Re: [Ganglia-general] Ganglia 4.x architecture planning

[Vladimir Vuksan]

We may want to clarify what services from Mongo are we gonna use. Is the intention to use it as a key/value store or use it for things like capped collections (http://docs.mongodb.org/manual/tutorial/use-capped-collections-for-fast-writes-and-reads/). Currently we do have support for memcached. I actually use memcached for most of our alerting at this point instead of querying gmetad. Vladimir On 03/28/2014 09:51 AM, Daniel Pocock wrote: On 28/03/14 14:39, Aaron Nichols wrote: On Fri, Mar 28, 2014 at 1:45 AM, Maxime Brugidou <maxime.brugi...@gmail.com> wrote: I don't understand why all this is necessary. I strongly disagree with the "horizontal scalability" of mongoDB ( i run a very large mongodb cluster in production in addition to other databases) and would rather suggest a pluggable backend with a simpler default (like text files or maybe postgresql or mysql). I'll just 2nd this opinion. My immediate reaction to the addition of mongo is rejection - my experience with it at scale has been awful & on a small scale it always feels like overkill. Further, since the data being served seems like a relatively small dataset it's unclear to me why there cannot be more options. For our use case (which is That has already been discussed in the thread - are you OK with the plugin idea? ~180k metrics) an in-memory datastore for metadata should be more than sufficient assuming it can be re-hydrated from some persistent source which could simply be waiting for a polling interval of gmetad. I have no need for the addition of nagios / rsyslog events.  Knowing you have 180k metrics is only one factor How many users (both humans and processes) want to view your data in real-time?  Is it just used by sysadmins, for example, or does everybody in every development or support team access Ganglia? I've never looked at Ganglia as a horizontally scalable system and accepted that because it made things simple. If I need to scale it out I need to partition my grids / clusters.  Agreed - that has always worked and it would continue to work that way too.  This is also good for people who have multiple small sites, etc.

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

_______________________________________________
Ganglia-general mailing list
Ganglia-general@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ganglia-general