Robert Milkowski wrote:
Hello zfs-discuss,
Subject says it all.
I first checked - no IO activity at all to the pool named thumper-2.
So I started replacing one drive with 'zpool replace thumper-2 c7t7d0
c4t1d0'.
Now the question is why am I seeing writes to other disks than c7t7d0?
Are you *sure* that nothing else is going on? Not even atime updates?
Do 'zfs umount -a' and see if there's still writes to other disks.
There may be some small amount of writes to update some metadata with
the resilvering status.
I just did this yesterday on a raidz2 pool and didn't see writes to the
other disks. Maybe the code has changed since you tried?
Also why in case of replacing a disk we do not just copy disk-to-disk?
It would be MUCH faster here. Probably 'coz we're traversing
meta-data? But perhaps it could be done in a clever way so we endup
just copying from one disk to another. Checking parity or checksum
here it's not necessary - scrub is for it. What we want in most cases
is to replace drive as fast as possible.
In some cases it would be faster but others not. For example, if the
pool is not very full, it would be slower. Also if the disk you're
replacing is not available, a straight disk-to-disk copy would not be
possible.
On another thumper I have a failing drive (port resets, etc.) so I
issued over a week ago drive replacement. Well it still hasn't
completed even 4% in a week! The pool config is the same. It's just
waaaay to slow and in a long term risky.
Are you taking snapshots? They cause scrubbing / resilvering to restart
(this is bug 6343667).
--matt
_______________________________________________
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
