[nycbug-talk] Statistical Monitoring

[Steven Kreuzer]

On Nov 4, 2008, at 3:05 PM, Matt Juszczak wrote:

>> I used Ganglia at a previous job and for the most part found it to  
>> be a nice way to quickly poll several thousand hosts.
>>
>> In addition to having it report on things like cpu and memory  
>> usage, it was integrated into one of our internally developed  
>> applications which allowed us to retrieve statistics specific to  
>> that application in a more efficient manner.
>>
>> Previously all the statistics were being exported by a tiny built  
>> in web server and then having a script make http get requests to  
>> each server. It did not scale too well.
>>
>> My only complaint with ganglia is that after running it for a  
>> while, it would stop responding to requests and require the service  
>> to be restarted.
>
> Your experience with Ganglia has been simply as a push agent,  
> correct? You can collect statistics via any method you choose and  
> just push to gmond?  I don't believe ganglia has a "pull" method  
> where it can combine querying snmp as well, does it?

We had each server running gmond, collecting information from the host  
and ganglia aware applications. gmond makes this information available  
by listening on a multicast channel and everyone once and a while a  
central server would poll this channel and archive the XML that every  
server responded with.

Its a pull method, but you only have to make one request to retrieve  
statistics from every machine in that broadcast domain as opposed to  
querying each and every host.

Steven Kreuzer
http://www.exit2shell.com/~skreuzer