another upgrade, another disaster

Neal Becker ndbecker2 at gmail.com
Thu May 31 19:08:54 UTC 2012


Adam Williamson wrote:

> On Tue, 2012-05-29 at 16:50 -0400, Tom Callaway wrote:
>> On 05/29/2012 04:46 PM, Corey Richardson wrote:
>> > I've heard nothing but bad things about preupgrade from lots of people,
>> > and I've heard the developers either never hear about it, ignore it, or
>> > don't care. I tried a preupgrade and it half-succeeded, I had to fix
>> > the initramfs and a few other things afterwards though. IMO, it isn't
>> > worth the pain.
>> 
>> We do test preupgrade as part of the pre-release testing. The problem is
>> that people do all sorts of wacky things to their systems that cause it
>> to not work well. I'm sure that QA would love to have more people
>> testing preupgrade during the beta->RC window.
>> 
>> So, your statement that devs don't care isn't valid. It may still not be
>> worth the pain, depending on how much you have installed outside of the
>> Fedora Package universe.
> 
> Not entirely valid, but if we're being completely honest, there is only
> one preupgrade maintainer - hughsie - and he has a couple of other
> full-time jobs. So I think it's fair to say it doesn't get as much
> active maintenance as it might. The preupgrade bug list is quite long -
> 127 bugs, http://bugz.fedoraproject.org/preupgrade - and I know for a
> fact it contains quite a few valid reports which could be fixed
> by...development.
> 
> Your broader point is of course also true. upgrading is a fundamentally
> unsupportable operation: the amount of variables in play when it comes
> to upgrading a system which a person has been using, normally, for some
> time is so huge as to be probably incalculable. What we test in a
> programmed way is that you can do a clean installation of F16, update it
> to current packages, then run 'preupgrade' and wind up with an F17
> system that boots. That is the extent of the preupgrade validation
> testing. Any further testing relies on people trying it and filing bugs,
> as Tom says.
> 
> The particular failure scenario involved here - 'user let
> SOME_THIRD_PARTY_THING screw up RPM-managed files' - is obviously quite
> a long way down the list of things we're going to care about. There has
> to be a limit to what we can achieve with the resources available to a
> project like Fedora. We don't have a 2,000 system test lab stashed away
> somewhere with a few hundred people in it running upgrade tests in
> various arbitrary scenarios, 24 hours a day. We just don't. All we can
> do is ask people to test and file bugs prior to release, and I think we
> certainly do enough of that...

But we can, and should, at least try to make our systems tolerant of failures.  
Just because we can't test everything.  Defensive programming.



More information about the devel mailing list