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

Re: Optimising Debian-CD

On Tue, 2002-04-09 at 13:39, Raphael Hertzog wrote:
> Le Tue, Apr 09, 2002 at 11:43:18AM +0100, Philip Hands écrivait:
> > If the CD building machine had a tendency to corrupt data comming off
> > it's disks (as happened with a previous incarnation of open) or the
> > mirroring run had failed in a subtle way, then a simple loop mount of
> > each image, and running md5sum -c on each, should catch that straight
> > away if the md5sums were not locally recalculated.
> You can do whatever you want, but we should try to limit the number of
> dependencies we create against the mirror. What I mean is that it should
> still work even if the md5sum.gz is missing from the local mirror...

Well, I've made it's use conditional on setting FASTSUMS=yes for the

It reads $BDIR/indeces/md5sums and $BDIR/indeces-non-US/md5sums, ignores
things with Packages, Sources, or Release in its name, then does a
(slightly stramelined) version of the find, uses the md5sum from it's
hash if there is one --- if not it generates it on the fly.

If the indeces*/md5sums were empty, it would simply md5sum the lot,
probably quicker than the old method, because its all done within perl,
so there's no forking cost for the many md5sums, so we can probably get
rid of the old code at some point.

I'll commit it to CVS once I've managed a successfull build run with it,
if that's OK.

> > A slightly more ambitious replacement would be to generate the Packages
> > files by taking the records out of the main archive's Packages files.
> I don't think it's worth it. On the contrary, building it again using
> the overrides file (local and from the mirror) is a good idea since it
> lets you define your own "standard" for example (overriding Debian's
> "standard" distribution).

Yeah, I was looking into that, and decided it was more trouble than it's

The fast_sums thing is definitely worth the effort though, since it runs
in just over a minute for i386 on open, as opposed to over 10 minutes
for the old approach.  For a full build that should come to a saving of
about an hour and a half.

> > Opinions?
> Yeah, you're one of the very few people who builds images for all
> arches, so it's an improvement for you since it would be faster ... but
> I don't see what other wins we'd have. The possibility that the files
> get corrupted between the mirror and the CD set is quite limited.

Well as I said it happened for 2.2r0 (IIRC) --- I can do without it
happening again.

Cheers, Phil.
Say no to software patents!  http://petition.eurolinux.org/

|)|  Philip Hands [+44 (0)20 8530 9560]    http://www.hands.com/
|-|  HANDS.COM Ltd.                    http://www.uk.debian.org/
|(|  10 Onslow Gardens, South Woodford, London  E18 1NE  ENGLAND

Attachment: signature.asc
Description: This is a digitally signed message part

Reply to: