Nick,
Well I'm glad its not just something I'm doing wrong
- thanks for the explanation. I'm not desperate for a
fix, just wanted to do some local development work on
the frontend. I'm looking forward to testing out
early versions including the libapr changes. For now
VMWare w/Ubuntu runs gmetad just fine on a MBP.
Thanks,
Gil
--- Nick Galbreath <[EMAIL PROTECTED]> wrote:
> yes, I've seen this on a mac too.
>
> The gmetad code is somewhat odd in that is uses
> "poll" one file descriptor
> at time, one thread at a time. which is not how it's
> normally used. I
> _THINK_ this was done to do read timeouts on the
> socket, however this can
> now be done safely with libapr. (it's really just a
> wrapper around
> setsockopt). on the mac, it appears there is a race
> condition where poll
> signals "yes I have data to read" and "yes the
> client disconnected" when
> gmetad and gmond are on the same box.
>
> Also the gmetad code is inefficient in that it
> reads 1K blocks at a time,
> most sockets have 4-8k buffers, and the typically
> gmond xml is _quite_
> large. This might aggravate the problem on the mac.
>
> This can all be cleaned up, but I'd need the
> go-ahread the ganglia crew (to
> convert native calls to use libapr).
>
> I think I might have a quick patch laying around
> (which just undoes the
> poll, which isn't needed) so it works on mac. If
> you are desperate let me
> know!
>
> --nickg
>
>
> On 5/9/07, Gilad Raphaelli <[EMAIL PROTECTED]>
> wrote:
> >
> > On a seperate note, are you running gmetad on a
> mac?
> > I've been unable to get it to run on an intel mac,
> > debug reports:
> >
> > The remote machine closed connection for
> [localhost]
> > data source after 1023 bytes read
> >
> > It usually reports 1023 or 2046 bytes but
> occasionally
> > it actually gets the full 3069 bytes or other non
> > 2^n-n number but never actually writes any rrds.
> This
> > is 3.0.4 built with Apple's default gcc4.
> >
> > Anyone else have any pointers?
> >
> > Gilad
> >
> > --- Matthew Chambers
> <[EMAIL PROTECTED]>
> > wrote:
> >
> > > Hi all,
> > >
> > >
> > >
> > > I recently started using Ganglia on our Mac Mini
> > > cluster and I noticed that
> > > the graphs are resized from their native size so
> > > that they fit the
> > > templates. This didn't make a lot of sense to
> me,
> > > so I edited the system so
> > > that the images returned by graph.php calls are
> used
> > > as-is. I added two
> > > variables "sx" and "sy" which are passed in the
> > > graph.php call specifying
> > > the desired width and height of the output
> image.
> > > This approach means that,
> > > at least for the image calls which are fully
> > > specified from the templates,
> > > the dimensions can be specified in the template
> and
> > > needn't be worried about
> > > anywhere else. When I made this modification I
> > > discovered that the width
> > > and height that rrdgraph takes are actually the
> > > dimensions of the canvas
> > > (the inside box that the actual graph is plotted
> > > on), and they don't include
> > > the outer box. Rrdgraph does provide a way to
> get
> > > the final image's
> > > dimensions (--imginfo), but it is not possible
> to
> > > use it while also using
> > > the standard output mode (which is what the
> > > graph.php call does). So I
> > > added a second call in graph.php which writes to
> > > /dev/null and simply parses
> > > the output dimensions and adjusts the dimensions
> > > given to the second call by
> > > the amount necessary to make the final
> dimensions
> > > match the "sx" and "sy"
> > > variables. Since the "z" variable is obsolete
> but
> > > was still used to change
> > > the graph style slightly for the host graphs, I
> > > changed it to a "formtype"
> > > variable ("f") with values of "compact" or
> "normal."
> > > Oh, I also changed the
> > > images to PNG instead of GIF since the rrdtool
> > > documentation says that PNGs
> > > are generated as fast or faster and are as small
> or
> > > smaller than GIFs.
