[OpenIndiana-discuss] Re2: ZfS migration scenario including zvols

Sebastian Gabler sequoiamobil at gmx.net
Sat Apr 13 22:02:00 UTC 2013


Am 13.04.2013 17:37, schrieb Sebastian Gabler:
> Hi Jim and Edward,
>
> thanks for your comments.
>
> Taking them into account, I decided to apply the following method:
>
> - clean up, that is all the file sets that should remain on the other
> pool in the future have been migrated using zfs send | receive. That
> worked well. Mount points were moved (the only local service using
> them is rsync, config is easy to update), share points were kept.
> - make sure that none of the data sets is currently mounted externally
> (because I can: I have a service window that should be large enough.)
> - iron out all existing snapshots (I can, because I don't have to keep
> them)
> - make a list of the properties of each data set node. (pool.list;
> nfs.list; iscsi.list; smb.list; nfs_filsetxx.list;
> iscsi_filesetxx.list; smb_.filesetxx.list) Each text file contains on
> child exactly. nfs, iscsi, and smb are the root branches.
> - snapshot -r each of the main branches (important)
> - zfs send mainbranch at 1 -R > /pool2/mainbranch.dmp for each nfs,
> iscsi, smb
> - destroy pool1, and apply the new vdev layout under the old name
> - zfs create pool1/mainbranch for each nfs, iscsi, smb
> - cat /pool2/mainbranch.dmp | zfs recv -d pool1/mainbranch
>
> Guess that will work. In case of doubt, I should be able to fix
> deviating file set properties manually.
>
> What I already found out is that there are some niggles with Comstar.
> So, I had to destroy the luns in sbdadm to be able to destroy the
> parent zvols. I dumped the parameters before doing that, so I hope
> that I will be able to get them back without too many complications.
>
> Thanks again for your help.
>
> Sebastian

I am done.
Saliences: I had to renew the view for the zvols, and restart CIFS 
server after the final zfs receive; all received data sets report by 1 
GB less in used size consistently. I am not missing anything, however.
Sending and receiving speed depended on content. Small-block stuff was 
slower.

BR

Sebastian





More information about the OpenIndiana-discuss mailing list