>I can't believe that a serious distribution like debian do not do
>tests before suggest an upgrade.
As Scott says, for production systems you either have to test things
yourself, or you accept that possibly an upgrade may break your
system. Debian (and the other distributions) do do a **LOT** of
testing - they have a whole QA setup just to handle testing etc.
BUT they rely on the package maintainers/contributors to do most of
the functional testing - the distribution testing is (I believe)
mostly to ensure everything installs cleanly, there aren't broken
The thing is, once you get past the trivial (and both the Linux
kernel and Xen are very much non-trivial) it is simply impossible to
test every possible combination of software and usage cases. That's
not "difficult", but "impossible" - unless you have a near infinite
number of systems and people to run the test scenarios.
I understand how frustrating it is for you, I've been hit by this
sort of thing before. You either have to have a test/QA setup and
test all upgrades before going live, or you upgrade live systems and
have a plan to backtrack what you upgraded. The last one I had was an
incompatibility between (IIRC) Squirrelmail and (IIRC) one of the
Perl elements it depends on - leading to inability to log in as a
user. I had to downgrade the bits I'd upgraded, then upgrade them one
at a time until I found the problem. Fortunately it got fixed at some