[OpenAFS] Re: OpenAFS services startup

Andrew Deason adeason@sinenomine.net
Thu, 1 Mar 2012 17:50:08 -0600

On Thu, 01 Mar 2012 17:58:48 -0500
Danko Antolovic <dantolov@indiana.edu> wrote:

> RPM installs occasionally fail because of kernel versions, incompatible 
> architecture and what not. In such cases, source build is usually 
> considered the preferred fallback.

...then it sounds like it's supposed to fail. If you have a mismatch in
kernel version or architecture, something's going to fail along the line
for the client, whether it be RPM complaining, or the kernel panicing.

If there is some case where this fails when you don't think it's
supposed to, that's something that should be fixed, but we can't fix
anything if we don't know about it. I don't think I've ever seen a valid
use case for not using the packages in a production environment.

> OpenAFS documentation already branches out to different Unix flavors
> in one or two places, so there is precedent for documenting
> OS-specific things.

Some of the installation instructions for various Unices are there
because there is (or was) no practical packaging for them; those manual
steps that you have to perform are (or were) the _only_ way to do an
installation. They are also for Unices that tend to not change in this
area very often, so in theory the documentation probably does not need
to change very much (in practice, we probably have no idea if it is
still correct for some platforms).

There is no way to provide accurate documentation for a manual "Linux"
install at the level of integrating with the init scripts and the system
configuration conventions. That integration is different between almost
every single Linux distribution and (sometimes) version that exists; to
some extent that is what makes them different distributions. It's
probably possible to provide instructions for "installing on RHEL 4.2"
and "installing on Debian Lenny update 2" and "installing on Fedora 12",
etc etc, but this is impractical. This is even more impractical for us,
since even without this we have trouble keeping up with documentation
upkeep. I think there may even be some "Linux"-specific instructions
somewhere in there (or there used to be), but it was only for some
version of pre-RHEL Red Hat or something, which today is pretty useless.

If you want an init script... which one do you want? The Debian one, the
'official' RHEL one, the other RHEL one, the SLES one...

If you want the RHEL-specific instructions, in my opinion they start
with "rpm -i" or (preferably) "yum install". Anything else is deviating
from the documented path.

> A startup service typically consists of placing the right script in
> init.d, a config file in /etc/sysconfig,

Says who?

Andrew Deason