boot time assembly of raid arrays

[Rick Stevens]

On Wed, 2005-10-19 at 14:27 -0400, Mark J Strawcutter wrote:
> >> rhel4
> >> 
> >> I've tried using combinations of md= and raid=noautodetect kernel parameters
> >> along with settings in /etc/mdadm.conf in an attempt to influence what devices/partitions
> >> get assembled into md3 at boot time.
> >> 
> >> So far, no luck.
> >> 
> >> md3 is root.  The correct level-1 array should be sda3 and sdc3.  However, sdd3
> >> has raid superblock left on it from previous use with preferred minor=3
> >> 
> >> It appears that something starts looking at the last/highest disk's partitions and works
> >> down.  It builds md3 from sdd3 then rejects sdc3 and sda3 as not matching.
> >> 
> >> Tried device and uuid options in md= and mdadm.conf but it doesn't help.
> >> 
> >> Is there something else that influences how raid arrays are assembled at boot time?
> > 
> > A normal SCSI scan starts from device 7 (or 15 on wide SCSI) and works
> > its way down in Linux.
> 
> And with multiple adapters, it appears to start with 1 then goes to 0
>  
> > Have you tried adding a line such as:
> > 
> > ARRAY /dev/md3 devices=/dev/sda3,/dev/sdc3 uuid=whatever
> > 
> > to your mdadm.conf file?
> 
> yes - didn't help
>  
> > How about destroying the RAID superblock on /dev/sdd3 via
> > 
> > mdadm --zero-superblock --force /dev/sdd3
> > 
> > and then mdadm can't do anything with it even if it wanted to (well,
> > unless you tell it to do a legacy MD array, then all bets are off).
> 
> That operation executes without error, and subsequent mdadm query
> says no superblock, but when it's unmounted at shutdown time the
> superblock gets re-written.

Hummm...well, at boot time, mdadm is run as "mdadm -A -s", which means
that it should look at /proc/mdstat first, then /etc/mdadm.conf.  You
might try the "mdadm /dev/md3 -r /dev/sdd3" to make sure /dev/sdd3 is
removed from the RAID.  Also check /proc/mdstat to see if the device is
still shown in there.
----------------------------------------------------------------------
- Rick Stevens, Senior Systems Engineer     rstevens at vitalstream.com -
- VitalStream, Inc.                       http://www.vitalstream.com -
-                                                                    -
-           Denial.  It ain't just a river in Egypt anymore!         -
----------------------------------------------------------------------