[svlug] Inits (was: On the process of picking up systemd)

Rick Moen rick at svlug.org
Fri Jan 2 15:41:33 PST 2015


Ivan Sergio Borgonovo wrote:

> OK, let me rephrase it.
> 
> sysvinit is special in the sense that it was the only one actually
> deployed at large.

Well, no.  OpenRC is default on a number of Linux distributions.  The BSD
init is default on several including Slackware.  Some implementations of the
BSD init have frequently used sequences called from a shell library, which
keeps the main init file very clean, and is particularly nice.

Advanced sparse inits like runit are often easily usable on any Linux
distributions, irrespective of whether the distro bundles them.

Of course, one should also talk about best-of-breed process service
software, as systemd has code in quite a number of functional categories,
and those include inits, service supervisors, and a number of other distinct
functions.  Good service supervisor codebases include supervisord, runit,
daemontools, and other such things.   I've extensively run supervisord
at work, and it's really good -- transparent, easy to debug, highly
reliable.  And you get to decide what processes run under its control vs.
the much simpler old-school PID system that is haphazard for run-state
enforcement but more than good enough for many things.  Thus, you retain the
freedom to easily architect the way processes run.

> The only exception I'm aware of was upstart 

No, I'm afraid you have a very limited acquaintance with inits (not to
mention service supervisors).

systemd is not best of breed in any of the categories it attempts to cover,
octopus-like.  Nor is SysVInit, of coursee, and Upstart certainly isn't, and
wouldn't be even if Canonical weren't attempting to play games with noxious
contributor licence agreements.

> Exim and postfix prosper, Courier seems to feel well, PowerDNS and
> Unbound are in good health. But they are pretty different beasts than an
> init system.

Oh dear!  My most important point, for which I mentioned those codebases,
seems to have completely eluded you.  The point I was making is that the
DJBware cult consistently compared Dan's MTA and his DNS daemon package with
only one single competitor, deliberately picking the most crufty alternative
and carefully ignoring the stronger and more modern options - such as
Postfix, Exim4, Courier-MTA, NSD, Unbound, MaraDNS, and PowerDNS.

And that is also exactly what you were doing.  Thus my point.

You would have understood this if you had taken my advice and read the
'sysvinit: the eternal red herring' section of
http://uselessd.darknedgy.net/ProSystemdAntiSystemd/.  Please do that now,
before posting more non-sequitur time-wasting arguments.  Thank you.





More information about the svlug mailing list