On 27 Mar 2009, at 23:00, Peter Tribble wrote:
So, a question for the weekend:
When administering something, do you prefer to issue a bunch of
commands
to set configuration parameters, or edit a configuration file that the
thing can read?
A couple of basic prerequisites for any kind of configuration changes
are:
1. They need to be usable on a "dead" system - you need to be able to
apply the change to a system whose filesystem you have access to but
on which you can't run commands (for instance a non-booted zone, or
any other kind of system image). So any command must support a "do
these changes to this system image" option.
2. You need to be able to keep track of changes to configuration in a
useful way: such that differences are meaningful (so you can use some
kind of source control system for your configurations), and such that
saved configurations can be rolled back in. So any command must
support a "dump out the current configuration in a standard, stable
format" as well as a "read in a configuration and apply it to the
system or image".
If you think a bit about the implications of (2), particularly, you'll
realise that the print/read formats that commands must support to be
useful in fact *are* configuration files. So there is no real
difference between the command-based and the file-based approach, if
the command-based one is done properly[1].
Unfortunately the command-based one is often not done properly, which
is a significant pain for people trying to look after large numbers of
machines in a reasonably coherent way.
--tim
[1] Actually there is a difference: the files do not have documented
names or locations.
_______________________________________________
sysadmin-discuss mailing list
sysadmin-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/sysadmin-discuss
