in-situ conversion to FT disk config?
Michael ODonnell
michael.odonnell at comcast.net
Thu Nov 17 19:17:01 EST 2005
How I ended up creating my RAID1:
Recall that hda1 and hdc1 are identically sized partitions
on identical disks. hda1 was full of data that I wanted
made redundant, hdc1 was empty. I first said:
mdadm -v -C /dev/md0 -l 1 --raid-devices=2 /dev/hdc1 missing
...which created the RAID1 device, named it md0, declared
it to consist of two "real" devices, mentioning one of them
(hdc1) while indicating that it was OK for the other one
not to be present at the moment (a "degraded" array).
mdadm said this:
mdadm: /dev/hdc1 appears to contain a reiserfs file system
size = 120053696K
mdadm: size set to 120053632K
...and then /dev/md0 was immediately available for use.
Interestingly, the entire reiserfs filesystem inside the
RAID appeared to be mountable and usable, but fsck.resierfs
was unhappy to discover that it could no longer see as far
into that device as its superblock indicated it should.
That's tantalizing; this indicates that if I had had the
presence of mind (and courage) to use something like "parted"
I could have done this with hda1 instead of hdc1, resized
that reiserfs filesystem inside that now-slightly-smaller
partition, shrinking it to fit and my original fantasy about
an in-situ RAIDification probably could have been realized. Dang...
The real story isn't nearly so thrilling, though - I ended
up just rebuilding a reiserfs inside that new device (md0)
and then mounted it and copied all the data over from hda1.
After I had convinced myself all data had been copied intact,
I threw The Big Switch by saying:
mdadm -v -a /dev/md0 /dev/hda1
...to which mdadm responded:
mdadm: hot added /dev/hda1
...and the md0_raid1 resync thread is now busily cloning hdc1
onto hda1, while md0 remains available and mounted. It would
appear that all I have left to do is to use fdisk to change
the partition-type cookies to 0xfd indicating "Linux raid
auto" so they'll supposedly be noticed and activated at bootup.
More information about the gnhlug-discuss
mailing list