[CentOS] CentOS 8 network-scripts

Sat Oct 5 15:40:20 UTC 2019
Valeri Galtsev <galtsev at kicp.uchicago.edu>

> On Oct 5, 2019, at 10:29 AM, Stephen John Smoogen <smooge at gmail.com> wrote:
> On Fri, 4 Oct 2019 at 18:11, Japheth Cleaver <cleaver at terabithia.org> wrote:
>> On 10/4/2019 8:17 AM, Lamar Owen wrote:
>>> On 10/4/19 10:40 AM, Valeri Galtsev wrote:
>>>> My impression is younger generation doesn't value rules that
>>>> programmers were following 2-3 decades ago. One of which is:
>>>> Do not make any changes [in the program] unless they are absolutely
>>>> necessary.
>>> I have in the past agreed with this assessment more than once. And I
>>> _am_ somewhat of an old hand at this, having run Unix and Unix-like
>>> systems for a bit over 30 years.
>>> The fact of the matter is that, even though some of the old ways work
>>> just fine and don't need to be changed, many more times I've seen
>>> that, if the old way was a kludge to begin with, maybe there really is
>>> a better way to do it.  Take the transition from horse and buggy to
>>> automobile for instance.  Iron rim tires work just great for the
>>> buggy, not so great for the automobile; a change had to be made in an
>>> old technology (the wheel) to meet the needs of the new automobile.
>>> Lots of wheelwrights probably fought that change, too.
>>> ...
>>> Today, I'm doing things with containers, virtualization, dynamic load
>>> balancing, software-defined infrastructure/IaaS, etc that the old ways
>>> simply cannot handle.  NetworkManager/systemd/etc in CentOS are far
>>> from perfect, but at least they're trying to solve the newer problems
>>> that the old ways in many cases simply cannot.
>> This is a bit orthogonal, though. (Witness the effort to remove systemd
>> requirements from containers.) An engineer is expected to understand the
>> component parts rationally to arrive at some sort of professional
>> conclusion that something is likely to work properly. This is not helped
>> by a switch from imperative and deterministic to declarative and
>> dynamic, which underlies many of the changes we've had to deal with in
>> the past decade. There is a time and place for the latter, and it's good
>> to have options available... but there are many times and places
>> (especially in the Enterprise space) where the opposite is necessary,
>> and it's FAR more reasonable to layer dynamic manipulation on top of a
>> deterministically-configured core than the other way around.
> On the other hand, most of the idea that the old config scripts were
> deterministic and imperative was built on a large amount of hacks to
> try and make it so. Having spent more time than I want dealing with
> systems which seem to be just like everything else but coming up with
> eth0 being eth4 (I am looking at you 40 Dell, HP and IBM boxes) on a
> reboot half the time.. I have come to see that a lot of scripts are
> full of race conditions and slowdowns to try and stop those race
> conditions from happening. If anything messed up from a kernel change,
> bios update, a switch update?, etc and you could be completely in the
> weeds wondering why imperative was failing. It was failing because it
> was never absolutely true.
> The problem is that as hardware rollouts have grown larger and larger
> spending time trying to figure out why 400 to 1000 out of 10,000
> systems are weird.. is too much time. You want something which will
> try to figure it out itself and do the 'right' thing.. even if it
> means that eth4 on those 400 boxes are now the main interface versus
> the eth0 on the 3600. And yes.. as an non-neurotypical person.. I find
> that incredibly infuriating.. however I have also realized that most
> businesses don't care I and others find it that way. They just want
> those 10,000 to 100,000 systems to come up and work.

And this is where big guys with thousand of boxes win, and us small guys with few dozens of boxes have to learn what works for big guys, and use it on our small scale. And no, this is not a rant, it is just realization of the reality, so I can adjust to it.

Thanks everybody, this was insightful - for me.


>> -jc
>> _______________________________________________
>> CentOS mailing list
>> CentOS at centos.org
>> https://lists.centos.org/mailman/listinfo/centos
> -- 
> Stephen J Smoogen.
> _______________________________________________
> CentOS mailing list
> CentOS at centos.org
> https://lists.centos.org/mailman/listinfo/centos

Valeri Galtsev
Sr System Administrator
Department of Astronomy and Astrophysics
Kavli Institute for Cosmological Physics
University of Chicago
Phone: 773-702-4247