Re: Removing an accidentally incorrect vdev from a ZFS pool

On 7/12/07, Steven Schlansker <stevenschlansker_at_berkeley.edu> wrote:
> Hello everyone,
> I've been using ZFS on -CURRENT for a few weeks now with quite decent
> results.  I'm really glad that this apparently awesome filesystem is
> available, and want to register my thanks for everyone who put work into
> it :)
>
> That being said, I've just run into a little snag.  I wanted to extend
> my zpool from 3 drives to 6 drives.
>
> My first hope was that I could automagically extend the RAIDZ.  My hope
> was to extend the vdev to be a raidz1 with 6 devices, and then restripe
> it to raidz2.  That would have been pretty neat, and a great way to
> manage drives.  After reading the man pages and not finding that
> functionality, I found many mailing list posts with everyone chatting
> about exactly that.  I hope the functionality finds its way into 7.0.
> I'd add it myself, but I tried to read the ZFS code and a good
> percentage off it flew over my non-filesystem-geek programmer's head :)
>
> So then I settled on the next-best thing - adding another raidz1 vdev
> with the three new drives.  Not quite as flexible, but still adequate.
>
>    I checked the dmesg to find my drives - ad12, ad16, ad19.  I ran
> sudo zpool add universe raidz1 ad12 ad16 ad19, and then checked zfs list
> to find my brand new double-sized zfs.... huh?  It only grew by 40G?
> That's strange.  Maybe the zpool grew... nope.  About here is where I
> started to panic a bit.  Checked the dmesg... oops!  it was ad18, not 19...
>
> Anyway, my questions now are:
> Am I correct in concluding there is no way to reshape a raidz1 to a
> raidz2?  Is this functionality planned?
>
> I now need to remove this broken vdev from my array.  I haven't added
> any data, so there shouldn't even be any data at all on it.  However all
> the remove/delete options to zpool seem to exclusively work on mirrors
> and hot spares.  I really need to get this vdev off the zfs - it's
> entirely useless.  How can I do that?  I've already taken out the
> accidental drive - I want to try to recover the old filesystem off of
> it.  Lucky it wasn't too important, though it'd be nice to have.  Now i
> have an array stuck permanently degraded.
>
> On a related note, perhaps zpool should do a bit of sanity checking.  I
> know the Linux md tools require you to 'force' array creation if the
> drives differ by +/- 5% or thereabouts.  I just created a 400G/400G/40G
> raidz, which is a totally stupid array layout.  Maybe it should try to
> catch it.  (Yes, I should have caught it, but it's easy to miss the
> extra digit when you see 4xxxxx and 4xxxx next to each other, especially
> when they don't line up)
>
> Thank you so much for any advice anyone can offer!
>
> Steven Schlansker

There is still no way to expand a raidz. The way ZFS was created makes
this a difficult task. Still the ZFS developers say they are working
on that.
Removing a device from a raidz is not supported as well. Again it's
being worked on.
I can't say if for sure about raidz to raidz2 being supported, but I
guess it isn't.
Everything else I pretty much learned a couple of months ago in Sun's
online docs and forum which I recommend you do browse.
As for new features hitting 7.0 I wouldn't get my hopes up, new
features come from Sun, not FreeBSD developers. You might see some
small updates and/or fixes for 7.0 but that's it. My words, don't take
them for granted :>

-- 
Joao Barros