Quoth Hans-Joerg on Sun, Aug 26, 2007 at 02:55:32AM -0700: > Right, the disk that crashed had an unmirrored swap partition. So it's > completely clear, that non_persist could becomet corrupted. > svc.configd crashed some seconds after the first hardware error for > the particular disk was reported. So far so good. > > But a way to switch SMF back to consistent state without rebooting the > system is IMHO highly desirable. I mean a single temporary file (which > is a compilation of some xml-files which are placed safely on > a mirrored fs) shouldn't cause such consequences. I think the reason > for requiring a reboot is the limited abilities to "restart" the SMF > system. It seems that it cannot handle this situation very well > (maintenance or uninitialized state for services that are fully > operatice).
I think you misunderstand svc_nonpersist.db. The data in it is not a compilation of the XML files in /var/svc/manifest. It is data which is not necessary after reboot, such as temporary configuration changes and runtime information, such as service states and information about processes. Like I said before, it may be possible to reverse-engineer that information from what the kernel has, but that would require a nontrivial amount of work and probably wouldn't be as fruitful as teaching the kernel to keep the file off of the swap device. David
