Hi Jon,
* Jon LaBadie <jon AT jgcomp DOT com> [20070926 17:49]:
> On Wed, Sep 26, 2007 at 04:26:50PM -0400, Jean-Francois Malouin wrote:
> >
> > One of the most important feature of amanda was (it still is but read
> > on) the possibility of doing a bare-metal restore with just a few
> > utilities. I say was because in my case the ever increasing amount of
> > data, the size of the DLEs and the number of chunks needed in order to
> > fit in my tapes (LTO-1 and LTO-3) was getting out of hand --too much
> > likely to 'forget' some data while including/excluding dirs in a DLE--
> > so I had to use the tape spanning feature. I get a much better tape
> > usage but now I can't do bare-metal restores in case something real
> > bad happens. I understand that I can't have both the cake AND the
> > cherry but I would be hard pressed to justify using amanda to our
> > computer people at my site when such an important feature is missing.
> >
>
> Jean-Francois,
>
> I think you lose less than you realize. I've not tested a bare-metal
> restore with my spanned dumps, but here is how I view the procedural
> difference from a non-spanned dump.
>
> For non-spanned:
>
> 1. you need to locate the specific tape file where the DLE begins.
> 2. you need to use mt to position your tape at the beginning of the file
> 3. you need to use dd to strip off the first 32KB of the tape file.
> 4. you need to feed the rest of the file to your restore command(s),
> eg. unzip, tar, ...
>
> For spanned:
>
> 1. the same except you also need to determine the number of chunks and
> whether it spreads to multiple tapes
> 2. the same
> 3. a modification
> 4. the same
>
> Recognize that the spanned pieces are all one after the other on the
> tape, and if crossing to another tape, at the beginning of that tape.
> But, if it crosses to another tape, the last chunk on the tape is
> not valid.
>
> So step 3 becomes not one "dd bs=32k skip=1 ..." but a loop through
> the chunks doing the same dd command on each chunk. In a scenario
> where you knew the dump was in 5 chunks, step 3, piped into step 4
> could be simply:
>
> for i in 1 2 3 4 5
> do
> dd bs=32k skip=1 if={your tape dev}
> done | {step 4 commands}
>
> Obviously it would be more complex if the DLE were split over
> multiple tapes, but then something like this is possible:
>
> (
> FOR LOOP FROM ABOVE
> tape change commands
> tape position commands (to skip over amanda tape header to first file)
> ANOTHER FOR LOOP
> ) | {step 4 commands}
>
> And I've ignored the bad last chunk.
>
> If sufficient disk space were available, you could just save
> each chunk with a sequential number. Then cat the chunks
> piping them into the step 4 commands.
>
> Amfetchdump tries to automate this.
> And if I understand the -i option, assists in locating the chunks.
> But it still can all be done manually.
>
> What you lose by spanning is the simplicity of having the entire
> dump in a single tape file. But the manual recovery procedure
> is still basically the same.
Yes, you're right that it is manually feasable. However conceptually
simple the task is, I feel that potential mistakes are bound to
happen, particularly when doing a restore under pressure! Remember,
you have the mob^H^H^Husers on your back, yelling bad words at you :)
I'll see if I can write a dirty little script that could somehow
automate the procedure...
Thank you for taking the time to comment!
jf
>
> --
> Jon H. LaBadie jon AT jgcomp DOT com
> JG Computing
> 4455 Province Line Road (609) 252-0159
> Princeton, NJ 08540-4322 (609) 683-7220 (fax)
--
<° ><
|