Quoth Hans-Joerg on Fri, Aug 24, 2007 at 04:23:59PM -0700: > Do you have any recommendations for me? I must say, that my sympathies > for the Solaris Volume Manager is somewhat limited these days. A hard > disk crash shouldn't arise such a situation.
Without knowing more about what happened in the past, I don't think I can comment on how well SVM performed. And even if I did, I don't know particularly much about SVM anyway. > Anyway there should be > a way to salvage this damn SQLite-file and get SMF in consistent state > without a reboot. Well the svc_nonpersist.db file is kept on a tmpfs filesystem, which is usually kept in memory, but I suppose could be swapped out to disk. So if a disk which swap is kept on fails, then I suppose the file could be corrupted. So we could teach SMF to try to be more resilient to disk failure by mirroring the file to another disk (such as the root filesystem), but I think it would be better for SMF to instruct the kernel that the file should not be swapped out to disk, which should reduce the liklihood of corruption down to the liklihood of memory failure. Or perhaps invent a way to reconstruct the essential information from whatever data the kernel keeps about processes. (Which would require storing the mapping between FMRIs and contract IDs in the kernel, to have any chance of being effective.) David
