Thanks for the pictures, they helped me visualize the problem.

I agree with you, even in non-scalable mode, gmetad should output a GRID tag. 
I have made the change to server.c, and initial tests here look good. 

I have committed the gmetad changes to CVS. No changes were made to the 
webfrontend.

-federico

On Thursday 11 September 2003 10:02, Jason A. Smith wrote:
> I have a single computer running gmetad, collecting data from 10
> separate clusters.  The webfrontend is running on that same computer,
> like this:
>
> webfrontend  -->  gmetad (not scalable)  -->  10 separate clusters
>
> I noticed the new scalable feature and tried it out when I saw the
> problem.  I have attached two webfrontend snapshots, the one called
> scalable is the normal old view.  If all I do is change the scalable
> option to off and restart gmetad, I get the second picture, where you
> can see that the total grid plots are now fourth from the top and don't
> have the grid name there anymore.
>
> That is why I suggested that the GRID tag should always be present in
> the xml output from gmetad, but ignore it on input, so the grid name is
> always present, for the web-frontend and other gmetads to use if they
> like.  It is up to the higher level gmetad to decide if it will be
> scalable or not and ignore the grid tag and its authority location.  The
> source gmetad can only offer the scalable option but the higher gmetad
> can choose to ignore it and reproduce all rrds locally.  Of course, it
> would also be nice if the source gmetad could report its grid name for a
> nicer tree layout, but somehow let the higher level gmetad know that it
> is not available for authority referrals, maybe because of firewall or
> other security issues.
>
> ~Jason
>
> On Wed, 2003-09-10 at 14:47, Federico Sacerdoti wrote:
> > Could you take a screen snapshot and send it to me?
> >
> > Just so I understand, you have a setup something like this:
> >
> > gmeta (not scalable) ------> gmeta (not scalable) ----> cluster1

Reply via email to