[lug] Thoughts on upgrading to CentOS 7

[Steve Litt]

On Wed, 28 Mar 2018 07:21:09 -0600
Rob Nagler <nagler at bivio.biz> wrote:

> On Tue, Mar 27, 2018 at 11:39 PM, Steve Litt wrote:
> 
> > > What are the poison pills?  
> >
> > Too numerous to describe, and most I never troubleshot down to the
> > bitter end root cause.
> >  
> 
> Your article is very interesting. Thanks for providing it.
> 
> I still don't see the "poison pills". I think sysv is extremely
> difficult to replace. 

My experience in late 2014 was that systemd was an order of magnitude
harder to replace than was sysvinit. Given that systemd has welded
itself tighter to software at all levels, from the initramfs to Gnome,
I'd imagine it's much harder today. The only way I can be more
persuasive is if you try replacing both with runit.

> CentOS 7 still has init scripts. From what I've
> seen of systemd, I don't see it as being difficult to replace. 

I had a lot of difficulty.

> I can
> see how it is very difficult to unhook systemd from a particular
> installation of a distro. 

Yeah, that's what I was saying. Once a distro has systemd, swapping in
runit or s6 becomes extremely difficult. I'm not sure what other
context one could view it from.

Imagine that installing Emacs not only disables Vim, but makes so many
changes to the surrounding software that Vim installation becomes a
long series of workarounds instead of apt-get install vim. That's how I
perceive the situation once systemd is on a distro. My perception is
based on actually doing it. One could say "all you need to do is get
rid of gnome, change all the dbus commands, refurbish initramfs to get
rid of systemd specific stuff, rewrite the shutdown script, back out
systemd networking (and doesn't systemd have their own dns now, or did
I dream that?), and completely revise your relationship with polkit.
But when you actually try to implement that seemingly doable todo list,
you're met with the death of a thousand cuts, and the todo list
eventually grows to probably 40 items. Get one wrong and, whoops.

> I don't see that as a poison pill for the
> distro, though. That's some work, but nothing that makes it
> impossible to move away from.

If the criteria is "possible", the entire discussion is moot. Given
enough resources, I could transform CPM into Linux. The point is it
would take a very knowledgeable person a day to switch Red Hat to
runit. It would take a very motivated average Linux power user a month.
These timings have a chilling effect on switching. I call that a poison
pill. 

Continuing the concept of "possible" as a criteria, consider that half
of systemd's raison d'être is to keep the command prompt phobic
snowflakes continuously in drag and drop, the gap between replacement's
"possible" and sans-systemd's "easy" (I think most on this list
consider elementary commands and shellscripts to be easy) shines a
light on the situation. Systemd introduces massive monolithic
entanglement to relieve us the need of shellscripts and the like.

https://www.youtube.com/watch?v=_ERAXJj142o#t=1471s

 
SteveT

Steve Litt 
April 2018 featured book: Troubleshooting Techniques
     of the Successful Technologist
http://www.troubleshooters.com/techniques