On Wed, Nov 5, 2014 at 9:57 AM, NICOLAS DUPEUX <nicolas.dup...@arkea.com> wrote:
> > >[...] > > Do you remember my POC with elasticsearch/livestatus ? > At configuration reload, we can do the linking on a new index and when > it's ready swap old and new index. Of course, users will not get update > while configuration reload but at least, they will have answer to their > livestatus queries. > > And, as a bonus, it solve livestatus performance issues. > > Regards, > Nicolas > > Hi, switching from memory backend to a ES one can be a solution. It can solve several problems in one move: * ES already do CPU threading (thanks java :) ). * will do the job for log backend... :) * can be more reactive for new objects for the last point, the idea can be to do the insert not on the broker, but on the arbiter side, just after it load and validate a configuration (in a thread for example in order to not lock the monitoring part). This module will have to insert new objects (hosts/services/commands/timeperiods) and only update old ones (remember that the arbiter do not know about states like OK/WARNING/CRITICAL, only PENDING, the retention is scheduler based). Of course it introduce a new dependencies, but for very large env it an be a solution. At least we should try and bench this solution with real data (so all hosts/services fields. Jean
------------------------------------------------------------------------------
_______________________________________________ Shinken-devel mailing list Shinken-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/shinken-devel
