zfs i/o error - all block copies unavailable

[Eugene M. Zheganin]

Hi.

I have a server, it was running 8.2-STABLE/i386 with zfs v28. All of a 
sudden, on last reboot I got a big bunch of "zfs i/o error - all block 
copies unavailable" messages and server was unable to boot. I decided 
that one disk is dying, detached it, and booted successfully. On a next 
boot I got this again. I booted from CD, replaced a zpool.cache, and 
booted successfully once again. On next reboot I got it again, and was 
unable to fix it. However (as you may already know) disks were fine, and 
all the data wasn't corrupted. Iread a couple of mailing list posts 
about this mentioning that this could be an i386 issue, and decided to 
deal with in a radical way: I've installed 10.0-BETA1/amd64 (booted from 
a LiveCD, mounted an NFS share with /usr/src and obj, and did the 
upgrade). Now everything is almost fine, except that I still get this 
message, but only once and it seems to be harmless, as the server can 
still be booted (I experimented and tried this like a dozen times).

So, questions:

- is it really harmless ?
- can I run with this ?
- is there any way to get rid of it ? (probably, without recreating a 
pool, because it holds several TBytes of user data)
- and, finally - what does it mean ?

Thanks.
Eugene.