Re: Objections to #156161?

[Thomas Hood <jdthood@yahoo.co.uk>]

tags 156161 wontfix
thanks

On Mon, 2003-08-25 at 17:43, Goswin von Brederlow wrote:
> The existence of starts scripts alone is sufficient to get the right
> behaviour:
> 
> old level | new level | affect
> ----------+-----------+--------
>           |           | no change (if it runs don't kill it)
> has start |           | stop service
>           | has start | start service
> has start | has start | no change (if it doesn't run, don't start it)

It doesn't actually suffice because the K symlinks also control
the order in which services are stopped.  But I get your point.
You have shown that one can put a service into "manually controlled"
mode simply by setting all its symlinks to K.  Then switching among
runlevels never starts or stops the service.  The presence of a K
symlink really signifies the disjunction: either the service is not
running or it was started manually. Implementing "manually controlled"
mode this way works fine until you do something (such as a package
restart on upgrade) that resets the service to the state corresponding
to the symlink.  If we had try-restart implemented everywhere then
we wouldn't need a third state (besides S and K) to inhibit
inappropriate restarts.  And even if we have the third state, we
still need try-restart in order to restart services that have been
started manually.  So you have convinced me that we don't need this
third state and invoke-rc.d doesn't need to be changed; we should
simply try to get try-restart implemented.

--
Thomas