Hi,
We have Software RAID1 (MD) mirrors on top of LVM volumes so that we
can grow the LVM volumes if we need one of the mirrors to grow and
then use mdadm --grow to grow the mirror followed by using xfs_growfs
to grow the xfs file system mounted on the mirror. Thus we can extend
the size of the mirrored file system without having to unmount it
which is great. This is what should happen in an ideal world anyway.
In reality it does not work. )-: This is perhaps because mdadm does
not realize the underlying devices have grown (using lvextend) and
therefore running:
mdadm --grow /dev/md3 --size max
Does not do anything.
And if I do:
mdadm /dev/md3 --fail /dev/vg/volume2
mdadm /dev/md3 --remove /dev/vg/volume2
mdadm /dev/md3 --add --write-mostly /dev/vg/volume2
This fails because MD complains that no superblocks are found on the
array or something like that. )-: Perhaps it has noticed that the
devices are bigger and now cannot find the MD super block as it is no
longer at the end of the device or something like that, not sure.
The only working way I have found to do what we want is to --fail & --
remove one device, then grow it with lvextend, then add it back to the
mirror. BUT this causes a full resync because it does not detect that
the device was part of the mirror before and no resync is thus
needed. Again, I assume because the MD superblock is now in the
middle of the volume instead of at the end.
I then have to wait for the resync to finish and then I have to remove
the other mirrored device with --fail & --remove, then lvextend that,
and then I can add it back to the mirror which AGAIN triggers a full
resync as before.
I then have to wait for the resync to finish and then I finally can
grow the mirror using:
mdadm --grow /dev/md3 --size max
And it works this time.
As a small shortcut, I can do the mdadm --grow after the first resync
is complete (see above), after breaking the mirror for the second
time, then the md3 device is grown with just one member device. Then
I add the extended other mirror target to the array with --add and
that then does a full resync on the already grown mirror.
So to grow a raid mirror takes ages because of the resync times when
you have large arrays!
Is there a better way to do this? I am hoping someone will tell me to
use option blah to utility foo that will do this for me without having
to break the mirror twice and resync each time. (-;
If not, please consider this a feature request for mdadm. (-: It
should have an option to detect that the underlying device has grown
and thus write a new superblock (or move the old one or whatever) at
the end of the newly grown device instead of complaining that it does
not exist. Or something! As it is, it is incredibly time consuming
and inefficient. )-:
Best regards,
Anton
--
Anton Altaparmakov <aia21 at cam.ac.uk> (replace at with @)
Unix Support, Computing Service, University of Cambridge, CB2 3QH, UK
Linux NTFS maintainer, http://www.linux-ntfs.org/
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
