Network availability systemd dependency failure at boot

[Sam Varshavchik]

Tom Horsley writes:

> On Sat, 05 Jul 2014 08:13:45 -0400
> Sam Varshavchik wrote:
>
> > Everything was always broken
>
> I'm pretty sure everything was always broken.
> I never had the combination of postfix, dovecot,
> and stunnel operational more than about 10% of
> the time with pure systemd.

For me, everything seemed to work fairly reliably until about a week ago,  
which, coincidentally, is when the last systemd update got pushed out. I'm  
pretty sure that entropy is in full control here; for a particular system,  
depending on which services it has installed and enabled, determining the  
order in which systemd will kick things off, and, together with some  
additional entropy-driven factors, like how long each start script runs,  
before it stops and systemd kicks off the next one – that determines, in  
the end, whether things will get started in the order that results in a  
working server, or a hopelessly bolloxed server. And, some minor change in  
systemd's internal logic made my entropy effect change from "always happens  
to work" to "always happens not to work". Everything was fine until a week  
or so, ago, maybe even a little longer – perhaps the previous, rather than  
the most recent systemd update. I don't recall. Everything was hunky dory,  
and now:

named only listens on 127.0.0.1, at boot
innd only listens on 127.0.0.1, at boot
httpd gives up and dies
dhcpd gives up and dies (lots of fun on my LAN)
privoxy gives up and dies (more fun)

Perhaps, technically, it's all these packages fault, for not installing the  
correct service file. Still, I'll point the finger at systemd. It's a direct  
consequence of its buzzword-compliant, but fundamentally broken design.

You could look at the old rc?.d directory, and at a glance see exactly what  
gets started at system boot, and in which order. Now, it's a big mystery,  
shrouded in a dark cloud. One of my servers does not use plymouth, and has  
boot messages turned on, and the number of errors spewed by systemd at boot  
time is quite impressive. The number of packages that have a systemd file  
that's broken in some way is staggering. So, is it all their fault, or  
systemd's?

> That at least works up to the day systemd decides
> no one needs rc.local and they drop support for
> it (a day that is sure to come :-).

You can bet on it.

Actually, I think we do have a sliver of hope, now that systemd has infested  
RHEL. As RHEL 7 rolls out, complete with the systemd clusterfrak, it's now  
going to get some exposure to Red Hat's paying customers. Expect some noise  
to slowly increase, in volume, over time. So, as RHEL 7 roll-out progresses  
over time, I think that it's going to end up dooming systemd. It's just a  
matter of time. I'm going to wager a 100 quatloos that we'll start hearing  
some talk about replacing systemd with something that actually works in,  
maybe, 3-4 years' time.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.fedoraproject.org/pipermail/users/attachments/20140705/f53b8932/attachment.sig>