Re: How to recreate a dmraid RAID array with mdadm

[Neil Brown <neilb@suse.de>]

I see the problem now.  And John Robinson was nearly there.

The problem is that after assembling the container /dev/md/imsm,
mdadm needs to assemble the RAID1, but doesn't find the
container /dev/md/imsm to assemble it from.
That is because of the
  DEVICE partitions
line.
A container is not a partition - it does not appear in /proc/partitions.
You need

  DEVICE partitions containers

which is the default if you don't have a DEVICE line (and I didn't have a
device line in my testing).

I think all the "wrong uuid" messages were because the device was busy (and
so it didn't read a uuid), probably because you didn't "mdadm -Ss" first.

So just remove the "DEVICE partitions" line, or add " containers" to it, and 
all should be happy.

NeilBrown



On Mon, 22 Nov 2010 13:07:10 -0500
Mike Viau <viaum@sheridanc.on.ca> wrote:

> 
> > On Thu, 18 Nov 2010 16:38:49 +1100 <neilb@suse.de> wrote:
> > > > On Thu, 18 Nov 2010 14:17:18 +1100  wrote:
> > > > >
> > > > > > On Thu, 18 Nov 2010 13:32:47 +1100 wrote:
> > > > > > > ./mdadm -Ss
> > > > > > >
> > > > > > > mdadm: stopped /dev/md127
> > > > > > >
> > > > > > >
> > > > > > > ./mdadm -Asvvv
> > > > > > >
> > > > > > > mdadm: looking for devices for further assembly
> > > > > > > mdadm: no RAID superblock on /dev/dm-3
> > > > > > > mdadm: /dev/dm-3 has wrong uuid.
> > > > > > > want UUID-084b969a:0808f5b8:6c784fb7:62659383
> > > > > > > Segmentation fault
> > > > > >
> > > > > > Try this patch instead please.
> > > > >
> > > > > Applied new patch and got:
> > > > >
> > > > > ./mdadm -Ss
> > > > >
> > > > > mdadm: stopped /dev/md127
> > > > >
> > > > >
> > > > > ./mdadm -Asvvv
> > > > > mdadm: looking for devices for further assembly
> > > > > mdadm: no RAID superblock on /dev/dm-3
> > > > > mdadm: /dev/dm-3 has wrong uuid.
> > > > > want UUID-084b969a:0808f5b8:6c784fb7:62659383
> > > > > tst=0x10dd010 sb=(nil)
> > > > > Segmentation fault
> > > >
> > > > Sorry... I guess I should have tested it myself..
> > > >
> > > > The
> > > > if (tst) {
> > > >
> > > > Should be
> > > >
> > > > if (tst && content) {
> > > >
> > >
> > > Apply update and got:
> > >
> > > mdadm: /dev/sdb is identified as a member of /dev/md/imsm0, slot -1.
> > > mdadm: /dev/sda is identified as a member of /dev/md/imsm0, slot -1.
> > > mdadm: added /dev/sda to /dev/md/imsm0 as -1
> > > mdadm: added /dev/sdb to /dev/md/imsm0 as -1
> > > mdadm: Container /dev/md/imsm0 has been assembled with 2 drives
> > > mdadm: looking for devices for /dev/md/OneTB-RAID1-PV
> >
> > So just to clarify.
> >
> > With the Debian mdadm, which is 3.1.4, if you
> >
> > mdadm -Ss
> > mdadm -Asvv
> >
> > it says (among other things) that /dev/sda has wrong uuid.
> > and doesn't start the array.
> 
> Actually both compiled and Debian do not start the array. Or atleast create the /dev/md/OneTB-RAID1-PV device when running mdadm -I /dev/md/imsm0 does.
> 
> You are right about seeing a message on /dev/sda about having a wrong uuid somewhere though.  I went back to take a look at my output from the Debian mailing list to see that the mdadm did change slightly from this thread has begun.
> 
> The old output was copied verbatim on http://lists.debian.org/debian-user/2010/11/msg01234.html and says (among other things) that /dev/sda has wrong uuid.
> 
> The /dev/sd[ab] has wrong uuid messages are missing from the mdadm -Asvv output but....
> 
> ./mdadm -Ivv /dev/md/imsm0 
> mdadm: UUID differs from /dev/md/OneTB-RAID1-PV.
> mdadm: match found for member 0
> mdadm: Started /dev/md/OneTB-RAID1-PV with 2 devices
> 
> 
> I still have this UUID message when still using the mdadm -I command.
> 
> 
> I'll attach the output of both the mdadm commands above as they run now on the system, but I noticed, but also that in the same thread link above, with the old output I was inqurying as to both /dev/sda and /dev/sdb (the drives which make up the raid1 array) do not appear to recognized as having a valid container when one is required.
> 
> What is take on GeraldCC (gcsgcatling@bigpond.com) assistance about /dev/sd[ab] containing a 8e (for LVM) partition type, rather than the fd type to denote raid autodetect. If this was the magical fix (which I am not saying it can’t be) why is mdadm -I /dev/md/imsm0 able to bring up the array for use as an physical volume for LVM?
> 
> 
> 
> >
> > But with the mdadm you compiled yourself, which is also 3.1.4,
> > if you
> >
> > mdadm -Ss
> > mdadm -Asvv
> >
> > then it doesn't give that message, and it works.
> 
> Again, actually both compiled and Debian do not start the array. Or atleast
> create the /dev/md/OneTB-RAID1-PV device when running mdadm -I
> /dev/md/imsm0 does.
> 
> >
> > That is very strange. It seems that the Debian mdadm is broken somehow, but
> > I'm fairly sure Debian hardly changes anything - they are *very* good at
> > getting their changes upstream first.
> >
> > I don't suppose you have an /etc/mdadm.conf as well as /etc/mdadm/mdadm.conf
> > do you? If you did and the two were different, the Debian's mdadm would
> > behave a bit differently to upstream (they prefer different config files) but
> > I very much doubt that is the problem.
> >
> 
> There is no /etc/mdadm.conf on the filesystem only /etc/mdadm/mdadm.conf
> 
> 
> > But I guess if the self-compiled one works (even when you take the patch
> > out), then just
> > make install
> 
> I wish this was the case...
> 
> >
> > and be happy.
> >
> > NeilBrown
> >
> >
> > >
> > >
> > > Full output at: http://paste.debian.net/100103/
> > > expires:
> > >
> > > 2010-11-21 06:07:30
> 
> Thanks
> 
> -M
>  		 	   		  
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html