RAID1 /dev/md4 lost one disk partion /dev/sdb4 after reboot, WHY??

[Bruno Wolff III]

On Mon, Mar 30, 2009 at 10:24:24 +1100,
  L <yuanlux at gmail.com> wrote:
> 
> Bruno,
> below is the output from dmesg. as you see, when generation /dev/md4,
> partition /dev/sdb4 as undetected at all. I did  curious tests (--fail
> --remove; reboot; then --add; reboot) on
> 
> /dev/sdb3 on /dev/md3
> /dev/sdb2 on /dev/md2
> /dev/sdb1 on /dev/md1
> 
> They were all detected and bind to RAID1.
> Of the 4 partitions, I re-formatted partition /dev/sdb4 on Fedora 10
> to ext3. The disk is on small NAS system. Hope this can stimulate some
> hints from you, for me.

While it probably doesn't have anything to do with your problem, you don't
want to reformat array elements. That should be done on the array itself.
In some modes when you add a failed array back in, it detects that it was
recently removed from the array and some shortcuts are taking with
resyncing. That could mess things up.
If you really wanted to clean out the array in the event some data on the
partition is causing a problem, copy from /dev/zero to /dev/sdb4 using dd.
That might also help if you are getting bad reads from a block. (And make
sure the array element, /dev/sdb4, has been failed out of the array before
starting this.)