On Tue, Dec 09, 2008 at 06:19:32PM +0100, Sven Rudolph wrote: > Jon LaBadie <[EMAIL PROTECTED]> writes: > > > On Mon, Dec 08, 2008 at 01:15:21PM +0100, Sven Rudolph wrote: > >> Nick Brockner <[EMAIL PROTECTED]> writes: > >> > >> > I have an interesting problem. I have old archive tapes on DAT72 > >> > tapes which I need to move over to my new LTO3 tapes. The dat72 > >> > changer is on a different server than the new lto3 changer. > >> > > >> > What is the best way to migrate the data? > >> > >> Probably not the best way, but I'll try to describe what I do: > >> > >> I copy the old tapes to disk, each in a separate directory named 1, 2 > >> and so on. Each "tape file" (the things separated by "file marks") is > >> copied into on disk file, they are numbered too. Example: > >> > >> mkdir 1 > >> cd 1 > >> > >> n=0 > >> > >> while dd if=$TAPE bs=32k of="$(printf '%03d' $n)" > >> do > >> n=$((n+1)) > >> done > >> > >> When I copied enough tapes that will fit on a new tape, I do some > >> checking, and after that I write the files to a new tape: > >> > >> for f in */* > >> do > >> dd if=$f of=$TAPE bs=32k conv=sync > >> done > >> > >> Afterwards some more checking... > >> > >> The resulting tape can be used with amrecover, but you can not use the > >> "amrecover_do_fsf" and "amrecover_check_label" options, because the > >> amanda database does not know the file positions of the new tape. Its > >> OK for me, restores from these old tapes are very rare. > >> > >> I started using this with DLT-IV (35 GB), and now I use LTO-4 (800 > >> GB), so one new tape replaces more than ten old tapes. > >> > > > > Interesting Sven, I'm trying to wrap my mind around how a > > typical amrecover session would work. Suppose I've used > > amrecover to mark several files/directories for recovery. > > Amrecover then uses its index and suppose it says that > > tape "old-7" is needed. I consult my migration TOC and > > see that tape "old-7" is one of the tapes migrated to > > tape "new-2". So I put tape "new-2" in the drive. > > > > What happens next? > > Or are my assumptions of the steps to this point faulty? > > An old tape contains these files > 1 Amanda Header (for TAPE001) > 2 First Data File > 3 Second Data File > > and the second old tape contains: > 1 Amanda Header (for TAPE002) > 2 Third Data File > > Now these are concatenated onto a new tape: > 1 Amanda Header (for TAPE001) > 2 First Data File > 3 Second Data File > 4 Amanda Header (for TAPE002) > 5 Third Data File > > When I want to recover the Third Data, I insert the tape, position it > with mtx and let amrecover read it. > > There is nothing like a "new-2" name, at least not known to > Amanda. Amanda does not know that these tapes reside on a new larger > tape. (Thats the reason why Amanda cannot check the tape header: It > rewinds, finds the TAPE001 header and decides that it is the wrong > tape.) > > The list mapping the Data File names to the positions is kept outside > of Amanda. (I keep it as a both as a file and as paper inside the > physical tape box.) >
I think my big problem was thinking that amrecover would rewind the tape before checking the label. We gotta find a way to redo the indexes and other files that are needed for recovery and continued dumping. jl -- Jon H. LaBadie [EMAIL PROTECTED] JG Computing 12027 Creekbend Drive (703) 787-0884 Reston, VA 20194 (703) 787-0922 (fax)
