On Thu, Mar 8, 2012 at 5:48 AM, Bob Doolittle <bob.doolit...@oracle.com> wrote:
> Wait, I'm not following the last few steps you suggest. Comments inline:
> On 03/07/12 17:03, Fajar A. Nugraha wrote:
>> - use the one new disk to create a temporary pool
>> - copy the data ("zfs snapshot -r" + "zfs send -R | zfs receive")
>> - destroy old pool
>> - create a three-disk raidz pool using two disks and a fake device,
>> something like http://www.dev-eth0.de/creating-raidz-with-missing-device/
> Don't I need to copy the data back from the temporary pool to the new raidz
> pool at this point?
Yes, I missed it :)
That's what you get for writing mail at 5 am :P
> I'm not understanding the process beyond this point, can you clarify please?
>> - destroy the temporary pool
> So this leaves the data intact on the disk?
Destroy it after the data is copied back, of course.
>> - replace the fake device with now-free disk
> So this replicates the data on the previously-free disk across the raidz
The fake disk was never written cause it was destroyed soon after
created (see the link), so the pool was degraded. The replace process
tells zfs to use the new disk to make the pool not degraded anymore by
writing the necessary data (e.g. raidz parity, although this might not
be the most accurate way to describe it) to the new disk.
> What's the point of the following export/import steps? Renaming?
> Why can't I
> just give the old pool name to the raidz pool when I create it?
Cause you can't have two pools with the same name. You either need to
rename the old pool first, or rename the new pool afterwards.
zfs-discuss mailing list