Re: [munin-users] hybrid graphing

[Matthew W. <mw...@zy...>]

I'll go ahead and get this done later tonight and get you the information.

Thanks,

-- 

Matt West



> From: Nicolai Langfeldt <ja...@re...>
> Date: Thu, 28 Oct 2010 13:36:50 -0700
> To: Matthew West <mw...@zy...>
> Cc: "mun...@li..." <mun...@li...>
> Subject: Re: [munin-users] hybrid graphing
> 
> On 28. okt. 2010 22:20, Matthew West wrote:
>> Lets just go with large at this point, over 500. I do have a couple of
>> multigraph enabled plugins on some select hosts, but that accounts for about
>> .25% of my total graphs created. To say I have 5000 - 10000 services being
>> "graphed" on this one machine wouldn't be crazy to say ;)
>> 
>> As I said, the revision 3901 works perfectly for me with fastcgi setup. And
> 
> The odd thing is that 3901 should be in all ways slower.  To start with
> it reads the datafile twice, then it mucks about with it making deep
> clones for each graph, and so on and on.  Lots of overhead that was
> removed or reduced in 3965.  If there is a 30 second timeout it should
> hit 3901 too.  (I should note that 3965 is unstable the fast-cgi
> processes need to be restarted if it/one of them crashes.  We've found
> that some bad error handling causes the instability, patch is comming up).
> 
>> I can clearly see the advantages over normal cgi graphing when compared to
>> my munin 1.2.x machine. However, making graphing calls even faster would
>> definitely be appreciated ;)
> 
> With your hooge site can you do us a big favour.  If you make a copy of
> munin-cgi-graph installed from the latest trunk, on the line after
> 
>    my $config = graph_startup(\@params);
> 
> around line 63 insert a "exit 0;".  This will let us look at the startup
> overhead only.  Then, as munin, run
> 
>    time munin-cgi-graph-copy
> 
> just to see how long the startup time is.  Then run
> 
>    perl -d:NYTProf munin-cgi-graph-copy
> 
>    nytprofhtml
> 
> And email me a tar file of the html hierarchy that last command
> generated.  I'm very curious to see the profile.
> 
> Thanks,
>    Nicolai