Problems with first attempts to use ZFS

[Mike Clarke]

Thinking that it's about time to learn a bit about ZFS I started with 
the tutorial at <http://flux.org.uk/howto/solaris/zfs_tutorial_01>

I'm currently running 8.1-RELEASE i386 with a generic kernel and 3 disk 
slices, each mirrored across 2 drives with gmirror. I appreciate the 
current configuration is far from ideal for ZFS but I could use a file 
based ZFS pool and the reduced performance shouldn't matter for a small 
learning exercise. Though it appears that running a disk based pool on 
a gmirror system was not a good thing.

First I created a couple of file based pools:
   mkdir /disk1/z
   cd /disk1/z
   dd if=/dev/zero of=disk1 count=128 bs=1m
   dd if=/dev/zero of=disk2 count=128 bs=1m
   zpool create trout mirror /disk1/z/disk1 /disk1/z/disk2
   zpool list
   zpool status trout

That looked ok, the ZFS filesystem /trout was created and I was able to 
create data on it with dd if=/dev/zero of=/trout/foo count=32 bs=1m. 
The next step was to test it by overwriting the first disk label with 
random data

   dd if=/dev/random of=/disk1/z/disk1 bs=512 count=1

Then force ZFS to check with the zfs scrub command.

   zpool scrub trout

At this point everything went pear shape, the whole system stopped and I 
finally had to force a reboot with a hard reset.

On rebooting gmirror reported that all 3 slices on one of the drives 
were degraded and rebuilt them.

Subsequent attempts to run zpool commands like list or status just lead 
to kernel panics followed by gmirror rebuilds on rebooting.

How do I clear up this mess either by fixing the existing test ZFS pool 
or just getting rid of it so that "zpool list" returns the "no pools 
available" message instead of crashing the system. And, yes, I do have 
backups so I can restore any files which might be needed.

-- 
Mike Clarke