On 2 May 2002, Philip Hands wrote: > On Wed, 2002-05-01 at 03:38, Mattias Wadenstein wrote: > > On 27 Apr 2002, Philip Hands wrote: > > > The structure is: > > > > > > The *.jigdo, *.template and MD5SUMS for each architecture are here: > > > > > > jigdo-area/<version>/jigdo/<arch>/ > > > > And the final images should go into /debian-cd/<version>/<arch>/ ? > > I was thinking of putting them in a subdirectory: > > /debian-cd/<version>/images/<arch>/
Ok. > Mainly because on open, the images will need to be on a different > partition from the snapshot (which has to live on /home because of the > hard-links to the archive). > > It would also make it more obvious for people that want to mirror only > the images, or perhaps all bar the images. Yeah, that sounds good. > Also, to start with at least, the images will only be a partial set on > open, because leting people rsync 90-odd CDs off open would be a > mistake. Yeah, but if we manage to borrow a nice temporary server for release, we should be able to have all online within 4 hours of getting the jigdo files. Note the "if" though. > AJ's latest suggestion, which seems like a good plan, is to point > cdimage.debian.org at raff.d.o instead of open, and have all the jigdo > files (which should have no non-US stuff in them, so raff being in the > USA os fine) and the main part of the This seems to be a truncated paragraph. > Then we could just have the 1_NONUS images on open (as well as all the > jigdo files) and call it something like nonus-cds.d.o Ok, because what I need to sync is the jigdo files and templates and then I could probably manage to build them fine (there should be no need for fallback if I build them the same day). > > > BTW does the fact that the template is pointing at cdimage in this file > > > mean that the whole world are going to try to pick up the templates from > > > there, rather than the mirror the grabbed the jigdo file from? > > > > Well, for the mirror situation, it (jigdo-mirror) seems to find them > > locally when you have mirrored it. Mirroring the snapshot is probably > > harder though. > > Having raff as the home of the main snapshot seems to be the way to go > here. Yeah, that seems resonable. I'd like to say that we could have a snapshot mirror over here, but we have scalability concerns with the number of inodes on our ftp server. Is there a script around to build a snapshot given a release? (If you are a bit late and miss a few files, that is a simple rsync to an existing snapshot location to fix..) > > > Anyway, have a play and tell me what breaks. The alpha and sparc images > > > are still not right, but I thought getting this out for testing was a > > > higher priority. > > > > Trying to build them now with (woody) jigdo-file and jigdo-mirror gives: > > Found 1322 of the 1356 files required by the template > > Copied input files to temporary file `image.tmp' - repeat command and > > supply more files to continue > > 2002-05-01 04:10:19: 34 parts still missing from image > > 2002-05-01 04:10:19: Too many files missing in local mirror > > Shouldn't it be getting the missing bits from the snapshot? Yes, the default parameter "maxMissing=25" was too small in jigdo-mirror. When I increased it, it worked fine. I managed to build a full set of the pre2 images (except for alpha that is). Richard: Have you made sure that it will never happen that non-US/Contents-i386.gz and main/Contents-i386.gz will be downloaded in the same fetch batch in jigdo-mirror? They have the same name, so wget might overwrite or something. Just tell me that you have thought about it and solved the problem. :) /Mattias Wadenstein -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
