In a message dated: Tue, 14 Nov 2000 16:54:04 EST
Benjamin Scott said:

>On Tue, 14 Nov 2000, Paul Lussier wrote:
>> Why would I want to NFS mount my /usr partition?
>
>  A cluster full of diskless workstations comes to mind...
>
>  I believe the Linux File System Standard actually specifies that /usr should
>be mountable read-only to support just this sort of operation.

Just because a "Standard" specifies doesn't mean it's actually used by anyone.
And yes, there probably are a few people out there who haven't discovered that 
disk space is cheap and keeping /usr local saves a ton of bandwidth (remember, 
people using diskless clients are also probably using 4Mb token ring networks 
too :)  So I'm not advocating for the removal of this "feature" just saying 
that anyone running any network with more than 10 hosts is not likely to be 
NFS mounting /usr!

>> RPM is designed with the misinformed notion that the user is installing to
>> *their* desktop system, *NOT* a centralized NFS server.
>
>  I think that is going too far.

I don't.  The people who designed it are programmers, not sysadmins.  I 
seriously doubt they've ever been in charge of a large network the NFS mounted 
everything from strange locations.

>  The default RPM database is for the local system, and assumes you are
>installing to it.
>
>  But you can quite easily create additional RPM databases for separate
>filesystem trees or whatever suits your application.

That is true, but also useless if I can't tell rpm to install the applications 
in a different location.

>> If it were truly designed for centralized installation, then --prefix and
>> --relocate would work on *every* pacakge.  It doesn't, because RPM wasn't
>> designed with the concept of an NFS server in mind.
>
>  The reason --prefix and --relocate don't work on every package is that most
>packages have compiled-in defaults for the locations of their files.  If you
>move the locations of those files, the programs stop working.  This isn't RPM,
>it's 99% of the software packages out there.  Open Source packages assume
>you'll recompile if you want to change a path (that's efficient!).  Closed
>Source packages are just stupid.

I understand why the *don't* work.  It's because those who maintain packages 
are essentially too lazy to check to see if they're package still works when 
it's "relocatable".  Most of the maintainers don't even set them as 
relocatable.  If RPM were truly designed for installation to a centralized 
path, the tool itself would have been designed to force package maintainers to 
configure their packages properly and allow for end-user site configuration.
It is precisely this lack of forethought that makes me prefer to install 
source tarballs for everything.  It's just a lot simpler than fighting with rpm
(or dpkg/apt, etc.).

>  Blaming RPM for this is an error.

I'm not blaming rpm.  I'm stating that rpm was not designed to be used in a 
centralized environment that spans more than a single system.  For that I 
blame the rpm developers and rpm package maintainers.

>> Yes, but how do I accomplish this if my NAS is exporting the path /nfs to
>> the world and I want to install all "centrally admin'ed pkgs" down that
>> path rather than /usr?
[...snip...]
>
>  If the package doesn't support relocation, either the packager was lazy, or
>the package itself has compile-time paths set.  Fix the appropriate problem.

That's exactly my point.  Installing the rpm db down a central location does 
me no good if I can't install my packages down the same central path.

>> It won't work.  RPM is not well designed with large scale networks in
>> mind.
>
>  Please explain how the above won't work in a large environment.

I already have, but if you insist upon an example, try this for yourself:

        get the rpm for just about any package, like openssh 2.2.0p1

        NFS mount a filesystem from some other system on /foo

        install said rpm down /foo.

It won't work.  Large environments usually don't NFS mount common paths.  If 
they do, at best, it's /usr/local.  Of course, there are hardly any rpms that 
install down /usr/local to begin with, since that damn file system standard 
insists that just about everything in the world belongs down /usr.

RPM is not flexible, it was not designed for large networks where people 
access software from paths that do not match the FSSTD, which is also not 
designed around large networks.  Both are meant for single systems.
-- 
Seeya,
Paul
----
           I'm in shape, my shape just happens to be pear!

         If you're not having fun, you're not doing it right!



**********************************************************
To unsubscribe from this list, send mail to
[EMAIL PROTECTED] with the following text in the
*body* (*not* the subject line) of the letter:
unsubscribe gnhlug
**********************************************************

Reply via email to