My original question is at the bottom of this message. I was looking for
a way to minimize downtime of our Oracle databases when we perform a cold
backup. My thought was to shutdown the databases, break the existing
mirrors,
remount the filesystems, startup the databases, and then backup the piece
of the mirrors that was broken off (and then resyncing that piece).
I heard almost immediately from alan_at_nabeth.cxo.dec.com and had two other
replies from Steve Hancock [hancock_at_decatl.alf.dec.com], and from
North, Walter [Wnorth_at_state.mt.us].
Steve wrote:
The procedure looks ok, except that the mount of the mirrored plex
will
not work as you have stated. The mount will fail, because you cannot
have two active domains with the same domain ID. The way around this
restriction, in V4.0D and higher, is to use the "-o dual" option to
the mount command. If you use this, you will see your procedure
work.
I tried this and it works! Woohoo! I don't know if we'll do this or
not, but I now know it's an option.
Alan suggested using cloning and then referred me to the man pages for
mount (lucky for me Steve gave me the option b/4 I had a chance to look).
I'm not sure cloning will work for us, as some of the filesystems are
higher than 95% used. Alan does raise some good points so I'm including
all 3 of his notes here:
Alan 1: Have you considered using AdvFS clones for the purpose? The
clonefset command creates a read-only snapshot of the fileset
metadata. Then as file pages (blocks) are changed, the
original is written to the clone so the original is kept
as a snapshot. This method saves you the trouble breaking
the mirrors (and the associated risk of not having mirrored
data) and the copy time to put the mirrors back together.
It has the same problem of making the file system quiet
enough to get a good backup, but that depends as much on
the data as the file system; low risk for a typical time-
sharing load off hours, higher risk for a 24 hour per
day database.
And, the domain is used for space for the clone. If the
domain is fairly full and there are lots of changes during
the lifetime of the clone it could run out of space. Finally
removing the clone can take a long time in some cases.
Alan 2:
Ah, you already knew about clones. Sort of. I hadn't
read all the message.
AdvFS does have domain unique information on its volumes,
but there is an option to mount or one of the other
domain related commands to have it ignore this piece
when you mount it. Check the mount manual page first.
The advantage of using AdvFS's cloneing ability is that
you don't lose the redundancy of your data during the
backup.
Alan3:
Estimate how much time it will take backup the data and
resync the mirrors. That's how long you won't have
mirrored data. Get someone in your finance department
to calculate how much that risk is worth. Disks are cheap
even when you need to buy twice as many to have a spare
volume to have enough free space for the clone.
Walt also pointed out that I missed a 'volume stop' command.
Original message:
8200, 4.0d PK2, ASE, LSM
I've got several advanced filesystem domains on mirrored LSM volumes
that I'd like to backup by breaking the mirrors, doing a vdump,
and then reattaching the removed plex for resyncing. Something along
the lines of:
umount /filesystem
volplex dis plex2
mount /filesystem
volmake -Ufsgen vol backup plex2
volume start backup
Now I'd need to mount the quiescent filesystem (if that's possible) and
do a vdump, so I think I need to:
mkdir /etc/fdmns/newdomain
ln -s /dev/vol/rootdg/backup /etc/fdmns/newdomain/backup
mount newdomain#oldfilesetname /mnt
If that works, then I'd issue a vdump command followed by:
volplex -f dis plex2
[should have a 'volume stop backup' command here that was missing from the
original note]
voledit rm backup
volplex att original_volume plex2
Does anyone know if this will work? DEC recommends using cloning for
AdvFS, but with a mirrored volume, the clone already exists. I'm not
sure what sorts of advfs metadata stuff exists on the volumes, whether
I can just create a new domain as suggested above and have a successful
mount. I think someone asked this question last month but I didn't
see a summary. TIA
Jeff Beck
jbeck_at_carewiseinc.com
206.749.1878
CareWise Inc.
701 5th Ave.
Suite 2600
Seattle, WA 98104-7015
Received on Tue Sep 21 1999 - 22:35:50 NZST