* Michael Tautschnig <m...@debian.org> [Sun Feb 21, 2010 at 09:47:46PM +0100]:
> > First, I want to clarify that this bug is NOT directly fault of fai, it > > is again due to other key packages and changes in fundamental behavior > > that drive me NUTS when it is changed with so little consideration, but > > it seems you may still wish to change things on your end to stay up with > > the curve. > > mdadm 3.1.1(-1?) changes the DEFAULT metadata flag from the prior 0.90, > > to 1.x; which effectively translates to 1.1 as the default/stable 1.x > > metadata format. This chokes up ALL attempts to use these raid devices > > as either boot or root in my testing at least. All of this behavior > > manifests itself when building a standard sid system using fai and the > > current tree, and I fear it will propagate down sooner or later, at > > least to squeeze if it's not already there too. > > Further evidence of this issue and 'reasons' as to why it's not fixed > > can be found in these bug reports: > > http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=569107 > > http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=492897 > > http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=554500 > > Long story short, since you control the mdadm calls in your fai build > > scripts, if you add the option > > -e 0.90 (or long tag --metadata=0.90) > > To the create/partition script when creating the mdadm raid arrays, > > specifically for boot/root partitions, I think we all win and you beat > > this for the long-term until mdadm/bootloader people get their crap > > straight and take the full user base into consideration. Right now, > > netinst is broken using raid1 and I'm having to manually destroy and > > re-create the partitions to boot any new software raid system setups.... > > grrrr. If you care about my stress levels please be sure to modify > > setup-storage as well as the old standard as I use setup-storage. Not > > sure if the old method even supports software raid, I forget =) > I'm not that fond of the idea of enforcing something outdated, it will likely > come back on us at some point. The problem is that once we enforce this, no > one > can easily override it. Assuming that the maintainers of all the involved > packages will do their job sooner or later, I think this is not an issue > forever. > Of course, however, we need a workaround. And that is already there: If you > just > add > mdcreateopts="-e 0.90" > to each of those raid volume definitions you should be fine. Now I understand > that this is some extra per-volume work for you, but it is a workaround only, > so > shouldn't be there forever :-) A full example would look like this: > disk_config raid > raid1 / hda1,sda1 ext3 rw mdcreateopts="-e 0.90" > raid1 swap hda2,sda2 swap sw mdcreateopts="-e 0.90" > Is that acceptable to you? Could you verify that it actually does the trick? AFAICT this issue won't affect squeeze itself anymore since grub in squeeze supports the 1.x metadata format. But: if someone deploys etch or lenny with squeeze or a newer Debian suite (where the 1.x format is the default) it won't work in the given situation. I therefore suggest to document this issue. Do you (especially Michael and Thomas) agree with me? regards, -mika-
signature.asc
Description: Digital signature
