ZFS Cheatsheet

This is a quick and dirty cheatsheet on Sun's ZFS

Directories and Files
error messages /var/adm/messages
console
States
DEGRADED One or more top-level devices is in the degraded state because they have become offline. Sufficient replicas exist to keep functioning
FAULTED One or more top-level devices is in the faulted state because they have become offline. Insufficient replicas exist to keep functioning
OFFLINE The device was explicity taken offline by the "zpool offline" command
ONLINE The device is online and functioning
REMOVED The device was physically removed while the system was running
UNAVAIL The device could not be opened
Scrubbing and Resilvering
Scrubbing Examines all data to discover hardware faults or disk failures, only one scrub may be running at one time, you can manually scrub.
Resilvering is the same concept as rebuilding or resyncing data on to new disks into an array, the smart thing resilvering does is it does not rebuild the whole disk only the data that is required (the data blocks not the free blocks) thus reducing the time to resync a disk. Resilvering is automatic when you replace disks, etc. If a scrub is already running it is suspended until the resilvering has finished and then the scrubbing will continue.
ZFS Devices
Disk A physical disk drive
File The absolute path of pre-allocated files/images
Mirror Standard raid-1 mirror
Raidz1/2/3

## non-standard distributed parity-based software raid levels, one common problem called "write-hole" is elimiated because raidz in ## zfs the data and stripe are written simultanously, basically is a power failure occurs in the middle of a write then you have the ## data plus the parity or you dont, also ZFS supports self-healing if it cannot read a bad block it will reconstruct it using the
## parity, and repair or indicate that this block should not be used.

## You should keep the raidz array at a low power of two plus partity
raidz1 - 3, 5, 9 disks
raidz2 - 4, 6, 8, 10, 18 disks
raidz3 - 5, 7, 11, 19 disks

## the more parity bits the longer it takes to resilver an array, standard mirroring does not have the problem of creating the parity
## so is quicker in resilvering

## raidz is more like raid3 than raid5 but does use parity to protect from disk failures
raidz/raidz1 - minimum of 3 devices (one parity disk), you can suffer a one disk loss
raidz2         - minimum of 4 devices (two parity disks), you can suffer a two disk loss
raidz3         - minimum of 5 devices (three parity disks) , you can suffer a three disk loss

spare hard drives marked as "hot spare" for ZFS raid, by default hot spares are not used in a disk failure you must turn on the "autoreplace" feature.
cache

Linux caching mechanism use what is known as least recently used (LRU) algorithms, basically first in first out (FIFO) blocks are moved in and out of cache. Where ZFS cache is different it caches both least recently used block (LRU) requests and least frequent used (LFU) block requests, the cache device uses level 2 adaptive read cache (L2ARC).

log

There are two terminologies here

  • ZFS intent log (ZIL) - a logging mechanism where all the data to be written is stored, then later flushed as a transactional write, this is similar to a journal filesystem (ext3 or ext4).
  • Seperate intent log (SLOG) - a seperate logging devive that caches the synchronous parts of the ZIL before flushing them to the slower disk, it does not cache asynchronous data (asynchronous data is flushed directly to the disk). If the SLOG exists the ZIL will be moved to it rather than residing on platter disk, everything in the SLOG will always be in system memory. Basically the SLOG is the device and the ZIL is data on the device.
Storage Pools
displaying

zpool list
zpool list -o name,size,altroot

# zdb can view the inner workings of ZFS (zdb has a number of options)
zdb <option> <pool>

Note: there are a number of properties that you can select, the default is: name, size, used, available, capacity, health, altroot

status zpool status

## Show only errored pools with more verbosity
zpool status -xv      
statistics zpool iostat -v 5 5

Note: use this command like you would iostat
history zpool history -il

Note: once a pool has been removed the history is gone
creating

## You cannot shrink a pool only grow it

## performing a dry run but don't actual perform the creation (notice the -n)
zpool create -n data01 c1t0d0s0

# you can persume that I created two files called /zfs1/disk01 and /zfs1/disk02 using mkfile
zpool create data01 /zfs1/disk01 /zfs1/disk02

# using a standard disk slice
zpool create data01 c1t0d0s0

## using a different mountpoint than the default /<pool name>
zpool create -m /zfspool data01 c1t0d0s0

# mirror and hot spare disks examples, hot spares are not used by default turn on the "autoreplace" feature for each pool
zpool create data01 mirror c1t0d0 c2t0d0 mirror c1t0d1 c2t0d1
zpool create data01 mirror c1t0d0 c2t0d0 spare c3t0d0

