Here I describe a way I use ZFS on Ubuntu. Usually I just create mirror partition (RAID 1) for storing and sharing some data.

Using almost any of given command will probably erase some data. If you’re not sure what you’re doing, check any good documentation or guide.

BE EXTREMELY CAREFUL. It’s very simple to lose data. If you don’t understand something or you want to just create a partition, don’t be afraid of asking questions on any resource you prefer, like stackoverflow or reddit.

Erasing partitions from a drive

First of all you need to erase all partitions from every drive. In order to do so I use fdisk.

Via fdisk -l you can see all drives and partition tables assigned to them. To erase all partitions you can simply do the following:

  1. Find correct drive:
    • sudo fdisk -l
    • (for example let’s use /dev/sdf)
  2. Open the drive using fdisk:
    • ``sudo fdisk /dev/sdf`
    • (now you’re using fdisk application)
  3. In order to check whether you’re using the right drive you can print the drive info and all partitions that are on current drive by pressing p.
  4. Delete partitions:
    • Press d for each partition that drive has or until fdisk prints the error that there are no partitions left.
  5. Write all changes by pressing w

After erasing partitions you can create a ZFS pool.

Creating a simple pool

It’s not recommended to create a ZFS pool via /dev/sd* devices because Ubuntu may reassign drives after reboot. But using drives by their id will eliminate this issue.

Drives ids are stored in /dev/disk/by-id/ directory. As a way to find required drive, just use ls and grep:

  • ls -l /dev/disk/by-id/ | grep sd

Usually drives are named using following convention: type-id-partition.

  • Example: ` scsi-SATA_ST2000DM008-2FR1_ZFL16E1K-part1`
  • Type may be: ata, scsi, wwn, etc. Usually it doesn’t matter what you choose. I prefer using scsi.
  • ID may include following characters: _, -.
  • If you’ve erased partitions from a drive, you won’t see -part*.

You don’t want to create a ZFS pool out of disk partitions.

To create a ZFS pool you need to use zpool tool.

Example of creating a ZFS pool out of a single drive:

  • sudo zpool create -f zfs-pool-something /dev/disk/by-id/scsi-SATA_ST2000DM008-2FR1_ZFL16E1K

Where:

  • zfs-pool-something - the name of the pool.
  • /dev/disk/by-id/scsi-SATA_ST2000DM008-2FR1_ZFL16E1K - the drive by id.

Also, you can create a pool with different RAID level:

  • sudo zpool create -f zfs-pool-something mirror /dev/disk/by-id/scsi-SATA_ST2000DM008-2FR1_ZFL16E1K /dev/disk/by-id/wwn-0x5000c50087237149

Where:

  • zfs-pool-something - the name of the pool.
  • mirror - type of the RAID.
    • It can be:
      • mirror - RAID 1, mirror.
      • nothing - RAID 0, stripe.
      • raidz, raidz2, raidz3 - RAID-Z, where 1, 2 or 3 drives can be broken or replaced without loosing data.
  • /dev/disk/by-id/scsi-SATA_ST2000DM008-2FR1_ZFL16E1K and /dev/disk/by-id/wwn-0x5000c50087237149 - used drives.

In order to check the pool you can use following commands:

  • zpool status
  • zpool list
  • zfs list

Creating a pool with tuning

The default ZFS setting are good, but depending on the situation I may change some of them.

Generally I create a zfs pool like that:

  1. I erase all partitions from a drive.
  2. sudo zpool create -o ashift=12 -f zfs-pool-something /dev/disk/by-id/
  3. sudo zfs set recordsize=128k zfs-pool-something
  4. sudo zfs set compression=lz4 zfs-pool-something
  5. sudo zfs set xattr=sa zfs-pool-something
  6. sudo zfs set atime=off zfs-pool-something
  7. I create datasets with tuning.

Alignment Shift (ashift) determines the size of a block that ZFS will use. ashift is a number that represents power of 2 (ashift=2^x).

ashift size description
0 auto-detect zfs should auto-detect the sector size.
9 512b I don’t recommend using it at all
12 4096b Good default value
13 8192b Good value that sometimes is supported by special SSDs (some of samsung SSDs may support this value)

Record size (recordsize) - maximum size of a logical block in a ZFS dataset.

recordsize description
64k I use it rarely
128k Default good value
1M Good value for media and torrent pools

Compression (compression).

compression description
off Disables compression, I don’t recommend it
on zfs should use and choose compression (usually it’s lz4)
lz4 Great default compression that I use all the time
gzip gzip compression which may provide better compression ratio but generally requires more resources in order to compress data. It’s good on systems with external compression hardware (like Intel QuickAssist)

Check a documentation and run some tests if you need to have some sophisticated compression. On my system I have approximately 5 compression algorithms with tuning:

  • I can use any of the next values: on, off, lzjb, gzip, gzip-[1-9], zle, lz4, zstd, zstd-[1-19], zstd-fast and zstd-fast-[1-10,20,30,40,50,60,70,80,90,100,500,1000].

