rc.d cleanup patch redux
Mike Makonnen
mtm at FreeBSD.Org
Mon Oct 1 15:34:56 PDT 2007
On Mon, Oct 01, 2007 at 02:13:36PM -0700, Doug Barton wrote:
> On Sun, 30 Sep 2007, Mike Telahun Makonnen wrote:
>
> >>
> >>3. Remove the comment from named_flags, to match all the other empty
> >>_flags variables *grumble*
> >
> >I don't understand what the purpose of all those empty foo_flags=""
> >variables is.
>
> I put in a commented out example for named so that users would know that
> it's a knob which is available for them to twiddle.
>
> >Maybe for 8-CURRENT we can get rid of them from etc/defaults/rc.conf?
>
> Personally I'd rather comment them out, but I'm open to suggestions.
Well, I was more concerned with bloat of /etc/defaults/rc.conf. I would
prefer to just remove them all and replace them with a generic man page
entry, but:
mtm at terra ~% grep '^[a-z,0-9]*_flags=""' /etc/defaults/rc.conf | wc -l
33
so I guess it's not that important.
> >I'm not sure about the usefullness of these. If all the daemon needs is
> >a simple kill -TERM, then I believe init already takes care of this. A
> >script
> >should make use of the shutdown keyword only if it needs to do additional
> >processing. For example, rc.d/amd doesn't do anything special on
> >shutdown. It just lets rc.subr(8) glue send a -TERM signal. The only
> >benefit I see to adding the shutdown keyword to these kinds of scripts
> >is that the shutdown occurs in reverse order of startup (as opposed to
> >init just killing them off all at once after rc.shutdown).
>
> Yeah, that's the main benefit I had in mind. I also have a sort of gut
> feeling that doing this would be a good practice to adopt, and can lead to
> other benefits down the road, but I could be wrong.
>
> Any other opinions?
>
> >This change only adds aditional processing during shutdown without any
> >real benefit.
>
> I don't see any measurable increase in processing time, but my laptop is
> still on the fastish side.
It's just that this little chage adds a little extra processig time here,
that change adds just a little extra there, etc... and before you know it
they all add up to a big difference (especially if we don't use faststop
like des suggested).
I'm not really against this change, I just wanted to voice my reservations.
In fact, I would be more comfortable with leaving this last part of
the change until after the code-freeze since it's also likely to
introduce more "foo is not running?" console spammage.
Cheers,
Mike.
More information about the freebsd-rc
mailing list