The zsync
daemon auto-snapshots ZFS datasets, replicates datasets, and
auto-deletes stale snapshots. The replication can be performed across machines
in either a push or pull mechanism using either SSH or local OS subprocesses.
This only tested with ZFS on Linux.
In order for the daemon to properly perform ZFS operations, the zfs allow
feature must be used to enable permissions on certain operations.
The following permissions should be granted:
# On all sources:
sudo zfs allow $USER send,snapshot,destroy,mount $DATASET
# On all mirrors:
sudo zfs allow $USER receive,create,mount,mountpoint,readonly,snapshot,destroy,mount $DATASET
The mountpoint
and readonly
properties are only need when performing the
first replication, where the receiving side sets mountpoint=none
and
readonly=on
.
The zpool
and zfs
tool must work without sudo
.
See the "Changes in Behavior" section of the ZoL v0.7.0 release.
go get -u github.com/dsnet/zsync
The instruction steps below assume $GOPATH/bin
is in your $PATH
,
so that the zsync
binary can be executed directly from the shell.
The operation of zsync
is configured using a JSON configuration file,
where the full list of features can be listed by running zsync -help
.
Here is an example configuration file (adjust as appropriate):
{
"SSH": {
"KeyFiles": ["/home/user/.ssh/id_rsa"],
"KnownHostFiles": ["/home/user/.ssh/known_hosts"],
},
"AutoSnapshot": {"Cron": "@daily", "Count": 7},
"Datasets": [{
"Source": "//localhost/tank/dataset",
"Mirrors": ["//user@remotehost:22/tank/dataset-mirror"],
}],
}
This example auto-snapshots daily and only keeps 7 snapshots.
It replicates tank/dataset
on the local machine to tank/dataset-mirror
on a remote machine using an SSH push.
Start the daemon with:
zsync /path/to/config.json