[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Bug#531154: mirrors: Same problem on ftp.uni-kl.de

Simon Paillard (simon.paillard@resel.enst-bretagne.fr) wrote on 14 June 2009 17:37:
 >reopen 531154
 >retitle 531154 libasound2_1.0.20-2_amd64.deb corrupted on ftp.de.d.o and consequences
 >On Sun, Jun 14, 2009 at 12:10:14PM +0200, Adrian Knoth wrote:
 >> Package: mirrors
 >> Followup-For: Bug #531154
 >> I don't know if I should open a new bug, but I see the same problem on 
 >> ftp.uni-kl.de:
 >> [..]
 >> Failed to fetch  http://ftp.uni-kl.de/debian/pool/main/a/alsa-lib/libasound2-dev_1.0.20-2_amd64.deb  
 >> Hash Sum mismatch
 >A short status:
 >* rsync by default uses only modification time and file size to sync
 >  files
 >* ftpsync doesn't use rsync checksum by default
 >* ftp.de.d.o (on top of many servers I guess) doesn't allow server
 >  checksum because of the impact on system load.
 >So as a consequence, if a file is corrupted (but mtime and size match)
 >on a rsync server somewhere between ftp-master.debian.org and a debian
 >mirror, fixing one primary rsync mirror is not enough.
 >The only ways to fix it are:
 >1/ on a rsync server with the fixed file: touch the modification time to
 >   force resync on downstream mirrors
 >2/ on every downstream mirror with bad file: touch the modification time
 >3/ on every downstream mirror with bad file: delete the file
 >Solution 2/ or 3/ can be automated by checking the MD5 sums of the whole
 >archive on a periodic basis, then delete them:

Yes. Mirrors, at least the primary ones, should verify the checksums
from time to time. We do it at least twice a year.

 >wget -q -O - http://some.mirror/debian/indices/md5sums.gz | gunzip -c - | LANG=C md5sum -c - 2>&1 | grep "FAILED$"

A mirror should already have it, wget is not necessary.

This will take a looong time... One might want to run several md5sums
in parallel. It'll be much faster but will put a heavy read load on
the disks.

Reply to: