TimeMachine style backups for ZFS users. ZFS-Timemachine creates incremental backups of zfs datasets on one host to datasets on another disk or host. This is done via sending snapshots, deleting old ones in time machine style. It works with FreeBSD and Macs (with TensCompliments ZFS implementation), but should work with other ZFS implementations as well.
How it works
- the script creates a snapshot on the source dataset every time it is called.
- then it figures out the last snapshot on the destination dataset that matches to one on the source dataset.
- it sends the snapshot from the source to the destination.
- removes old snapshots on the source - it keeps just n-snapshots.
- removes old snapshots on the destination - time machine fashion : 5min/last day, 1 hour last week, 1 day last 3 months, 1 week thereafter
It requires perl and the Time::Local and Date::Parse libraries. If you are on a Mac you can install them by using the command line:
$export PERL_MM_USE_DEFAULT=1 ; perl -MCPAN -e 'install Date::Parse' 'install Time::Local'
If you are on a different OS (like linux or bsd) everything should work.
How to use
The simplest use of the script requires just two options, --sourcedataset and --destinationdataset options. Like this:
$ sudo zfstimemachinebackup.perl --sourcedataset=tank --destinationdataset=root/backup
Usually you want the script to create a snapshot on the sourcedataset when it is called, so add the --createsnapshotonsource option. To see all options use the --help commandline option.
As there are quite a few options, let's go through them in detail:
- --sourcehost (string): hostname where the source dataset is on.
- --sourcehostoptions (string): options given to ssh (default: -c blowfish -C -l root).
- --sourcedataset (string): the source dataset that is to be backed up.
- --destinationhost (string): hostname where the destination dataset is on.
- --destinationhostoptions (string): options given to ssh (default: -c blowfish -C -l root)
- --destinationdataset (string): the destination dataset where backups should be stored.
- --createsnapshotonsource (flag): When set the script will create a new snapshot on the source dataset everytime it is called.
- --snapshotstokeeponsource (number): How many snapshots we should keep on the source dataset. More datasets on source will be deleted (oldest beeing deleted first). If set to 0 no snapshots will be removed on the source. See also --minimumtimetokeepsnapshotsonsource option.
- --minimumtimetokeepsnapshotsonsource (string): Minimum time how long snapshots should exist on the source. With this set snapshots on the source will be kept at least that long even if there are more than the number of snapshots given in the --snapshotstokeeponsource option. (Eg: 1week, 1month or something like that).
- --replicate (flag): Only needed for the very first backup. It will replicate all snapshots from the source to the destination.
- --recursive (flag): Should we backup all decendent datasets on the source to the destination.
- --datasetstoignoreonsource (string): If you are recursivly backing up, you can disable backing up datasets that match this comma seperated list of datasets.
--deletesnapshotsondestination (flag): Should old snapshots on the destination be deleted.
--keepbackupshash (string): A comma seperated list of value pairs that define the granularity of how many snapshots are kept on the destination when they are getting older. The default is 24h=>5min,7d=>1h,90d=>1d,1y=>1w,10y=>1month which means:
24h=>5mi for snapshots younger than 24hours: keep not more than one per 5 minutes 7d=>1h for snapshots younger than 7 days: keep not more than one snapshot per 1 hour . . .
--maximumtimeperfilesystemhash (string) default: A comma seperated list of value pairs that define the granularity of how old snapshots can get on the destination. Special datasets might not be as important as others. Default of .*=>3months,.+/(Dropbox|Downloads|Caches|Mail Downloads|Saved Application State|Logs)$=>1month means:
.*=>10yrs keep everything 10 years by default - after that snapshots are removed .+/(Dropbox|Downloads|Caches|Mail Downloads|Saved Application State|Logs)$=>1month remove snapshots older than one month for datasets ending with the regex.
- --configurationfilename (string): config.ini filename the defaults are read from. Only works if you Config::Inifiles installed.
- --debug (number): debugging level of the script itself. When set will also enable verbose.
- --verbose (flag): showing more verbosely what is going on.
- --help (flag): Shows all options of the script and all values without starting the script.
