The problem seems to be that the SCSI/libata devices do not seem to be there at just the right moment, and I cannot get udevtrigger to ignore the "block" class devices (ie LVM, evms etc), so they get either assembled degraded, or not assembled at all.
I went in at premount one day, did the udev initialization part, and my RAID10s with 4 drives had only 1 active drive. I just decided to mdadm -S it, and enable it with mdadm manually, and it worked. No "failed" drives, everything just assembled correctly. Much better than usually having to re-add the drives. I propose the following workaround for now: Depending on whether udevtrigger returns immediately or waits till events are processed, afterwards (whatever that means in relation the the preceding statement), we should stop all arrays, and let the mdadm script from the initrams assemble the arrays manually. Hopefully I can test something soon and post a diff. ** Tags added: initramfs udev ** Bug watch added: Debian Bug tracker #403136 http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=403136 ** Also affects: udev (upstream) via http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=403136 Importance: Unknown Status: Unknown ** Changed in: initramfs-tools (Ubuntu) Sourcepackagename: initramfs-tools => udev -- kernel/initramfs >2.6.20-6 fails to init RAID/LVM root https://launchpad.net/bugs/86384 -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
