On 7/10/13 1:56 AM, "Ole Tange" [off-list ref] wrote:

I tested reshaping a raid0 on 2 devices to 4 devices.

It seems the reshape first converted to RAID4 and then quickly
converted to RAID0.

I have now done that on a bigger array. The only change that I am aware
of is:

* The 2+2 devices are much larger (25 TB each compared to 1 GB each)
* The system has crashed during the reshape

So right now the system looks like this:

Personalities : [raid6] [raid5] [raid4]
md3 : active raid4 md1[0] md5[3] md4[4] md2[1]
     109396484096 blocks super 1.2 level 4, 512k chunk, algorithm 0
[5/4] [UUUU_]

which looks like the RAID4 just before the final step.

I then tried:

# mdadm --grow /dev/md3 -n 4 -l 0 --backup-file reshape.bak

But that seems to cause the reshape to go through the full 100 TB again:

root@lemaitre:/lemaitre-internal# cat /proc/mdstat
Personalities : [raid6] [raid5] [raid4]
md3 : active raid4 dm-0[0] dm-3[3] dm-2[4] dm-1[1]
     109396484096 blocks super 1.2 level 4, 512k chunk, algorithm 0
[5/4] [UUUU_]
     [>....................]  reshape =  0.0% (28100/27349121024)
finish=32428.7min speed=14050K/sec

So I cancelled that and rolled back to the situation before (this was
possible because I ran this on overlay files):

Personalities : [raid6] [raid5] [raid4]
md3 : active raid4 md1[0] md5[3] md4[4] md2[1]
     109396484096 blocks super 1.2 level 4, 512k chunk, algorithm 0
[5/4] [UUUU_]

Can I convert that to RAID0? Can I do that without having to wait the
2-3 weeks a full reshape takes?

All you need to do to directly convert to RAID0 is:

echo 0 > /sys/block/md3/md/level

Sam