Re: proposed minimum function of /etc/init.d scripts

[Klee Dienes <klee@mit.edu>]

Ian Jackson <ian@chiark.greenend.org.uk> writes:
> Bdale:
> > I would like to propose that all scripts placed in /etc/init.d should
> > be required to support at least the four arguments
> > 'start|stop|reload|restart'.  In cases where there is currently no
> > 'restart', it could easily be formed as the sequence of actions to
> > perform a stop followed by a start.  In cases where there is no
> > 'reload' and the daemon(s) in question can't do on-the-fly reloading
> > of config files, the 'reload' token would degenerate to an alias for
> > 'restart'.
> 
> I agree, FWIW.  Does anyone have any counterarguments ?

I definitely agree regarding 'start|stop|restart'.  However, I believe
that 'reload' should take action only if properly supported by the
system in question.  When I use the 'reload' argument, it's for the
specific purpose of making the system reload it's configuration files
without interrupting service --- if it's not available, I don't
necessarily want the system to go ahead and restart the process
anyway.

For example, the AppleShare server 'afpd' does not to my knowledge
support the 'reload' operation.  If I type '/etc/init.d/netatalk
reload', I'd like to be sure that netatalk won't try to "help" me by
kicking all the connected users off the AppleShare servers and
restarting them.

If I don't mind the process being killed and restarted, I can always
habitually type '/etc/init.d/<script> restart' and avoid the issue
entirely.

Scripts that want to be clever about restarting systems should use
'/etc/init.d/<script> reload || /etc/init.d/<script> restart'.  This
makes script-writing slightly more complicated, but I'd argue that
this will be a much less common case than the interactive usages.