>b) Wait for multi-tape support in amanda (ETA ?)
I'm reasonably (but not 100%) sure it will be done this millennium :-).
>c) Split the filesystem on the server
Here's what we (I) do (and it's not at all pleasant):
* Have enough holding disk space to handle a file system that does
not fit, preferably a lot more space.
* Added a small hack to driver to fail without trying any file
system larger than the tapetype "length" parameter.
* File systems that fail fall into two categories, those that would
fit with software compression (which we don't normally use) and
those that won't fit no matter what.
* File systems that would fit with software compression are run
through "amrestore -chp <input-image> > out.xxx". Then the original
images are removed, the "out.xxx" image renamed and amflush run to
write the compressed image to tape. Hardware compression is manually
disabled for that one tape.
Restoring from these tapes is transparent -- Amanda just does the
right thing.
* For the rest, I have the holding disk chunksize set to ~90% of
tape capacity. So when one of these fails to go to tape, I have one
big image and one smaller one. I move the smaller one up a directory
level so amflush won't process it, then run amlink (I posted that
a while ago) to reset the header on the big chunk so it does not
look like it continues on to the smaller chunk. Then amflush is
used to write that chunk to tape. The smaller chunk is written to
a non-Amanda tape with dd.
Restoring these file systems is a PITA. First, they have to be
recognized as being split onto two tapes. Then the pieces have to be
brought back into the holding disk in two chunks as though they had
just been dumped there. Then amlink is run to link them together.
Finally, a small patch (already in CVS) to amindexd (or amrecover,
I forget which) makes amrecover see the holding disk instead of the
tape version and the rest works OK.
>d) let amanda believe that the filesystem is split
> - create two dumptype, each with a seperate exclude list
You can do this in the disklist without separate dumptypes:
<client> <disk> {
... dumptype stuff ...
... dumptype stuff ...
... dumptype stuff ...
} [ spindle [ interface ]]
See the amanda(8) man page.
> - add two entries to the disklist with separate dump type (will
> amanda backup the same partition twice ?)
Amanda will not allow the same client/disk being mentioned more than once.
The usual trick is to use mount point names (e.g. "/var") instead of
disk names and add on a trailing "/.", e.g.:
<client> /var <some-dumptype> ...
<client> /var/. <some-dumptype> ...
I'm not 100% certain if this works with dump. I do know it will work
with GNU tar, and since you're talking about exclusion lists, you're
already committed to that anyway.
>Do you think d) can be made to work ?
Other people have done it this way.
>Kris
John R. Jackson, Technical Software Specialist, [EMAIL PROTECTED]
