[conspire] Fwd: wheezy is OUT. tonight. Re: ANNOUNCEMENT: BAD meeting for 2013-05 (and mini-installfests, and ...)

[Rick Moen]

Quoting Nick Moffitt (nick at zork.net):

> Historically you advised extra caution when making large jumps in
> package sets, e.g:

However, it is the job of the Debian Release Manager to ensure that
going from stable n to stable n+1 via the approved process is a smooth
transition, and they have 18 years of consistent success in making that
happen.  Moreover, the disadvantages to remaining pegged to the named
branch that (at first) bears the functional name of oldstable (that used
to be stable) are serious and accumulate, notably the very common error
of forgetting for a long time that you've just sidetracked your system
to a track that will soon be completely unmaintained.  This is one of of
the more reliable ways to sabotage a Debian system I know of.  More to
the immediate point, the alternative of just leaving
/etc/apt/source.list[.d/*] tracking 'stable' Just Works, so don't fool
with it.

Mind you, in the name of extra caution when making large jumps in
package sets, if the user suddenly realises after he/she hears about a
Debian release that his or her 'stable' system hasn't gotten apt updates
for six months or a year, I _would_ advise some extra measures, but not
what Tony advised:

1.  First and foremost, Don't Do That, Then.  Avoid the problem in
advance by fetching package updates on at least a halfway regular basis,
like every week or fortnight.

2.  Second, since you failed to do that and did the Dumb Thing of
lagging behind package maintenance for a long time, yes, you _are_ best
advised to catch up to the latest of what was _formerly_ the stable
branch.  But you must remember to switch back to stable, after that,
because the system won't remember for you.

What I'd personally do if caught in that regrettable scenario is edit
'stable' to 'oldstable' in /etc/apt/source.list[.d/*], do 'apt-get
update && apt-get dist-upgrade', change /etc/apt/source.list[.d/*] back,
repeat.  That way, your system at the end of the first update very
closely approximates the tested pre-upgrade state, the one the Debian
Release Manager contemplates system being in before moving to stable
branch n+1.

The two cases I'm aware of, of packages that did not upgrade well from
Debian stable branch n to n+1 were both major server packages apache
(1.3.x) and exim (3.x).  In both cases, Debian sysadmins had been put on
notice in various ways for years that these were obsolete packages that
were being replaced without a direct upgrade path, to related but not
directly compatible packages apache2 and exim4.  Moreover, the old
packages still were around, remained functional if previously installed,
and could in a pinch just be kept in their old installed functional
states.  They just weren't going to get any more updates, and sysadmins
eventually would need to craft new conffiles manually for the
replacement packages, and remember to install the new package and remove
the old one.


In a similar case I personally encountered, Ubuntu Server badly muffed
the obsoleting and replacement by an incompatible successor of a crucial
server package during the jump from LTS release 10.04 'Lucid Lynx' to
LTS release 12.04 'Precise Pangolin'.  I refer to the authoritative DNS
server package 'nsd', which simply broke silently and completely upon
LTS -> LTS+1 upgrade, leaving the sysadmin (me) completely in the lurch
and obliged to figure out the problem, figure out 'nsd' had just become a
non-functional transition package, figure out that successor package
'nsd3' needed to be manually installed, and manually craft a new
conffile with radically different syntax, in new directory /etc/nsd3, and
also to move the zonefiles over to that new tree.

Not impressed with that experience.  It broke SVLUG's DNS service.
http://lists.svlug.org/archives/volunteers/2013q2/004348.html
http://lists.svlug.org/archives/volunteers/2013q2/004362.html

And yes, I most certainly did do 'sudo do-release-upgrade', with
additional precautions and a special kernel step necessary to follow
Linode's steps at
https://library.linode.com/upgrading/upgrade-to-ubuntu-12.04-precise .


> But of course I get the sense that most Debian Developers rarely use
> stable itself, and that makes testing rather more compelling if you can
> support it.

Yes, I fully agree, and I actually urge that for anyone but the very
most nervous and least able to cope with occasional small upgrade
glitches.