## setting up a log device and mirroring it
zpool create data01 mirror c1t0d0 c2t0d0 log mirror c3t0d0 c4t0d0

## setting up a cache device
zpool create data 01 mirror c1t0d0 c2t0d0 cache c3t0d0 c3t1d0

## you can also create raid pools (raidz/raidz1 - mirror, raidz2 - single parity, raidz3 double partity)
zpool create data01 raidz2 c1t0d0 c1t1d0 c1t2d0 c1t3d0 c1t4d0

destroying zpool destroy /zfs1/data01

## in the event of a disaster you can re-import a destroyed pool
zpool import -f -D -d /zfs1 data031
adding zpool add data01 c2t0d0

Note: make sure that you get this right as zpool only supports the removal of hot spares and cache disks, for mirrors see attach and detach below
Resizing ## When replacing a disk with a larger one you must enable the "autoexpand" feature to allow you to use the extended space, you must do this before replacing the first disk
removing zpool remove data01 c2t0d0

Note: zpool only supports the removal of hot spares and cache disks, for mirrors see attach and detach below
clearing faults zpool clear data01

## Clearing a specific disk fault
zpool clear data01 c2t0d0
attaching (mirror) ## c2t0d0 is an existing disk that is not mirrored, by attaching c3t0d0 both disks will become a mirror pair
zpool attach data01 c2t0d0 c3t0d0
detaching (mirror) zpool detach data01 c2t0d0

Note: see above notes is attaching
onlining zpool online data01 c2t0d0
offlining

zpool offline data01 c2t0d0

## Temporary offlining (will revent back after a reboot)
zpool offline data01 -t c2t0d0

Replacing ## replacing like for like
zpool replace data03 c2t0d0

## replacing with another disk
zpool replace data03 c2t0d0 c3t0d0
scrubbing

zpool scrub data01

## stop a scrubbing in progress, check the scrub line using "zpool status data01" to see any errors
zpool scrub -s data01

Note; see top of table for more information about resilvering and scrubbing

exporting

zpool export data01

## you can list exported pools using the import command
zpool import

importing

## when using standard disk devices i.e c2t0d0
zpool import data01

## if using files in say the /zfs filesystem
zpool import -d /zfs

## importing a destroyed pool
zpool import -f -D -d /zfs1 data03

getting parameters zpool get all data01

Note: the source column denotes if the value has been change from it default value, a dash in this column means it is a read-only value
setting parameters zpool set autoreplace=on data01

Note: use the command "zpool get all <pool>" to obtain list of current setting
upgrade ## List upgrade paths
zpool upgrade -v

## upgrade all pools
zpool upgrade -a

## upgrade specific pool, use "zpool get all <pool>" to obtain version number of a pool
zpool upgrade data01

## upgrade to a specific version
zpool upgrade -V 10 data01
Filesystem
displaying

zfs list

## list different types
zfs list -t filesystem
zfs list -t snapshot
zfs list -t volume

zfs list -t all -r <zpool>

## recursive display
zfs list -r data01/oracle

## complex listing
zfs list -o name,mounted,sharenfs,mountpoint

Note: there are a number of attributes that you can use in a complex listing, so use the man page to see them all

creating

## persuming i have a pool called data01 create a /data01/apache filesystem
zfs create data01/apache

## using a different mountpoint
zfs create -o mountpoint=/oracle data01/oracle

## create a volume - the device can be accessed via /dev/zvol/[rdsk|dsk]/data03/swap
zfs create -V 50mb data01/swap
swap -a /dev/zvol/dsk/data01/swap

Note: don't use a zfs volume as a dump device it is not supported

destroying zfs destroy data01/oracle

## using the recusive options -r = all children, -R = all dependants
zfs destroy -r data01/oracle
zfs destroy -R data01/oracle
mounting

zfs mount data01

# you can create temporary mount that expires after unmounting
zfs mount -o mountpoint=/tmpmnt data01/oracle

Note: there are all the normal mount options that you can apply i.e ro/rw, setuid

unmounting zfs umount data01
share

zfs share data01

## Persist over reboots
zfs set sharenfs=on data01

## specific hosts
zfs set sharenfs="rw=@10.85.87.0/24" data01/apache

unshare zfs unshare data01

## persist over reboots
zfs set sharenfs=off data01
snapshotting

## snapshotting is like taking a picture, delta changes are recorded to the snapshot when the original file system changes, to
## remove a dataset all previous snaphots have to be removed, you can also rename snapshots.
## You cannot destroy a snapshot if it has a clone

## creating a snapshot
zfs snapshot data01@10022010

## renaming a snapshot
zfs snapshot data01@10022010 data01@keep_this

## destroying a snapshot
zfs destroy data01@10022010

rollback ## by default you can only rollback to the lastest snapshot, to rollback to older one you must delete all newer snapshots
zfs rollback data01@10022010
cloning/promoting

## clones are writeable filesystems that was upgraded from a snapshot, a dependency will remain on the snapshot as long as the
## clone exists. A clone uses the data from the snapshot to exist. As you use the clone it uses space separate from the snapshot.

