Re: RAID1 assembly requires manual "mdadm --run"

[Molle Bestefich <molle.bestefich@gmail.com>]

On 7/8/05, Neil Brown <neilb@cse.unsw.edu.au> wrote:
> On Thursday July 7, molle.bestefich@gmail.com wrote:
> > Mitchell Laks wrote:
> > > However I think that raids should boot as long as they are intact, as a matter
> > > of policy. Otherwise we lose our  ability to rely upon them for remote
> > > servers...
> >
> > It does seem wrong that a RAID 5 starts OK with a disk missing, but a
> > RAID 1 fails.
> >
> > Perhaps MD is unable to tell which disk in the RAID 1 is the freshest
> > and therefore refuses to assemble any RAID 1's with disks missing?
> 
> This doesn't sound right at all.
> 
> "--run" is required to start a degraded array as a way of confirming
> to mdadm that you really have listed all the drives available.
> The normal way of starting an array at boot time is by describing the
> array (usually by UUID) in mdadm.conf and letting mdadm find the
> component devices with "mdadm --assemble --scan".
> 
> This usage does not require --run.
> 
> The only time there is a real reluctance to start a degraded array is
> when it is raid5/6 and it suffered an unclean shutdown.
> A dirty, degraded raid5/6 can have undetectably data corruption, and I
> really want you to be aware of that and not just assume that "because
> it started, the data must be OK".

Sounds very sane.

So a clean RAID1 with a disk missing should start without --run, just
like a clean RAID5 with a disk missing?

Nevermind, I'll try to reproduce it instead of asking too many questions.
And I suck a bit at testing MD with loop devices, so if someone could
pitch in and tell me what I'm doing wrong here, I'd appreciate it very
much (-:

# mknod /dev/md0 b 9 0
# dd if=/dev/zero of=test1 bs=1M count=100
# dd if=/dev/zero of=test2 bs=1M count=100
# dd if=/dev/zero of=test3 bs=1M count=100
# losetup /dev/loop1 test1
# losetup /dev/loop2 test2
# losetup /dev/loop3 test3
# mdadm --create /dev/md0 -l 1 -n 3 /dev/loop1 /dev/loop2 /dev/loop3
mdadm: array /dev/md0 started.

# mdadm --detail --scan > /etc/mdadm.conf
# cat /etc/mdadm.conf
ARRAY /dev/md0 level=raid1 num-devices=3
   UUID=1dcc972f:0b856580:05c66483:e14940d8
   devices=/dev/loop/1,/dev/loop/2,/dev/loop/3

# mdadm --stop /dev/md0
# mdadm --assemble --scan
mdadm: no devices found for /dev/md0

// ^^^^^^^^^^^^^^^^^^^^^^^^  ??? Why?

# mdadm --assemble /dev/md0 /dev/loop1 /dev/loop2 /dev/loop3
mdadm: /dev/md0 has been started with 3 drives.

// So far so good..

# mdadm --stop /dev/md0
# losetup -d /dev/loop3
# mdadm --assemble /dev/md0 /dev/loop1 /dev/loop2 /dev/loop3
mdadm: no RAID superblock on /dev/loop7
mdadm: /dev/loop7 has no superblock - assembly aborted

// ^^^^^^^^^^^^^^^^^^^^^^^^  ??? It aborts :-(...
// Doesn't an inactive loop device seem the same as a missing disk to MD?

# rm -f /dev/loop3
# mdadm --assemble /dev/md0 /dev/loop1 /dev/loop2 /dev/loop3
mdadm: cannot open device /dev/loop7: No such file or directory
mdadm: /dev/loop7 has no superblock - assembly aborted

// ^^^^^^^^^^^^^^^^^^^^^^^^  ??? It aborts, just as above...

Hm!