add another vdev (this widens the pool, it does not add redundancy)
zpool add mypool mirror <disk3> <disk4>

# add a hot spare, a read cache (L2ARC) and a separate intent log
zpool add mypool spare <disk>
zpool add mypool cache <ssd>
zpool add mypool log mirror <ssd1> <ssd2>

# destroy a pool — irreversible
zpool destroy mypool

🔍 Inspecting

# pools, their size and health
zpool list

# per-vdev breakdown
zpool list -v

# layout, errors and any running scrub or resilver
zpool status mypool

# only the pools that need attention
zpool status -x

# live throughput, refreshed every 5 seconds
zpool iostat -v 5

# every property of the pool
zpool get all mypool

# what has been done to this pool, ever
zpool history mypool

🛠 Vdev Types

Support vdevs: cache (L2ARC read cache), log (SLOG, for synchronous writes), special (metadata and small blocks), spare (hot spare).

🩺 Health & Maintenance

# verify every block against its checksum
zpool scrub mypool

# stop or pause a running scrub
zpool scrub -s mypool
zpool scrub -p mypool

# forget errors that have been dealt with
zpool clear mypool

# swap a failing disk for a new one
zpool replace mypool <old-disk> <new-disk>

# turn a single disk into a mirror, or break one up
zpool attach mypool <existing-disk> <new-disk>
zpool detach mypool <disk>

# take a disk out of service and bring it back
zpool offline mypool <disk>
zpool online mypool <disk>

# tell the SSDs which blocks are free
zpool trim mypool

# enable the on-disk features of a newer OpenZFS release
zpool upgrade mypool

💡 Scrub monthly on consumer disks, weekly if the data matters. zpool status reports what the last scrub found.

📤 Import & Export

# release a pool so it can be moved to another machine
zpool export mypool

# list the pools that could be imported
zpool import

# import by name, preferring the stable device names
zpool import -d /dev/disk/by-id mypool

# import every pool it can find
zpool import -a

# import a pool that was not exported cleanly
zpool import -f mypool

# import under a different name, or under an alternate root
zpool import mypool newname
zpool import -R /mnt mypool

⚙️ Pool Properties

# trim automatically as blocks are freed
zpool set autotrim=on mypool

# grow the pool when the underlying disks grow
zpool set autoexpand=on mypool

# replace a failed disk with a spare automatically
zpool set autoreplace=on mypool

# what to do when the pool faults: wait | continue | panic
zpool set failmode=continue mypool

# read one property
zpool get health,capacity,fragmentation mypool

📁 Datasets & Filesystems

# create a dataset
zfs create mypool/data

# create a nested dataset, making the parents as needed
zfs create -p mypool/data/projects/web

# create a block device (zvol) for a VM or iSCSI target
zfs create -V 50G mypool/vm-disk

# list datasets, snapshots and volumes
zfs list
zfs list -t all -r mypool

# common properties
zfs set compression=lz4 mypool/data
zfs set quota=100G mypool/data
zfs set reservation=10G mypool/data
zfs set atime=off mypool/data
zfs set recordsize=1M mypool/media

# read properties, including where they were inherited from
zfs get -r compression mypool
zfs get all mypool/data

# mount, unmount, and destroy
zfs mount mypool/data
zfs unmount mypool/data
zfs destroy -r mypool/data

📸 Snapshots & Replication

# take a snapshot (instant, and free until the data diverges)
zfs snapshot mypool/data@2026-08-22

# snapshot a dataset and everything under it
zfs snapshot -r mypool/data@nightly

# list snapshots and what they cost
zfs list -t snapshot -o name,used,refer

# go back in time — discards everything written since
zfs rollback mypool/data@2026-08-22

# read a single file out of a snapshot instead
ls /mypool/data/.zfs/snapshot/2026-08-22/

# a writable copy of a snapshot
zfs clone mypool/data@2026-08-22 mypool/data-copy

# send a full snapshot to another machine
zfs send mypool/data@snap1 | ssh user@host zfs receive backup/data

# send only what changed since the previous snapshot
zfs send -i mypool/data@snap1 mypool/data@snap2 | ssh user@host zfs receive backup/data

# delete a snapshot
zfs destroy mypool/data@snap1

🧠 Tips & Best Practices

• Address disks by /dev/disk/by-id/..., never /dev/sdX — kernel names move between boots.
• Set ashift=12 at creation time for any modern drive; it cannot be changed afterwards.
• Give ZFS whole disks, not partitions, so it can manage the write cache itself.
• Keep pools below ~80% full — performance falls off a cliff above that, and fragmentation is permanent.
• Use ECC RAM where you can: ZFS trusts what is in memory.
• Do not mix vdev types or widths in one pool; the pool is only as good as its weakest vdev.
• A snapshot is not a backup — replicate it somewhere else with zfs send.

📚 Resources

• OpenZFS documentation
• zpool manual page
• zfs manual page
• Workload tuning guide