Switch branches/tags
Nothing to show
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
107 lines (93 sloc) 5.5 KB
This is a backup script to replicate a ZFS filesystem and its children to
another server via zfs snapshots and zfs send/receive over ssh. It was
developed on Solaris 10 but should run with minor modification on other
platforms with ZFS support.
It supplements zfs-auto-snapshot, but runs independently. I prefer that
snapshots continue to be taken even if the backup fails. It does not
necessarily require that package -- anything that regularly generates
snapshots that follow a given pattern will suffice.
Command-line options:
-n debug/dry-run mode
-v verbose mode
-f file specify a configuration file
-r N use the Nth most recent local snapshot rather than the newest
-h, -? display help message
Basic installation: After following the prerequisites, run manually to verify
operation, and then add a line like the following to zfssnap's crontab:
30 * * * * /path/to/ [ options ]
(This for an hourly sync -- adjust accordingly if you only want to back up
daily, etc. zfs-backup now supports commandline options and configuration
files, so you can schedule different cron jobs with different config files,
e.g. to back up to two different targets. If you schedule multiple cron
jobs, you should use different lockfiles in each configuration.)
This aims to be much more robust than the backup functionality of
zfs-auto-snapshot, namely:
* it uses 'zfs send -I' to send all intermediate snapshots (including
any daily/weekly/etc.), and should still work even if it isn't run
every hour -- as long as the newest remote snapshot hasn't been
rotated out locally yet
* 'zfs recv -dF' on the destination host removes any snapshots not
present locally so you don't have to worry about manually removing
old snapshots there.
1. zfs-auto-snapshot or equivalent package installed locally and regular
snapshots enabled (hourly, daily, etc.)
2. home directory set for zfssnap role (the user taking snapshots and doing
the sending):
# rolemod -d /path/to/home zfssnap
3. ssh keys set up between zfssnap@localhost and remuser@remhost:
# su - zfssnap
$ ssh-keygen
Copy the contents of .ssh/ into ~remuser/.ssh/authorized_keys on
remhost. Test that key-based ssh works:
$ ssh remuser@remhost
4. zfs allow done for remuser on remhost:
# zfs allow remuser atime,create,destroy,mount,mountpoint,receive,rollback,snapshot,userprop backuppool/fs
This can be done on a top-level filesystem, and is inherited by default.
Depending on your usage, you may need to also allow further permissions such
as share, sharenfs, hold, etc.
5. an initial (full) zfs send/receive done so that remhost has the fs we
are backing up, and the associated snapshots -- something like:
zfs send -R $POOL/$FS@zfs-auto-snap_daily-(latest) | ssh $REMUSER@$REMHOST zfs recv -dvF $REMPOOL
Note: 'zfs send -R' will send *all* snapshots associated with a dataset, so
if you wish to purge old snapshots, do that first.
6. zfs allow any additional permissions needed, to fix any errors produced in step 5
7. configure the TAG/PROP/REMUSER/REMHOST/REMPOOL variables in this script or in a config file
8. zfs set $PROP={ fullpath | basename | rootfs } pool/fs
for each FS or volume you wish to back up.
Given the hierarchy pool/a/b,
* with 'fullpath' (zfs recv -d), this is replicated to backupserver:backuppool/a/b
* with 'basename' (zfs recv -e), this is replicated to backupserver:backuppool/b
This is useful for replicating a sub-level FS into the top level of the backup pool;
e.g. pool/backup/foo => backuppool/foo (instead of backuppool/backup/foo)
* with 'rootfs' set on pool (the root filesystem in the pool; uses zfs recv -d
with target set to $REMPOOL), pool is replicated to backupserver:backuppool.
It is an error to set this property value on any child filesystem.
WARNING: This can be dangerous -- any filesystems in $REMPOOL which do not
exist in the source will be deleted! For reasons of safety and simplicity,
it is usually preferable to work with ZFS filesystems rather than the root fs,
or use the 'fullpath' property value, which will receive a root filesystem
into a child filesystem of the same name, otherwise replicate all children
into top-level child filesystems, and not touch any unknown filesystems.
If this backup is not run for a long enough period that the newest
remote snapshot has been removed locally, manually run an incremental
zfs send/recv to bring it up to date, a la
zfs send -I zfs-auto-snap_daily-(latest on remote) -R $POOL/$FS@zfs-auto-snap_daily-(latest local) |
It's probably best to do a dry-run first (zfs recv -ndvF).
Note: I use daily snapshots in these manual send/recv examples because
it is less likely that the snapshot you are using will be rotated out
in the middle of a send. Also, note that ZFS will send all snapshots for a
given filesystem before sending any for its children, rather than going in
global date order.
Alternatively, use a different tag (e.g. weekly) that still has common
snapshots, possibly in combination with the -r option (Nth most recent) to
avoid short-lived snapshots (e.g. hourly) being rotated out in the middle
of your sync. This is a good use case for an alternate configuration file.
* find newest local hourly snapshot
* find newest remote hourly snapshot (via ssh)
* check that both $newest_local and $latest_remote snaps exist locally
* zfs send incremental (-I) from $newest_remote to $latest_local to dsthost
* if anything fails, set svc to maint. and exit