Extended Attributes (xattr) - how zfs should handle Linux extended attributes in a file system. I use xattr=sa.

Access time (atime) - updating atime on a file read. I turn it off atime=off.

You can see the list of all available zfs settings by using zfs get.

In order to see what settings are used by the pool or dataset, you can use zfs get:

  1. zpool get all | grep ashift
  2. zfs get recordsize zfs-pool-something
  3. zfs get compression zfs-pool-something
  4. zfs get xattr zfs-pool-something
  5. zfs get atime zfs-pool-something

Creating a dataset with tuning

It’s simple. Just choose the tuning settings and use zfs create:

  • sudo zfs create -o recordsize=128k -o compression=lz4 -o dedup=off -o sync=standard -o atime=off -o xattr=sa zfs-pool-something/dataset

Where:

  • zfs-pool-something - the name of the pool on which you want to create a dataset.
  • dataset - the name of the dataset.

Creating an encrypted dataset with tuning

There are different ways to create an encrypted dataset. It’s possible to use certs and hardware keys, but easiest one is to use a passphrase:

  • sudo zfs create -o recordsize=128k -o compression=lz4 -o dedup=off -o sync=standard -o atime=off -o xattr=sa -o encryption=on -o keyformat=passphrase zfs-pool-something/encrypted-dataset

Where encryption=on and keyformat=passphrase tell zfs that you want to use encryption and use a passphrase as a key.

To mount an encrypted partition do the following:

  1. sudo zpool import zfs-pool-something/encrypted-dataset
  2. sudo zfs load-key zfs-pool-something/encrypted-dataset
  3. sudo zfs mount zfs-pool-something/encrypted-dataset

To unmount an encrypted partition:

  1. sudo zfs umount zfs-pool-something/encrypted-dataset
  2. sudo zfs unload-key zfs-pool-something/encrypted-dataset

Deleting and exporting a pool

To destroy a pool use zpool destroy. It will destroy the chosen pool.

Sometimes you want to just disconnect the pool from a system. One of the most often reasons I do so when I want to move a pool from one server to another. And to do so I use zpool export zfs-pool-something.

Creating a snapshot

zfs snapshot allows you to create a snapshot. In order to create a recursive snapshot (a snapshot for all descendent file systems) you need to add -r option.

Example: zfs snapshot -r zfs-pool-something@2024-10-09 Where:

  • zfs snapshot -r - create a recursive snapshot.
  • zfs-pool-something - the name of the pool.
  • 2024-10-09 - the name of the snapshot.

A simple script for creating snapshots

Snapshots are useful when you need to save a read-only copy of a file system or volume. Even though you can create snapshots manually, there are at least two simple ways that can help you automate this process:

  • Can create a bash script and crontab job.
  • Use something similar to zfs_autobackup.

My bash script for creating recursive snapshots looks like that:

#!/bin/bash

date=$(date '+%Y-%m-%d_%H:%M')

/usr/sbin/zfs snapshot -r zfs-pool-something@$date

This script will create recursive snapshots of the zfs-pool-something pool with the name of the date they were taken and it will look like this: zfs-pool-something@2024-10-09_01:00.

In order to show snapshots for the specific pool you can use zfs list:

  • zfs list -t snapshot zfs-pool-something

And to show all snapshots of all pools, simply write the following:

  • zfs list -t snapshot -o name

Deleting snapshots

Snapshots will be deleted by zfs destroy command with specifying the snapshot you want to delete: zfs destroy zfs-pool-something@snapshot.

If you created snapshots with the script from above, you would be able to destroy a range of snapshots that follow a certain pattern. Usually I delete snapshots recursively except the last 16 or the last one.

To keep only the last 16 I use the following command:

  • zfs list -t snapshot -o name | grep ^zfs-pool-something@2024-10- | tac | tail -n +16 | xargs -n 1 zfs destroy -r

And to keep only the last one:

  • zfs list -t snapshot -o name | grep ^zfs-pool-something@2024-10- | tac | tail -n +2 | xargs -n 1 zfs destroy -r

Where:

  • zfs-pool-something - the name of the pool.
  • 2024-10- - a period of time that follows a specific pattern. Sometimes I want to delete all snapshots except the last one for a certain month (in this example for October 2024). Sometimes the last one of a certain year. Then the pattern will look equivalent to 2024-.
  • tail -n +16 and tail -n +2 - describe the amount of snapshots to keep.
Author: Savelii Pototskii
Date:
Edit this page on GitHub
Copyright: CC-BY
Categories: linux