On Sat, 2011-11-12 at 20:55 -0500, Christopher Browne wrote:
> 
> On Nov 12, 2011 1:54 PM, "Guillaume Lelarge" <guillaume at lelarge.info>
> wrote:
> >
> > On Wed, 2011-11-09 at 18:14 -0500, Christopher Browne wrote:
> > > [...]
> > > Does anyone feel strongly about this?  If not, then my inclination
> is
> > > to have just two behaviours:
> > >   a) Run the script on ALL nodes, as a default behaviour
> > >   b) Run on a specified list of nodes, e.g. - EXECUTE ONLY
> ON='2,3,4'
> > >
> >
> > I don't feel strongly about it, but I guess having a and b are
> enough. c
> > may be hard to code, will be cumbersome, and for no real value.
> 
> I'm most keenly concerned about the "cumbersome" part, actually.  That
> would establish c) as a clear anti-feature.
> 
> But that's assuming clumsiness.
> 
> Let's consider (loosely, I'm not consulting docs to make this up)
> 
> Execute script (event node=5, script='/tmp/Foo.slonik', only on sets
> ='1,4');
> 
> Or
> .... only on set=1);
> 
> That's neither necessarily *huge* implementation effort nor horribly
> clumsy.
> 
> Zero additional effort has its merits, but that's not enough to rule
> out the couple ideas above.
> 
> I suppose I'm "against" c) more out of not wanting to add more options
> for people to puzzle through than are necessary.
> 

More code means harder to maintain. People don't seem to want that
feature, so no need to add more complexity (even if it's low), more
code, etc, if people don't ask specifically for this feature (I'm not
going to say that this little survey gives a lot of informations from
users, but at least, it doesn't seem to attract many people).

> On reflection, there's some complication to log shipping tests, but I
> don't think that depends on whether c) gets added or not.  Rather, we
> need a test to validate that log shipping gets the DDL iff the node
> feeding log shipping was on the executor list.  No real logic
> difference whether that node was:
> I. Subscribing to the set (c), or
> II. In the "execute only on" node list. 
> 
> That's not an argument pro or con, rather a "let's not forget needful
> testing" aside. 
> 
> 
> > > Though I'm ready to argue "but if you don't know what your set of
> > > nodes are, I think you're in deep, deep trouble..."
> >
> > Exactly my thought.
> 
> So I'm not crazy, always good to know!  :-)
> 
> Glad to get feedback, thanks!
> 

You're welcome :)


-- 
Guillaume
  http://blog.guillaume.lelarge.info
  http://www.dalibo.com