On Thu, Mar 20, 2008 at 03:26:11PM +1100, Malcolm Herbert wrote: |I've only recently started working with SVM and there's probably a |'better' way to do the following, but it still shouldn't have failed in |the manner in which it did, imho.
Thanks to Peter J. Cherny and Nathan Kroenert for their prompt pointing out where I'd completely come off the rails ... :) Peter pointed me at some useful forum threads: |http://www.opensolaris.org/jive/forum.jspa?forumID=16 |http://www.opensolaris.org/jive/thread.jspa?threadID=40831&tstart=0 | |http://www.gods-inc.de/barbie/solaris/sds_mirroring_howto.txt Nathan broke down the failure mode very succinctly: On Thu, Mar 20, 2008 at 04:08:46PM +1100, Nathan Kroenert wrote: |Consider that if SVM is not told that a disk is gone, it assumes that |it's there... | |You likely booted from the good disk, then as soon as it tried to mount |read-write, which includes a mount -o remount of the metadevice, it read |and possibly wrote garbage immediately based on the bogus crap on the |second . (to things like the message log etc.). | |Doing what you did without a metadetach / metaclear operation is just |asking for much pain. | |I'm not the least bit surprised this happened. :) | |Patient - "Doctor - It hurts when I poke the knife in here..." |Doctor - "Then don't poke there..." | |You should have been able to just split the mirror... |even a metaoffline would have worked, I think, or perhaps a metadetach... In this case the problem was that I'd booted into an environment where I didn't have functioning meta* commands and was interested in seeing what I could do in such a situation and whether SVM would bite me ... It did, but certainly was not unprovoked ... :) SVM certainly seems to cope with devices simply being non-contactable, so in future I'll be a little more circumspect or arrange to break the mirror using admin tools first ... Regards, Malcolm -- Malcolm Herbert This brain intentionally mjch at mjch.net left blank