## clones cannot be created across zpools, you need to use send/receive see below topics

## cloning
zfs clone data01@10022010 data03/clone
zfs clone -o mountpoint=/clone data01@10022010 data03/clone

## promoting a clone, this allows you to destroy the original file ssytem that the clone is attached to
zfs promote data03/clone

Note: the clone must reside in the same pool

renaming

## the dataset must be kept within the same pool
zfs rename data03/ora_disk01 data03/ora_d01

Note: you have two options
-p creates all the non-existing parent datasets
-r recursively rename the sanpshots of all descendent datasets (used with snapshots only)

Compression

## You enable compression by seeting a feature, compressions are on, off, lzjb, gzip, gzip[1-9] ans zle, not that it only start
## compression when you turn it on, other existing data will not be compressed
zfs set compression=lzjb data03/apache

## you can get the compression ratio
zfs get compressratio data03/apache

Deduplication

## you can save disk space using deduplication which can be on file, block or byte, for example using file each file is hashed with a
## cryptographic hashing algorithm such as SHA-256, if a file matches then we just point to the existing file rather than storing a
## new file, this is ideal for small files but for large files a single character change would mean that all the data has to be copied

## block deduplication allows you to share all the same blocks in a file minus the blocks that are different, this allows to share the
## unique blocks on disk and the reference shared blocks in RAM, however it may need a lot of RAM to keep track of which blocks
## are shared and which are not., however this is the preferred option other than file or byte deduplication. Shared blocks are
## stored in what is called a "deduplication table", the more deduplicated blocks the larger the table, the table is read everytime
## to make a block change thus the table should be held in fast RAM, if you run out of RAM then the table will spillover onto disk.

## So how much RAM do you need, you can use the zdb command to check, take the "bp count", it takes about 320 bytes of ram
## for each deduplicate block in the pool, so in my case 288674 means I would need about 92MB, for example a 200GB would need
## about 670MB for the table, a good rule would be to allow 5GB of RAM for every 1TB of disk.

## to see the block the dataset consumes
zdb -b data01

## to turn on deduplicate
zfs set dedup=on data01/text_files

## to see the deduplicatio ratio
zfs get dedupratio data01/text_files

## to see the histrogram of howm many blocks are referenced how many time
zdb -DD <pool>

getting parameters

## List all the properties
zfs get all data03/oracle

## get a specific property
zfs get setuid data03/oracle

## get a list of a specific properites for all datasets
zfs get compression

Note: the source column denotes if the value has been change from it default value, a dash in this column means it is a read-only value

setting parameters

## set and unset a quota
zfs set quota=50M data03/oracle
zfs set quota=none data03/oracle

Note: use the command "zfs get all <dataset> " to obtain list of current settings

inherit ## set back to the default value
zfs inherit compression data03/oracle
upgrade

## List the upgrade paths
zfs upgrade -v

## List all the datasets that are not at the current level
zfs upgrade

## upgrade a specific dataset
upgrade -V <version> data03/oracle

send/receive

## here is a complete example of a send and receive with incremental update

## create some test files
mkfile -v 100m /zfs/master
mkdir -v 100m /zfs/slave

## create mountpoints
mkdir /master
mkdir /slave

## Create the pools
zpool create master
zpool create slave

## create the data filesystem
zfs create master/data

## create a test file
echo "created: 09:58" > /master/data/test.txt

## create a snapshot and send it to the slave, you could use SSH or tape to transfer to another server (see below)
zfs snapshot master/data@1
zfs send master/data@1 | zfs receive slave/data

## set the slave to read-only because you can cause data corruption, make sure if do this before accessing anything the
## slave/data directory
zfs set readonly=on slave/data

## update the original test.txt file
echo "`date`" >> /master/data/text.txt

## create a second snapshot and send the differences, you may get an error message saying that the desination had been
## modified this is because you did not set the slave/data to ready only (see above)
zfs snapshot master/data@2
zfs send -i master/data@1 master/data@2 | zfs receive slave/data

---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

## using SSH
zfs send master/data@1 | ssh backup_server zfs receive backups/data@1

## using a tape drive, you can also use cpio
zfs send master/data@1 > /dev/rmt/0
zfs receive slave/data2@1 < /dev/rmt/0
zfs rename slave/data slave/data.old
zfs rename slave/data2 slave/data

