Everhart, Glenn
From:	Glenn C. Everhart [Everhart@gce.com]
Sent:	Wednesday, March 10, 1999 7:44 PM
To:	Info-VAX@Mvb.Saic.Com
Subject:	Re: Backing up a live shadowed disk
It will be necessary for full safety to stop the application for
a moment. Might as well dismount the shadowset as well at that
point. True, you lose a safety measure for a few minutes but
you get a clean backup.

The issue here is that you have, in general, no way to know what
the application considers a single transaction. Suppose it were
writing a file structure, just for illustration. When writing to
such a thing, you need in general to update a storage map of some
kind, an index structure or structures, and data. (The same happens
for database managers.) If your app is inclined to cache indices
in memory, it might update data on disk but not the index on disk
till it reaches whatever it considers a synchronizing point. It might
be like what would happen if writes to a storage bitmap were
deferred, or worse, writes to an index file. This kind of thing
is totally application driven and unless the app is designed to
work no matter when the plug might get pulled, taking a backup
by unshadowing is equivalent to pulling the plug suddenly. The
disk will look like it did at some instant of time, but the apps
that were running may have all sorts of transactions partly done
and by the way will be thinking all was well so they may blow
away logging info you'd need to recover them.

You are in general far better off stopping the app momentarily
to ensure it writes things out to disk before you split the shadow
set. Otherwise you may wind up with a totally corrupt application
storage area even though the disk as a file structure may be
completely sensible.

Something that would shadow all writes to a file that started
before starting an online /ignore=interlock backup, and that
stopped after this was done, could allow you to have between
the backup and the shadow journal an image of the disk at the
instant the journal finished. The DEC snapshot capability gives
something more complex but equivalent...that you can buy.

However, getting a single-instant snapshot of the disk is only
part of the problem. Getting a snapshot of application storage
that can be used by the app if you ever have to restore it
is the other and more difficult problem.

I've backed up lots of disks with /ign=inter and been able to
use the backups fine, but I also do it a LOT and figure that
any database I might be running during this activity can be
possible toast. If you aren't willing for this to be the
case and to maybe have to go WAY back to find a good copy,
shut the thing down before backup. Where I've had to work with
DBMS systems, I either did this or figured I wasn't backing up
the DBMS that time out.

Even a DBMS that claims it's fine if the plug is pulled at any
instant could have problems unless your snapshot includes
ALL of whatever it uses for journals in its storage. If the
journal is on another disk, your image may not match anything
that's left in the journal and you might as well start over.

Loss of resilience during backup is a tiny issue. Data integrity
of apps that may use many disk read/write operations in a single
transaction is a huge one.

Have fun.

;-)

Glenn Everhart

POWERS John wrote:
> 
> Hi.
> 
> I have received a request to do an image backup of a disk that has files
> always open on it. The users do not wish to stop the application that
> opens the files. The disk is shadowed. They have asked if it is possible
> to unshadow the disk, then backup the copy of the disk on the shadow and
> then re-shadow the disk, without interruption to any processes running
> using files on the disk.
> 
> I have pointed out that while the disk is unshadowed they would lose the
> resilience that is the reason they shadowed, but they are prepared to
> live with that for the time it takes to backup, if this will work. Has
> anybody tried this idea, and will it work? The problem with using the
> simple /IGNORE=INTERLOCK, is that different files will be backed up
> at different times, and they are worried the backup will be unreliable.
> Unshadowing the disk, they hope, will mean all the files are in the
> same state. Has anybody got any advice, a pointer to the page in the
> great white wall, or any horror stories of people who tried this and
> came unstuck, that I can use to talk them out of it.
> 
> Many thanks for any help.
> 
> John.