Hello:

I will giving a BOF at Decorum on "AFS Disaster Recovery." Of course,
this involves rebuilding a new cell from the existing UBIK databases and
backup tapes and some of the issues with doing that. Very similar to
cloning a cell.

One of the assumptions I make in recovering a cell is that the cell name
will remain the same, but I do explore the question of changing the cell
name. You can do it but, as Tony pointed out, the passwords for all the
users will also need to be changed because the keyfile stored within the
kas database is based on the user's password and the cell name.

Excerpts from transarc.external.info-afs: 16-Feb-98 Re: cloning a cell
to a new.. Rudolph [EMAIL PROTECTED] (910)

> How about having CellServDB on the test servers and clients include the
> test IP addresses but for the real cell name?  That's what we intend to
> do for testing.

> It's not too convenient to not be able to access both cells at the same
> time, but it's only for testing...

Dangerous, confusing, and gives you little benefit (other than users
keep their own password). Actions on volumes (backup, move, remove, ...)
within the test cell can occur on volumes within your real cell creating
loss of data or incorrect location data in the real cell's VLDB.
Changing the keyfile in the test cell is HIGHLY recommended though
results in tokens getting discarded as they access data in the real cell.

Moving the VLDB is the cause of the danger. As you try to clean up the
test cell's VLDB, the changes may occur in the real cell (like removing
ROs or BKs or volumes not restored into the test cell or restoring
volumes).

Excerpts from transarc.external.info-afs: 18-Feb-98 Re: cloning a cell
to a new.. Ken [EMAIL PROTECTED] (284)

> Actually, this begs another question ... if I wanted to change the AFS
> cell name, forgetting for the moment about the Authentication Server ...
> could I do it?  I know pathnames would break, but I wonder if the cell
> name is embedded into anything in the AFS volumes themselves.

Only if the mountpoint is defined as a cross cell mountpoint. Otherwise,
it contacts the local VLDB (/usr/vice/etc/ThisCell) looking for the
volume.

        - John Morin
          Advanced Member of Technical Staff
          Transarc Corporation

Reply via email to