On Tue, Feb 3, 2009 at 22:46, martin f krafft <madd...@debian.org> wrote: > I found a bit of time to package up mdadm 2.6.7.2-1, which fixes two > RC bugs, but I cannot test it. I've built unofficial packages for > i386 and amd64 and put them at > > http://debian.madduck.net/repo/pool/main/m/mdadm/ > > so please try them out if you can, otherwise I won't be able to > upload them soon, which might delay the lenny release. > > mdadm (2.6.7.2-1~unreleased.1) unstable; urgency=low > > * New upstream release, created for Debian lenny: > - fixes assembly of arrays that are being reshaped (closes: #512475) > - this bug was also responsible for other assembly problems > (closes: #498505, #499643, #496334) > Again, many thanks to Neil Brown for being such an awesome upstream. >
Hi Martin, I've installed your unofficial x86-64 package, and tried to reproduce the failure in #496334. I think I reproduced the circumstances quite well (4 out of 6 devices kicked out from a raid10 array). [*] mdadm --assemble --force no longer segfaults, and reassembles the array, so #496334 is fixed. The reassembled array has the 4 drives as spares though they never were spares, so the array is not functional (would need 3 active, has 2). In the end I had to recreate the array manually to make those "spare" drives active again. I didn't loose any data, but I'm not too keen to try reproducing this bug a third time. Anyway mdadm was not segfaulting, so I could recover the array, unlike with the buggy version that was segfaulting and I couldn't do much with. [*] Here is how I made the array fail, DON'T TRY THIS: # for i in /sys/class/scsi_host/*; do echo min_power >$i/link_power_management_policy; done # sync # echo max_power >/sys/class/scsi_host/host2/link_power_management_policy # sync # dmesg ... see how the drives get kicked out, and all your filesystems unmounted .. reboot .. mdadm --assemble --force /dev/mdX .. array should be back again ... Best regards, --Edwin -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org