Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Fri, 24 Apr 2009, Kristinn Soffanias Runarsson wrote:
> On Fri, Apr 24, 2009 at 10:39 PM, Tom Brown <tbrown at baremetal.com> wrote:
>
>> On Fri, 24 Apr 2009, Kristinn Soffanias Runarsson wrote:
>>
>> Yeah I could do pvmove, that is what I will be using on machines that are
>>> using LVM. But most of them are just using plain devices. And LVM
>>> doesn't
>>> support non-destructive import of a device into LVM (but Solaris SVM does
>>> :)
>>> and I've had good results with that :)
>>>
>>> so drbd out of the question? :)
>>>
>>
>> I have no idea. I don't know how you're consuming the storage, whether you
>> can tolerate a reboot or two, whether this is the root filesystem of a
>> device that is literally booting off it (as compared to a virtual machine
>> using it for /).
>>
>> MD _might_ be able to help. You can build a mirror of an existing device,
>> but you need to be able to tell your storage "consumer" to switch from using
>> /dev/sda1 to /dev/md0 ... same as you would for drbd. The only real
>> difference between using drbd and md is that drbd wants to sync to a remote
>> server. MD wants to sync two local devices... but one or both of those could
>> be network attached storage. lookup mdadm (the build option) if that seems
>> interesting.
>>
>> That said, I've neither done this, nor do I have time to test it.
>>
>> -Tom
>>
>
> Yeah I can survive a few reboots, but not lengthy outages. As far as MD is
> concerned.. it cannot consume a device that already contains data without
> destroying it.
Bull.
Here's a test. Build yourself a pair of block devices. Format one. This
will represent your existing physical device. Format it, put data on it,
etc. Here I've created two logical values /dev/vg1/part1 and part2
Now setup the raid1 in degraded mode
mdadm --build /dev/md1 -l raid1 --raid-devices=2 /dev/vg1/part1 missing
/dev/md1 _is_ /dev/vg1/part1 ... which happens to be the device I setup...
e.g. formatted put a filesystem on it, and filled it up. But you can check
at the low level if you wish:
[root at tom2 mnt]# md5sum /dev/md1 /dev/vg1/part1
47c1572e9214fd60f95a450c53d7afdd /dev/md1
47c1572e9214fd60f95a450c53d7afdd /dev/vg1/part1
from here you'd just add whatever you wanted to mirror to.
root at tom2 mnt]# mdadm --manage /dev/md1 --add /dev/vg1/part2
mdadm: added /dev/vg1/part2
[root at tom2 mnt]# cat /proc/mdstat
Personalities : [raid0] [raid1]
md1 : active raid1 dm-7[2] dm-6[0]
1048576 blocks super non-persistent [2/1] [U_]
[==>..................] recovery = 10.0% (106176/1048576) finish=0.7min speed=21235K/sec
after a while it'll be fully mirrored.
[root at tom2 mnt]# cat /proc/mdstat
Personalities : [raid0] [raid1]
md1 : active raid1 dm-7[2] dm-6[0]
1048576 blocks super non-persistent [2/2] [UU]
3f23ad4b69d0a552042d1ed0f4399857 /usr/src/linux-2.6.24.tar.bz2
3f23ad4b69d0a552042d1ed0f4399857 /mnt/linux-2.6.24.tar.bz2
64921b5ff5cdadbccfcd3820f03be7d8 /usr/src/linux-2.6.29.tar.bz2
64921b5ff5cdadbccfcd3820f03be7d8 /mnt/linux-2.6.29.tar.bz2
[root at tom2 /]# mdadm --stop /dev/md1
mdadm: stopped /dev/md1
fini.
But it is your data. DO as you wish. I just wish you wouldn't ask
for advice and then tell the people that provide advice that they
don't know what they are talking about.
hhmm, seems to me I can probably do a better job of proving the
not "destroying it" point if I avoid running the proactive
filesystem checks which modify the md5sums
# ugh, this is slow...
[root at tom2 /]# dd if=/dev/urandom of=/dev/vg1/part1
dd: writing to `/dev/vg1/part1': No space left on device
2097153+0 records in
#snip
[root at tom2 /]# mke2fs -j /dev/vg1/part1
[root at tom2 /]# fsck -f /dev/vg1/part1
[root at tom2 /]# md5sum /dev/vg1/part1
30c3e250f68b33ec8f2d32d9c90bfc78 /dev/vg1/part1
root at tom2 /]# mdadm --build /dev/md1 -l raid1 --raid-devices=2
/dev/vg1/part1 missing
mdadm: array /dev/md1 built and started.
[root at tom2 /]# mdadm --manage /dev/md1 --add /dev/vg1/part2
[root at tom2 /]# # wait for it to sync
[root at tom2 /]# cat /proc/mdstat
Personalities : [raid0] [raid1]
md1 : active raid1 dm-7[2] dm-6[0]
1048576 blocks super non-persistent [2/2] [UU]
md0 : active raid1 sdb2[0] sda2[1]
5124608 blocks [2/2] [UU]
[root at tom2 /]# md5sum /dev/vg1/part?
30c3e250f68b33ec8f2d32d9c90bfc78 /dev/vg1/part1
30c3e250f68b33ec8f2d32d9c90bfc78 /dev/vg1/part2
So, is that sufficient proof that this statement is simply wrong?
> As far as MD is
> concerned.. it cannot consume a device that already contains data without
> destroying it.
?
-Tom