I haven't looked _REAL_ close yet and actually setup powerd, but it
appeared to me that if you do a shutdown in response to power going
out and then the power comes back before the battery on the UPS
actually goes dead your system will not come back up on its own.
If instead you bring the system down in a reliable manner by shutting
down processes and then umounting or remounting R/O the file systems
then powerd can still monitor for returned power until the computer
actually looses power.
I do see a general problem of being unable to robustly umount or
remount R/O filesystems.
I don't yet know how much it will help, but I took someones advice
and backgrounded the 'umount -a -t nfs' call in my rc scripts.
Ty
>
> >More generally, whenever one partition cannot be unmounted, all of the
> other
> >partitions should be unmounted anyway. I used to have a problem with an
> >unreliable disk and/or scsi controller that would force me to fsck every
> >partition on every disk just because trying to shut down one partition got
> >stuck. This was the old buslogic driver, not the new one, but I think the
> >problem is not specific to a single scsi driver.
>
> I am wondering whether umount should be made more robust so that
> it essentially forks prior to each unmount attempt. When all the children
> have died we know that everything is unmounted, and we can proceed. If
> one of the umounts hangs, then at least all of the remaining partitions
> are cleanly unmounted.
>
> If some NFS partition cannot be unmounted, can we still force the
> root partition back to R/O status so that it is also marked clean?
>
> -Eric