[mirror-admin] Seeding newkeys with old content?

[Chuck Anderson]

On Sat, Sep 06, 2008 at 08:24:30PM +0300, Axel Thimm wrote:
> On Sat, Sep 06, 2008 at 11:51:15AM -0400, Chuck Anderson wrote:
> > On Sat, Sep 06, 2008 at 10:44:32AM +0200, Adrian Reber wrote:
> > > On Sat, Sep 06, 2008 at 08:34:46AM +0300, Axel Thimm wrote:
> > > > does it make sense to seed the new rpms with the old signed ones?
> > > > Would rsync note the change of a few bytes (the lead-in/out) and could
> > > > it save the sync from redownloading the contents? Did anyone give it a
> > > > test?
> > > 
> > > Doing cp -al before running rsync helps a lot on my side.
> 
> I'm trying this, still syncing.
> 
> > Excellent.  Can the master .newkey directories please start out as a 
> > hardlinked tree of the original directories for a few days before 
> > pushing out the re-signed content?
> 
> That's impossible, the files *are* different. cp -al is just for a
> faster seeding, it wouldn't make an rsync difference if it were w/o
> the -l.

Well the idea would be to have the master do cp -al from x to x.newkey 
everywhere, let the mirrors sync that with rsync -H which won't 
actually download anything--just hardlink the content.  Then after a 
day or two, replace those hardlinked files on the master with the 
actual new files.  Then the next rsync -H the mirrors do will only 
download the tiny parts of the files that changed due to the new 
signatures.

Granted, this does create a problem for those who use rsync without 
-H, or FTP/HTTP to mirror.  They will end up downloading everything 
twice.  Perhaps we should require mirrors to use rsync with -H.

--