My current setup looks like this:
$ zfs list puddle 207Gi 214Gi 864Ki /Volumes/puddle puddle/Local 207Gi 214Gi 2.50Gi /Local puddle/Local/Users 204Gi 214Gi 891Mi /Local/Users puddle/Local/Users/jolly 204Gi 214Gi 50.4Gi /Local/Users/jolly puddle/Local/Users/jolly/Disks 22.4Gi 214Gi 22.3Gi /Local/Users/jolly/Disks puddle/Local/Users/jolly/Downloads 1.62Gi 214Gi 1.62Gi /Local/Users/jolly/Downloads puddle/Local/Users/jolly/Dropbox 3.53Gi 214Gi 3.53Gi /Local/Users/jolly/Dropbox puddle/Local/Users/jolly/Library 44.6Gi 214Gi 28.3Gi /Local/Users/jolly/Library puddle/Local/Users/jolly/Library/Caches 2.05Gi 214Gi 2.04Gi /Local/Users/jolly/Library/Caches puddle/Local/Users/jolly/Library/Logs 72.2Mi 214Gi 70.4Mi /Local/Users/jolly/Library/Logs puddle/Local/Users/jolly/Library/Mail 13.8Gi 214Gi 13.7Gi /Local/Users/jolly/Library/Mail puddle/Local/Users/jolly/Library/Mail Downloads 868Ki 214Gi 868Ki /Local/Users/jolly/Library/Mail Downloads puddle/Local/Users/jolly/Library/Saved Application State 50.8Mi 214Gi 9.38Mi /Local/Users/jolly/Library/Saved Application State puddle/Local/Users/jolly/Pictures 80.4Gi 214Gi 80.4Gi /Local/Users/jolly/Pictures ocean 1.24Ti 567Gi 266Ki /Volumes/ocean ocean/puddle 635Gi 567Gi 187Ki /Volumes/ocean/puddle ocean/puddle/Local 635Gi 567Gi 1.76Gi /Volumes/ocean/puddle/Local ocean/puddle/Local/Users 632Gi 567Gi 539Mi /Volumes/ocean/puddle/Local/Users ocean/puddle/Local/Users/jolly 631Gi 567Gi 49.8Gi /Volumes/ocean/puddle/Local/Users/jolly ocean/puddle/Local/Users/jolly/Disks 48.0Gi 567Gi 22.1Gi /Volumes/ocean/puddle/Local/Users/jolly/Disks ocean/puddle/Local/Users/jolly/Downloads 1.62Gi 567Gi 1.62Gi /Volumes/ocean/puddle/Local/Users/jolly/Downloads ocean/puddle/Local/Users/jolly/Dropbox 4.42Gi 567Gi 3.47Gi /Volumes/ocean/puddle/Local/Users/jolly/Dropbox ocean/puddle/Local/Users/jolly/Library 93.7Gi 567Gi 22.6Gi /Volumes/ocean/puddle/Local/Users/jolly/Library ocean/puddle/Local/Users/jolly/Library/Caches 1.90Gi 567Gi 1.90Gi /Volumes/ocean/puddle/Local/Users/jolly/Library/Caches ocean/puddle/Local/Users/jolly/Library/Logs 65.2Mi 567Gi 65.1Mi /Volumes/ocean/puddle/Local/Users/jolly/Library/Logs ocean/puddle/Local/Users/jolly/Library/Mail 18.6Gi 567Gi 11.2Gi /Volumes/ocean/puddle/Local/Users/jolly/Library/Mail ocean/puddle/Local/Users/jolly/Library/Mail Downloads 210Ki 567Gi 208Ki /Volumes/ocean/puddle/Local/Users/jolly/Library/Mail Downloads ocean/puddle/Local/Users/jolly/Library/Saved Application State 12.4Mi 567Gi 5.83Mi /Volumes/ocean/puddle/Local/Users/jolly/Library/Saved Application State ocean/puddle/Local/Users/jolly/Pictures 85.7Gi 567Gi 73.8Gi /Volumes/ocean/puddle/Local/Users/jolly/Pictures
/Local is where my home directory lives. The script is called as follows
$ sudo ./zfstimemachinebackup.perl --sourcedataset=puddle --destinationdataset=ocean/puddle --snapshotstokeeponsource=100 --createsnapshotonsource --recursive
So puddle is set as source, ocean/puddle will receive the snapshots from puddle and 100 snapshots are kept on puddle itself.
I'm also sending backups from the backupdisk to a remote machine with less space, so I keep backups only for 3 months:
$ sudo ./zfstimemachinebackup.perl --sourcedataset=ocean/puddle --destinationdataset=backups/puddle --destinationhost=server.example.com --recursive --maximumtimeperfilesystemhash='.*=>3months,.+/(Dropbox|Downloads|Caches|Mail Downloads|Saved Application State|Logs)$=>1month'
My backupserver usually sleeps and so it might take a day or two before finishing a download. The autoscrub script will scrub the given pool after the time given.
usage: sudo ./autoscrub.perl --scrubinterval=14
This will scrub your pools every 14 days. If you cancel a scrub that will be recognized but also it will be scrubed after the scrubinterval passed, in case you forgot that you canceled it.
You can start it for different pools as well.
I'm using it in a crontab entry:
1 * * * * cd ~jolly/Binaries/ZFSTimeMachine;./autoscrub.perl >/dev/null 2>&1
Mac OS X only stuff
The following is only relevant to those who use Macs.
The checkbackup.perl script checks if your backupscript is working correctly. As I do sleep my machine it will check if the snapshots are beeing done within the last 2*snapshotinterval+snapshottime seconds since the last wake or reboot. Exit code is correct depending if the snapshot is there or not. If the checkbackupscript can't find out the last sleep and boot time it will bug you about backups beeing too old when the machine has beeing powerd off for some time.
It has three options :
--datasets which dataset(s) to use comma separated list --snaphotinterval how often do you create snapshots --snapshotime how long it usually take for a snapshot to complete $[checkbackup.perl] module options are : --configurationfilename (string) default: config.ini current: not used as Config:IniFiles module not present --debug (number) default: 0 --help (option) default: current: 1 --datasets (string) default: puddle --snapshotinterval (number) default: 300 --snapshottime (number) default: 10
I'm currently using a script at crontab (executing as root) to tell me when things go wrong:
#!/bin/zsh ./checkbackup.perl --datasets="puddle/Local,puddle/Local/Users,puddle/Local/Users/jolly,puddle/Local/Users/jolly/Library,puddle/Local/Users/jolly/Disks,puddle/Local/Users/jolly/Pictures" --snapshotinterval=7200 || say -v alex "dataset snapshot on local host is too old" ./checkbackup.perl --datasets="example.com:pond/puddle/Local,example.com:pond/puddle/Local/Users,example.com:pond/puddle/Local/Users/jolly,example.com:pond/puddle/Local/Users/jolly/Library,example.com:pond/puddle/Local/Users/jolly/Disks,example.com:pond/puddle/Local/Users/jolly/Pictures" --snapshotinterval=7200 || say -v alex "dataset pond snapshots on example.com are too old"
TimeMachine backups to ZFS Volumes
For those of you that want to use MacOS X's TimeMachine to backup to a ZFS volume you can create the needed sparsebundle with the following commands. I now have moved everything except for my boot partitions to ZFS. To have some backup of the root drive I'm backing that of with Apples provided TimeMachine and here is how I do it:
Create a zfs filesystem for the TimeMachine backups for several machines:
sudo zfs create ocean/TimeMachine
Create a 100Gb sparsebundle for TimeMachine (my root is rather small, your mileage may vary):
hdiutil create -size 100g -library SPUD -fs JHFSX -type SPARSEBUNDLE -volname "tmmachinename" /Volumes/ocean/TimeMachine/tmmachinename.sparsebundle
Set up crontab to mount the sparsebundle every 20 minutes if it's not mounted yet. This is needed as TimeMachine will unmount the backup disk if it's a sparsebundle after backing up.
*/20 * * * * if [ ! -d /Volumes/tmtinkerbell ] ;then hdiutil attach /Volumes/ocean/TimeMachine/tmmachinename.sparsebundle; fi </dev/null >/dev/null 2>&1
Set up TimeMachine to use the sparsebundle:
tmutil setdestination -p /Volumes/tmmachinename