Re: swap-on-raidframe vs raidctl -P

[Greg Oster <oster%cs.usask.ca@localhost>]

Manuel Bouyer writes:
> On Tue, Jan 22, 2008 at 04:45:34PM -0500, Christos Zoulas wrote:
> > On Jan 22, 10:40pm, bouyer%antioche.eu.org@localhost (Manuel Bouyer) wrote:
> > -- Subject: Re: swap-on-raidframe vs raidctl -P
> > 
> > | No, but forcibly closing the raid device could solve other problems too,
> > | like (I suspect, I didn't try it) cgd on raid.
> > 
> > Not while processes are swapping on it? Are you planning to kill them?
> > Also swap is usually on top [who mounts things on top of swap?], so
> > removing swap before unmounting should work.
> 
> I'm not sure the cgd device gets unconfigured on shutdown, even if it's not
> in use any more. and I'm sure we could find other examples. If we had a way
> to say "there won't be any more I/O anyway so cleanup any raid devices
> (and others that need cleanup) and close them", it'd solve them all.

In the case of "marking a RAID set as clean", there will be I/O to an 
underlying component, and that underlying component could be a RAID set.

All that's needed for RAIDframe right now is for raidclose() to get 
called... but if there is a way to unscramble this mess with some 
sort of alternate solution (e.g. shutdownhooks or whatever) I'm happy
to look at that solution as well..

I think something like this has been proposed before:

 while(!done) {
     foreach mounted filesystem
       - attempt to unmount each
     foreach configured device
       - attempt to unconfigure each unused (unopened) device
     if (no mounted filesystems and no configured drives)
       - we're done
 }

but it's probably not enough to deal with really 'loopy' cases either..

Later...

Greg Oster