## you can also save incremental data
zfs send master/data@12022010 > /dev/rmt/0
zfs send -i master/data@12022010 master/data@13022010 > /dev/rmt/0

## Using gzip to compress the snapshot
zfs send master/fs@snap | gzip > /dev/rmt/0

allow/unallow

## display the permissions set and any user permissions
zfs allow master

## create a permission set
zfs allow -s @permset1 create,mount,snapshot,clone,promote master

## delete a permission set
zfs unallow -s @permset1 master

## grant a user permissions
zfs allow vallep @permset1 master

## revoke a user permissions
zfs unallow vallep @permset1 master

Note: there are many permissions that you can set so see the man page or just use the "zfs allow" command

Quota/Reservation

## Not strickly a command but wanted to discuss here, you can apply a quota to a dataset, you can reduce this quota only if the
## quota has not already exceeded, if you exceed the quota you will get a error message, you also have reservations which will
## guarantee that a specified amount of disk space is available to the filesystem, both are applied to datasets and there
## descendants (snapshots, clones)

## Newer versions of Solaris allow you to set group and user quota's

## you can also use refquota and refreservation to manage the space without accounting for disk space consumed by descendants
## such as snapshots and clones. Generally you would set quota and reservation higher than refquota and refreservation

  • quota & reservation - properties are used for managing disk space consumed by datasets and their descendants
  • refquota & refreservation - properties are used for managing disk space consumed by datasets only

## set a quota
zfs set quota=100M data01/apache

## get a quota
zfs get quota data01/apache

## setup user quota (use groupquota for groups)
zfs set userquota@vallep=100M data01/apache

## remove a user quota (use groupquota for groups)
zfs set userquota@vallep=none data01/apache

## List user quota (use groupspace for groups), you can alsolist users with quota's for exampe root user
zfs userspace data01/apache
zfs get userused@vallep data01/apache

ZFS tasks
Replace failed disk

# List the zpools and identify the failed disk
zpool list

# replace the disk (can use same disk or new disk)
zpool replace data01 c1t0d0
zpool replace data01 c1t0d0 c1t1d0

# clear any existing errors
zpool clear data01

# scrub the pool to check for anymore errors (this depends on the size of the zpool as it can take a long time to complete
zpool scrub data01

# you can now remove the failed disk in the normal way depending on your hardware

Expand a pools capacity # you cannot remove a disk from a pool but you can replace it with a larger disk
zpool replace data01 c1t0d0 c2t0d0
zpool set autoexpand=on data01
Install the boot block # the command depends if you are using a sparc or a x86 system
sparc - installboot -F zfs /usr/platform/`uname -i`/lib/fs/zfs/bootblk /dev/rdsk/c0t1d0
x86    - installgrub /boot/grub/stage1 /boot/grub/stage2 /dev/rdsk/c0t1d0s0
Lost root password

# You have two options to recover the root password

## option one
ok> boot -F failsafe
     whne requested follow the instructions to mount the rpool on /a
cd /a/etc
vi passwd|shadow
init 6

## Option two
ok boot cdrom|net -s (you can boot from the network or cdroml)
zpool import -R /a rpool
zfs mount rpool/ROOT/zfsBE
cd /a/etc
vi passwd|shadow
init 6

Primary mirror disk in root is unavailable or fails

# boot the secondary mirror
ok> boot disk1

## offline and unconfigure failed disk, there may be different options on unconfiguring a disk depends on the hardware
zpool offline rpool c0t0d0s0
cfgadm -c unconfigure c1::dsk/c0t0d0

# Now you can physically replace the disk, reconfigure it and bring it online
cfgadm -c configure c1::dsk/c0t0d0
zpool online rpool c0t0d0

# Let the pool know you have replaced the disk
zpool replace rpool c0t0d0s0

# if the replace above fails the detach and reattach the primary mirror
zpool deatch rpool c0t0d0s0
zpool attach rpool c0t1d0s0 c0t0d0s0

# make checks
zpool status rpool

# dont forget to add the boot block (see above)

Resize swap area (and dump areas)

# You can resize the swap if it is not being used, first record the size and if it is being used
swap -l

# resize the swap area, first by removing it
swap -d /dev/zvol/dsk/rpool/swap
zpool set volsize=2G rpool/swap

# Now activate the swap and check the size, if the -a option does not work then use "swapadd" command
swap -a /dev/zvol/dsk/rpool/swap
swap -l

Note: if you cannot delete the original swap area due to being too busy then simple add another swap area, the same procedure is used for dump areas but using the "dumpadm" command