[TriLUG] what happened to static-routes ?

[Aaron S. Joyner]

Oh my.  This has the smackings of a distribution flame war written all 
over it.  Quite an innocuous way to start one, too.  Kudos Ryan.

The basic, non-partisan answer, goes something like this.  Distributions 
evolve over time.  Their goal is often to allow you easy package 
management, simple configuration, and good integration, for a myriad of 
tools which have little or no association with each other.  In order to 
accomplish this, they attempt to abstract things into manageable chunks 
which are scripting-friendly.  This philosiphy applies equally well to 
the base system, as to individual package config file setup and 
placement.  Consider the following thought experiment.

You're going to install package Foo, a control package for a network 
appliance of some sort.  In the installer / configurator / defaults, it 
has been requested that Foo:
1) Start at boot up
2) Add an entry to the Gnome Applications menu
3) Create an alternate interface on your primary network card
4) Create a static route to it's remote device on that network card

In order to accomplish this in a minimilast distro (say, Slackware, for 
example), the package would probably provide short instructions at the 
end of the installation, or perhaps in the INSTALL file, describing what 
changes should be made where, in order for the package to function.  In 
more full-service distros, ala RedHat, Mandrake, Debian, etc it would be 
preferred for those actions to happen automagically.  To automate them 
with the way Slack has things laid out would require very cumbersome and 
error-prone scripted parsing and editing of configuartion files.  If you 
recall back to the `linuxconf` days, this is very error prone.  If done 
poorly, it can cause all kinds of trouble, ranging from blowing up your 
entire configuration file, stripping all of the comments, or dropping 
just that important setting you made yesterday.

So, how best to solve this problem?  One way (the traditional RedHat 
way) is to break the configuration files down into configuration 
*directories*, which are parsed by a configuration file.  The config 
file simply says, look into that directory over there, and parse every 
file as config entries in order.  This allows automated systems to 
easily add reliably and remove bits of the configuration, with out fear 
of messing up other aspects of the config.  It's a very simple, elegant 
solution to an otherwise rather complicated problem.

As you pointed out, distributions even vary internally over time.  RH 
has gradually broken apart more and more things, in their continued 
quest to provide better automated configuration.  With the current setup 
in RHAS, you can add your static routes to one interface, and have them 
come and go with that interface.  A good example is a firewall, with a 
dedicated dialup to a remote location for emergency connectivity 
situations.  Normal routing to RemoteLocationA goes over the "big 
pipe".  Bring up that ppp0 interface via the simple "ifup ppp0", and up 
with it comes the static route to RemoteLocationA over the dialup link - 
nothing behind the firewall needs to change, and it works intelligently 
as it should.

Hopefully that's a thorough explanation that doesn't bore anyone too 
much.  :)  As Mike M. pointed out (while I was typing this up), much 
good discussion (and many heated arguments) can be had over topics like 
this.  I often end up debating amongst my close circle of friends the 
merits and disadvantages of these configuration styles, though so often 
I'm just "out numbered".  :)  People often prefer one style to the other 
not for maintenance or configuration reasons, but simply because it's 
what they're comfortable with.

Aaron S. Joyner


Ryan Leathers wrote:

>Hi all.  I have wondered from time to time why configuration files are
>located in one place or another, and why this location isn't more
>consistent between ditributions.  
>
>Here is an example.  In Redhat 9 static routes are stored in the file
>aptly named static-routes found under /etc/sysconfig.  In RHAS 3 this
>same information is stored under /etc/sysconfig/networking/devices in an
>interface specific file like eth0.route
>
>I can see the benefit of associating static routes to specific
>interfaces, but why not just do that in one file?  This isn't a big
>deal, but if anyone knows WHY this was done I'd enjoy hearing the tale. 
>I'm also just generally curious about how these kinds of decisions come
>to be.
> 
>
>
>
>
>  
>