ideal config management framework [was: CPU Frequency Scaling]
Davide Bolcioni
db-fedora at 3di.it
Wed Dec 6 12:11:04 UTC 2006
Thomas M Steenholdt wrote:
> Also - Think of the additional error scenarios when init is gconf'ed.
> The same complexity is added for everything else that adopts a registry
> based configuration engine (since that engine has to be operational for
> it to work in the first place) but may not be quite as aparrent.
As I suggested in another message in the original thread, if this is
changed, i.e. everything adopts a registry-based configuration *if
one is found* but happily reads plain old configuration files from
/etc if not, this problem can be made less aggravating.
Personally, a scheme where (a) if the registry or policy daemon is not
operational it is supplemented by the configuration in /etc, and (b)
said configuration *overrides* unconditionally whatever the registry or
policy daemon spits out, would offer the right mix of safety and
functionality to give it a try on a non-production host, which seems the
sweet spot for Fedora. The scheme might get fancy and differentiate
between /etc/foo.conf which overrides unconditionally and
/etc/fallback/foo.conf which is read if the registry or policy daemon is
not available (and possibly overridden shortly thereafter).
For production use, however, the registry or policy daemon *must*
preserve comments and be manageable using whatever version control is in
use at a site (not everybody uses CVS or SVN); very often, there is a
reason behind setting something in a configuration and having to record
it separately is just one of the many shortcomings of registry-based
approaches.
Just my $0.02,
Davide Bolcioni
--
There is no place like /home.
More information about the fedora-devel-list
mailing list