Moving a ZFS filesystem from one pool to another
Solution 1:
I'm repeating much of what jlliagre said, but with additions for descendent file systems. (Mostly so I have a reference when I forget.)
If you have sub-filesystems you'll want to use the -r
flag on the zfs snapshot command, and the -r
or -R
flag on the zfs send
command. The uppercase -R
moves all properties, snapshots and clones.
To move an entire pool:
zfs snapshot -r sourcepool@moving
zfs send -R sourcepool@moving | zfs receive -F destpool
then to sync the changes for a shorter migration period shut down applications, shutdown samba, nfs
zfs snapshot -r sourcepool@moving2
zfs send -Ri sourcepool@moving sourcepool@moving2 | zfs receive -F destpool
If you are creating multiple intermediary snapshots for the migration, read the man page about the zfs send -I
switch.
Solution 2:
The question about disabling services essentially depends on the services themselves:
- Are they actively writing to the file system you plan to move ?
- Are they storing persistent data you want to keep.
In any case, sending a ZFS file system might take a long time. It is possible to minimize service unavailability by keeping them online most of the time that way:
create a snapshot
send that snapshot the way previously suggested but while keeping all services active
when the fs is received on the other pool, disable the critical services bounds to that file system. Make sure the new file system on the destination pool is not modified as changes will be discarded later anyway.
create a second snapshot (e.g. snapshot2)
send that second snapshot incrementally, that will be much quicker that the previous transfer. e.g.:
zfs send -i rpool/filesystem@snapshot rpool/filesystem@snapshot2 \ | zfs receive -F destinationpool/filesystem
when done, move the filesystem mount point from the old dataset to the new one. e.g.:
zfs set mountpoint=/application/directory.old rpool/filesystem zfs set mountpoint=/application/directory destination/filesystem
You need to make sure no process is bound to
/application/directory
(e.g.: accessing files or having it as its current directory) to achieve that.re-enable the service(s) and you are done.
Solution 3:
I don't think this can be performed online, but my process would be to shut the necessary services down, take a snapshot and use zfs send/receive to make the migration. Something like:
zfs send rpool/filesystem@snapshot | zfs receive destinationpool/filesystem
Edit:
If you install the "pipe view" command (pv), you can monitor status of the transfer with:
zfs send rpool/filesystem@snapshot | pv | zfs receive destinationpool/filesystem
Solution 4:
You'll also find some benefit in using a tool like "mbuffer" in the pipeline; apparently ZFS send/recv performance improves dramatically if both ends can mostly stream data continuously, and without mbuffer (or something similar) you get a ping-pong effect where one continually blocks on the other.