init managed or rc script managed services

[Marc Evans]

OK, fair enough answers. Thank you. Can I ask if your opinion changes if
the system in use was an appliance, e.g. you as the owner don't know that
Linux is on the inside unless you go out of your way to try to discover
it? In other words, no sysadmin will ever see the contents of the inittab
or rc scripts, nor be able to influance them, other then to perform a
firmware upgrade to the appliance?

- Marc

On Thu, 23 Jan 2003, Tom Buskey wrote:

>
> Marc Evans said:
> >Hello -
> >
> >I am pondering the decision logic that is used to decide what processes
> >are started through the use of inittab versus rc scripts. It seems to me
> >that several of the distros have chosen to use primarily rc scripts. From
> >a robustness of service perspective however, it is my opinion that many of
> >the services that are started by rc scripts really should be managed by
> >the init process through the /etc/inittab file. I am looking for opinions
> >on the subject.
> >
>
> >From an admin point of view, using /etc/inittab is (kindof) like using
>  /etc/rc.local.  Imagine you have to update the startup script for a
> daemon on 100 workstations.  If it's in rc.local, you'll need to ensure
> you don't stomp on other startup scripts embedded in it.  If it's in
>  /etc/init.d, it'll have its own little script, nicely isolated.
>
> Also, init.d entries usually have a nice start/stop/status mechanism.
> IMO this is the best part of a SysV vs BSD environment.  NetBSD also
> has an /etc/init.d folder instead of putting everything in rc &
> rc.local.
>
> >As I have presented this to some people, I have heard the comment that the
> >convience of being able to run "service {foo} stop" would be lost by
> >moving items from rc scripts to inittab. To counter that, I present a
> >trivial perl script that can be installed in the /etc/rc.d/init.d
> >directory which manages the entries in /etc/inittab, producing the same
> >behavior as sysadmins have experienced prior to such changes.
>
> Duplicating the effort.....
>
> >
> >You might ask, why do I care about this? Well, I have found that on rare
> >occasions, some of these services terminate, such as due to being out of
> >memory. Specifically, cron and anacron have been observed to experience
> >this, and unless the sysadmin is careful after the problem is corrected,
> >they may never get restarted. By moving these to inittab, I assert that
> >the system is more robust in its services.
>
> I'd put as little as possible into inittab (and no more so).  Then I'd
> set up a monitor for the buggy service so I can track and be alerted
> when it has problems.
>
> Once that was set up, I'd stop looking at the symptom (stopped
> services) and start looking at the problem (no enough memory, etc).
>
> In general, part of "good sysadmin practices" is to put stuff in /etc/
> init.d, not inittab.  As a sysadmin coming to a new system, that's where
> I'd look first.
>
> I've done admin on a network of BSD systems (> 400) with everything in rc.local
> and I'd rather not go back to something similar....
>
>
>
> --
> -------
> Tom Buskey
>
>