>>> On 7/10/2007 at 7:29 PM, in message
<[EMAIL PROTECTED]>, "Bernard Li"
<[EMAIL PROTECTED]> wrote:
> Brad:
> 
> I got some questions regarding the python support module.
> 
> Does it currently only support Python 2.4?  Would it work with < 2.4 &
>> 2.4?  I am currently developing on CentOS 4 and it only comes with
> 2.3.4.  I do have Python 2.5.1 installed some place else so if there
> is a way for me to specify the alternate location of Python libraries
> that might help.

I don't think that there is anything in the code itself that requires python 
2.4.  The python module is fairly straight forward.  Version 2.4 was just the 
version that we had to work with and I wasn't sure how to fix up configure so 
that it would allow multiple version.  The spec file would also have to be 
fixed up to allow it as well.

> 
> How does this implementation compare with Nick G's work?
> 
> http://www.mail-archive.com/[email protected]/msg02865.
> html
> 

Giving his implementation a quick glance, the main difference is in-process vs 
out-of-process.  Any implementation of gmetric still has to be scheduled 
outside of gmond which means that your metrics are being configured in two 
different places.  That's not a bad thing because the advantage of gmetric is 
that you can easily add new metrics to a running gmond since it doesn't require 
any kind of configuration change.  However you do have the overhead of being 
out-of-process.  

Both the DSO and python methods for building metric modules allow you to do 
some things that I believe, weren't really available before.  Primarily the 
ability to create dynamic metrics that are based on hardware, service or 
application discovery.  For example, the multicpu DSO module and the multidisk 
python module are both able to discover the number of cpus or disks on the box 
during init time and then represent each cpu or disk as an individual metric 
that can be included in a collection group just like any other metric.  There 
is also no ramp up or tear down overhead because everything is run in-process.  
In addition, both the DSO and python modules can maintain state since they are 
never unloaded and reloaded.  

Bottomline is that it comes down to which method servers your particular needs 
better.  I see advantages to both methods.

> I am trying to build Ganglia SVN trunk under CentOS 4 to try out the
> DSO stuff -- I got around the hurdles of apr-1 and libconfuse, but now
> am somewhat stuck with the Python 2.4 requirement.  BTW, is it
> possible to build trunk using libconfuse from srclib/ (but libapr is
> linked dynamically?)

I'm sure that there is a way to do it, I'm just not an autotools guru so the 
problem is really how to fix up the configure script to allow you to specify 
where to build from.  Since libconfuse doesn't ship on SuSE either, I made a 
few fixes to the libconfuse spec file which allows me to easily build a 
libconfuse RPM directly from the tarball.  Once I build the RPM, the rest is 
actually simpler and the whole build process for ganglia goes a lot quicker.  

Brad


-------------------------------------------------------------------------
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

Reply via email to