[ I'm adding the mdadm maitainer to the CC-list, maybe he has some useful comments. ]
On Mon, 19 Mar 2007 09:58:43 -0600 Luis Rodrigo Gallardo Cruz <[EMAIL PROTECTED]> wrote: > Package: uswsusp > Version: 0.6~cvs20070202-1 > Severity: important > > s2disk does not mark (software) raid arrays as clean when shutting > down the machine. I'm not to familiar with software raid, but this sounds like the way it should be. I guess this is similar to the state a journaling filesystem can be in during suspend; Part of the state is still in memory, part is on disk. Maybe the mdadm people can tell if it is possible to leave a raid array in a `clean state' during resume. I would guess that means stopping the whole array, wouldn't it? That is of course out of the question... > On the next bootup, the mdadm support in the initrd starts up the arrays > and, finding them dirty, starts a background reconstruction. > When, afterwards, uswsusp tries to resume, after properly finding and > loading the resume image, I think this is the problem, we should start resume, before the raid array is up. ( This would probably mean you can't have your swap on a raid array ... not sure ) Anyway, this will get us in an ordering problem: we do want to allow things like swap on lvm or crypt-partitions, this means I want to start resume after they have been setup, e.g. after local-top. But then also the mdadm hook has been run. A solution would be to have the raid 'discovery' phase in local-top so we know what partitions are on it, but split of the start of the reconstruction phase to anther hook in local-bottom, just like mounting filesystems is done after premount. Because reconstruction is done in the background, I presume it can wait for a little while before it is started. What do the mdadm maintainers think? Is this doable? grts Tim
signature.asc
Description: PGP signature
