:: [DNG] network configuration framewo…
Top Page
Delete this message
Reply to this message
Author: Martin Steigerwald
Date:  
To: dng
Old-Topics: Re: [DNG] bookworm without systemd
Subject: [DNG] network configuration frameworks (was: bookworm without systemd)
jeremy ardley via Dng - 23.07.23, 09:33:45 CEST:
> On 23/7/23 14:26, Manfred Wassmann via Dng wrote:
> > So what is your intention? Having Devuan swich 5o systemd?
>
> It would be nice if systemd-networkd could be extracted as an
> independent package.
>
> It's not a required part of systemd and no-one has told me yet about
> any better alternative.


Whoa, this thread splits out into discussing multiple different topics.

Well that is one of my main issues with Systemd I have: It is all or
nothing (expect a few components). This monolithic approach does not
work for me. The other issue is questionable and not configurable policy
decisions by Systemd developers who think they know it better than me. A
third issue is: Its too big. I do not like to have a > 1 MiB executable
with large libraries as PID 1.

runit currently on Devuan Ceres is 22992 bytes and it only needs libc.
How sweet is that? It sounds like a software that cannot be broken
easily. A software with manageable, low complexity.

That all written, I have no idea whether systemd-networkd is any good.
Considering different distributions I'd conclude that networking
configuration in Linux distributions generally is a mess. Red Hat uses
Network Manager on servers. Are you even kidding me? SUSE thought it
would be best to invent yet another tool and called it Wicked, cause you
know, network configuration is wicked. Ubuntu developers thought: What do
we know? Let's use NetPlan that either configures ifupdown or systemd-
networkd. Yes, that is right, we are talking about a network
configuration framework that configures a network configuration framework.
Why not even adding another framework into the queue? Who cares?

For servers I still prefer ifupdown or… Python based ifupdown2 which is
used on Proxmox VE, which unfortunately uses Debian with Systemd.
However… PVE basically works, and as a hypervisor I would not make it
directly accessible to the internet. I am not happy with PVE using
Systemd, but I also do not want to spend the time at the moment to make
it work without it. ifupdown2 does not need Systemd. It allows for
safely reloading on changes with "ifreload INTERFACE" or "ifreload -a"
for all interfaces.

Maybe systemd-networkd is well done, I just don't know.

The idea of Lennart to unify between different distributions is not all
bad. I would love network configuration to work in a similar way across
(many) distributions. However the execution of the idea left a lot to be
desired. It needs discussing with one another and a good approach to
come to a conclusion that works for everyone involved. It does not need
power play. In my perception introducing Systemd involved a lot of power
play and manipulative communication. And due to that it introduced a
lot of bad design decisions into many distributions¹. Mind you, Lennart
is clearly not the most competent developer on earth. Thing is, as it
appears to me: He does not know. And that is what makes it dangerous.

[1] https://suckless.org/sucks/systemd/

Best,
--
Martin