[OpenIndiana-discuss] ZfS migration scenario including zvols

[Jim Klimov]

On 2013-04-10 20:52, Sebastian Gabler wrote:
> The rest will temporarily be shifted starting with 'pool1/',  to
> Are there any hints what I should take into account to
> retain all object parameters once pool1 is back?

Beside the use of zfs-send replication mode to propagate dataset
attributes, you might also use "zfs list -o all pool/dataset |
grep local" on both the old and new copy of the dataset to validate
that the attributes you need got carried over.

Conversely, you might want to revise the compression and blocksize
properties (the latter - for datasets with databases, disk images,
etc., be it zvols or filesystems). For example, you might benefit,
at least space-wise, by enabling gzip-9 compression on the target
dataset while it is being sent, and perhaps enabling the compression
level you need for active IO. Old data would be compressed and WORM,
and future new data might be less compressed allowing faster writes
per cpu cycles. YMMV.

You might also want to reorganize some data (split or merge datasets),
in which case rsync might be your better friend than zfs-send.

Do take notice if your datasets used ACLs - older rsyncs might not
transfer them properly (or at all); 3.0.10 is rumoured to do it well.

HTH,
//Jim