Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 21 Jul 2013 15:08:36 -0700
From:      Peter Wemm <peter@wemm.org>
To:        hubs@freebsd.org
Subject:   FYI: Upcoming cvsup changes
Message-ID:  <CAGE5yCqbtGfFN_BGFjncD6%2BAhZ9knA8fL6cgorxmk=O-oopwaA@mail.gmail.com>

next in thread | raw e-mail | index | archive | help
Right now, we have a rather fragile cvsup setup that's a bit painful
to deal with.

cvsup-master depends on a scan file for efficiency.  So we have to
double-cvsup internally to generate the scan files.

It looks like this:

www -> via cvsup(rsync mode) -> cvsup-in -> scan file -> cvsup-master -> mirrors
www -> via rsync -> internal rsync mirror -> (not published)
cvs -> via cvsup(cvs repo mode) -> cvsup-in -> scan file ->
cvsup-master -> mirrors
mail-> via cvsup(rsync mode) -> cvsup-in -> scan file -> cvsup-master -> mirrors
mail -> via rsync -> internal rsync mirror -> (not published)
gnats -> not in cvsup anymore, but it was like www/mail.
gnats -> via rsync -> internal rsync -> semi-published on bit0
portsindex -> via rsync -> internal rsync -> semi-published on bit0

This means we have cvsupd on many more systems than we'd like.  I'm
planning to change it like this:

www -> via rsync -> internal rsync mirror -> published
mail -> via rsync -> internal rsync mirror -> published
gnats -> via rsync -> internal rsync -> published
portsindex -> via rsync -> internal rsync -> published

and on a backend worker in the cluster:

www -> rsync -> stage area
mail -> rsync -> stage area
gnats -> rsync -> stage area
cvs(local file system) -> stage area

then:
stage area -> cvsupd (loopback) -> create scan file -> publish as cvsup-master.

This should be mostly transparent for cvsup mirrors.

As of today there is still some data that can only be obtained via
cvsup and this change will finally fix it.  After the change, cvsup
will be 100% second-hand and/or converted data.

cvsup has had a very good run, but this will likely be the final step
in completing the deprecation of the system.

Note that I did NOT say "turning off" - there is a fair bit of effort
going into keeping it alive.  This work is for sustaining it so it can
wind down gracefully.  But this is probably the last major chunk of
work being spent on it at our end.  (FYI, remember CTM? It's still
going due to a madman^H^H^H^H^H^Hvolunteer. Said in the nicest
possible way of course :)

cvsup-master will likely be changing IP addresses but we'll reduce the
dns TTL's to keep this as transparent as possible.  Naming for the
rsync infrastructure is internal right now but that'll be sorted out
as part of this and documented.

cvs tree status:
6.x (and earlier) hasn't been exported to cvs for quite a while.
7.x hasn't been exported to cvs for a few months.
8.x is still being exported once an hour. (including security support branches)
9.x is still being exported once an hour. (including security support branches)
10.x is not exported from svn to cvs.
ports has not been exported since February.
doc/www was never exported from svn to cvs.

BTW; Some mirrors have been rapidly updating multiple times per hour.
Please don't.  Once an hour is more than enough - there will never be
more than one cvsup-master update per hour.
-- 
Peter Wemm - peter@wemm.org; peter@FreeBSD.org; peter@yahoo-inc.com; KI6FJV
UTF-8: So you can \342\200\231 .. for when a ' just won't do
<brueffer> ZFS must be the bacon of file systems. "everything's better with ZFS"



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAGE5yCqbtGfFN_BGFjncD6%2BAhZ9knA8fL6cgorxmk=O-oopwaA>