Leon Brooks <[EMAIL PROTECTED]> writes:
> Since the last (7.2) update to BIND, I've been seeing complaints in
> syslog about it being unable to write the PID file in
> /var/run/named.pid - to cut a long story short, since it's now running
> as user ``bind'' it no longer has write permission in /var/run/ and
> the killproc function in /etc/rc.d/init.d/functions has to rely on
> pidof (essentially the same as doing a killall).
>
> To get around this, may I suggest having services write their PID into
> /var/run/$SERVICE/$SERVICE.pid? - for example /var/run/named/named.pid
> -
> that way, /var/run/$SERVICE/ (/var/run/named/ in this case) can be
> owned by the user running the service. The code to add this to
> /etc/rc.d/init.d/functions is trivial.
>
> As an alternative, perhaps some processes could use
> /var/run/$SERVICE/$PID.pid to permit several of each kind of process
> to run alongside each other. The killproc function (which refers to
> the basename of the service as ${base}) need only look for
> /var/run/$SERVICE/*.pid and kill everything it finds a PIDfile for
> there.
>
> I think a special file named something like
> /var/run/$SERVICE/nokillall.pid would also be useful here to avoid
> killproc nailing services not started through the SysV scripts by
> accident if it does a killall on not finding a .pid file, unless
> killproc accepted an empty /var/run/$SERVICE/ directory as evidence
> that the service existed but was not running.
>
> How say you?
This look like a good idea for me but i think this kind of dicsussion
need to be done on the lsb mailing list see :
http://www.linuxbase.org/
--
MandrakeSoft Inc http://www.chmouel.org
--Chmouel
