The Netyeti's Journal

ZFS Backup

4 min read

Here’s a breakdown of the process of how to use ZFS to backup data to another system.

Create a ZFS Target for Receiving Snapshots 

  1. Access the Remote System: Log in to the system where you want to store the snapshots.
  2. Create a New ZFS Pool
    • Use the zpool create command with the desired pool name and the devices that will comprise the pool. For example, to create a pool named backup_pool on devices sda and sdb:
zpool create backup_pool /dev/sda /dev/sdb

This creates the fundamental storage pool. 3. Create a ZFS Dataset: - Inside the new pool, create the specific ZFS dataset that will hold the incoming snapshots. This dataset will be the target for the zfs receive command. - For example, to create a dataset named rpool/snapshots within the new backup_pool: 

zfs create backup_pool/snapshots

 

NOTE

If you are setting up for incremental sends, this dataset must exist on the receiving end.

Determine the Source

Use zfs list to view the pools on the source system and select one to use for the following process.

Create a Snapshot

On the Source System (where you want to back up from):

zfs snapshot -r source_pool/dataset@backup_snapshot_name

Example on my system:

sudo zfs snapshot -r rpool/USERDATA/home_dpwraq@home_20250916

This command creates a recursive snapshot (-r) of the source_pool/dataset and names it backup_snapshot_name. Replace source_pool/dataset with your actual ZFS dataset name and backup_snapshot_name with a descriptive name for your snapshot. 

Send the snapshot

(to a remote server via ssh from the Source System) Pipe the zfs send output to the zfs recv command on the destination system, typically through SSH.  - For a full (initial) send: The destination dataset does not need to exist beforehand. 

Show Progress

To see the progress of the transmission, you will need to ensure that pv (Pipeline Viewer) is installed on the source machine.

zfs send -R source_pool/dataset@backup_snapshot_name | ssh user@backup_server zfs recv -F backup_pool/backup_dataset
 
# Showing progress as it works
sudo zfs send -R rpool/USERDATA/home_dpwraq@home_20250916 | pv | ssh nexus zfs recv -F zfs10/gearboxsnapshots

This command sends the snapshot stream to the remote server (backup_server) via ssh. 

  • -R sends a replication stream, preserving all properties and descendent datasets.
  • user is the username on the backup server.
  • backup_server is the hostname or IP address of the backup server.
  • backup_pool/backup_dataset is the ZFS dataset on the backup server where the data will be received.
  • -F forces the receive, allowing it to overwrite existing data if needed.

Incremental Backup

After the initial backup, you will want to use a much faster and lower resource intensive option - incremental backups.

For an incremental send

The destination dataset must exist and have the starting snapshot already received. This sends only the changes between two snapshots, which is much more efficient for subsequent backups. 

# On the source server:zfs send -i <source_pool>/<dataset>@<previous_snapshot> <source_pool>/<dataset>@<new_snapshot> | pv | ssh <user>@<remote_host> zfs recv -F <destination_pool>/<dataset>

-i: Sends an incremental stream from the <previous_snapshot> to the <new_snapshot>. 

Example with Compression

For limited bandwidth, you can add compression to the pipeline. 

# On the source server:zfs send -i <previous_snapshot> <new_snapshot> | gzip | ssh <user>@<remote_host> "gunzip | zfs recv -F"
  • gzip: Compresses the data stream on the sending side. 
  • gunzip | zfs recv -F: Decompresses the data on the receiving side before it’s received by zfs recv.

Important Considerations:

  • Encryption: For secure backups, especially over a network, consider using SSH with encryption or encrypting the ZFS pool itself on the backup server.
  • Incremental Backups: For subsequent backups, you can send incremental snapshots using zfs send -i or zfs send -I. This significantly reduces transfer time and storage space.
  • ZFS Versions: Ensure both the source and backup systems are running compatible ZFS versions.
  • Space Requirements: ZFS snapshots can consume significant disk space. Regularly prune old snapshots to manage storage.
  • Tools: Consider using tools like syncoid (part of the zfs-auto-snapshot package) or zfsnap for automating backups, snapshot management, and incremental transfers.
  • Firewall: If using ssh, ensure that the firewall on the backup server allows incoming connections on the SSH port (usually 22).
  • Network Bandwidth: For large datasets, consider using compression or tools like mbuffer to manage network bandwidth during transfers.

Graph View