I cannot say when this regression has been introduced. I rebooted one of my servers (stable, KEYWORDS=amd64) last night (last reboot 23-10-2011 IIRC) and it did not boot anymore (/dev/mapper/root is not a valid root device....) because the two md soft raid did not get initialized anymore. Boith raids were detected as md127 and md126 becaus I had a config error where my /dev/mdadm.conf contained wrong UUIDs for md0 and md1. But: The system always bootet correctly. Not I had to have a closer look, detected the wrong UUIDs, fixed them, and the system was bootable again. So I cannot 100% say if you have to have a wrong mdadm.conf or no file at all to make it NOT boot. Reproducible: Always
To clarify: "md127 and md126" were the device names of the md devices before the reboot, when the system booted correctly with a wrong mdadm.conf with wrong UUIDs for both of my raids mdadm.conf: DEVICE /dev/sd* ARRAY /dev/md0 UUID=13ccb9d4:9569b90f:332c2c25:004bd7b2 ARRAY /dev/md1 UUID=1d22da7f:2b0de180:332c2c25:004bd7b2 MAILADDR root@localhost
I think as soon as genkernel finds a /etc/mdadm.conf, it skips the autodetection. This is by purpose, and is for example handy when you want full control over what arrays genkernel starts. So can someone close this?
(In reply to comment #2) > I think as soon as genkernel finds a /etc/mdadm.conf, it skips the > autodetection. > This is by purpose, and is for example handy when you want full control over > what arrays genkernel starts. Quoting "man 8 genkernel": --mdadm-config=<file> Use <file> as configfile for MDADM. By default the ramdisk will be built without an mdadm.conf and will auto-detect arrays during boot-up. > So can someone close this? Closing. Marc, please re-open if necessary.