> > > Here's the code:
> > >
> > >
> > >
> > > #
> > >
> > > # Generate the rrdtool graph command.
> > >
> > > #
> > >
> > > if ($formtype != "compact")
> > >
> > > {
> > >
> > > $command = RRDTOOL . " graph
> /dev/null
> > > --start $start --end $end
> > > ".
> > >
> > > "--width $width --height $height
> > > $upper_limit $lower_limit ".
> > >
> > > "--title '$title' $vertical_label
> > > $extras $background ".
> > >
> > > " --imgformat 'PNG' --imginfo
> > > '%s\n%lu %lu' ".
> > >
> > > $series;
> > >
> > >
> > >
> > > $imginfo = exec($command);
> > >
> > > $dims = explode(" ", $imginfo);
> > >
> > > $width = $width - ($dims[0]-$width);
> > >
> > > $height = $height -
> ($dims[1]-$height);
> > >
> > > }
> > >
> > >
> > >
> > > $command = RRDTOOL . " graph - --start $start
> --end
> > > $end ".
> > >
> > > "--width $width --height $height $upper_limit
> > > $lower_limit ".
> > >
> > > "--title '$title' $vertical_label $extras
> > > $background ".
> > >
> > > #"--imgformat 'PNG' ".
> > >
> > > "--step 5 ".
> > >
> > > $series;
> > >
> > >
> > >
> > >
> > >
> > > There are a few problems with this approach: 1)
> > > /dev/null is platform
> > > specific (are there other parts of Ganglia that
> are
> > > already platform
> > > specific so that this isn't such an issue?); 2)
> if
> > > the layout of the graph's
> > > legend changes when the dimensions are adjusted,
> the
> > > adjustments won't be
> > > correct, and I'm not sure what the best approach
> is
> > > to fix this. Let me
> > > know what you guys think about this approach.
> If
> > > it's satisfactory, I'll
> > > submit the changes to the repository.
> > >
> > >
> > >
> > > Also, has there been any discussion about using
> SVG
> > > instead of raster
> > > formats for the graph output? It would
> introduce
> > > serious user agent
> > > compatibility considerations, but it would
> provide
> > > quite a lot of potential
> > > benefits (interactivity, anyone?). Rrdtool's
> SVG
> > > output would need to be
> > > improved, though.
> > >
> > >
> > >
> > > Thanks,
> > >
> > > Matt Chambers
> > >
> > > >
> >
>
-------------------------------------------------------------------------
> > > This SF.net email is sponsored by DB2 Express
> > > Download DB2 Express C - the FREE version of DB2
> > > express and take
> > > control of your XML. No limits. Just data. Click
> to
> > > get it now.
> > > http://sourceforge.net/powerbar/db2/>
> > _______________________________________________
> > > Ganglia-developers mailing list
> > > [email protected]
> > >
> >
>
https://lists.sourceforge.net/lists/listinfo/ganglia-developers
> > >
> >
> >
> > __________________________________________________
> > Do You Yahoo!?
> > Tired of spam? Yahoo! Mail has the best spam
> protection around
> > http://mail.yahoo.com
> >
> >
>
-------------------------------------------------------------------------
> > This SF.net email is sponsored by DB2 Express
> > Download DB2 Express C - the FREE version of DB2
> express and take
> > control of your XML. No limits. Just data. Click
> to get it now.
> > http://sourceforge.net/powerbar/db2/
> > _______________________________________________
> > Ganglia-developers mailing list
> > [email protected]
> >
>
https://lists.sourceforge.net/lists/listinfo/ganglia-developers
> >
> >
-------------------------------------------------------------------------
> This SF.net email is sponsored by DB2 Express
> Download DB2 Express C - the FREE version of DB2
> express and take
> control of your XML. No limits. Just data. Click to
> get it now.
> http://sourceforge.net/powerbar/db2/>
_______________________________________________
> Ganglia-developers mailing list
> [email protected]
>
https://lists.sourceforge.net/lists/listinfo/ganglia-developers
>
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
_______________________________________________
Ganglia-developers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/ganglia-developers
