I’m hitting a resource constraint with the checkers on my satellites, and I’m trying to avoid building using sub-zones. Is it possible to configure command_endpoint to use a host group?
I don’t think so but can you add a second satellite to the zone or give them more CPUs?
After 2 masters and 2 satellites per zone, I suspect you have no other choice then to add more sub-zones.
Also is it possible to install the agent on the target and thus run the checks on localhost? IMHO, this scales the best.
Thanks for the response.
Unfortunately, I’m already at 2 satellites in the zone and can’t build them up any further. The targets are all networking devices and checks are snmp, but there’s about to be over 1000 devices added, each with multiple checks. This will prove too much I think. I guess sub-zones it is then.
If you don’t mind asking a side question…
What is the resource constraint with the checkers and how did you determine that?
I would like to monitor the checkers I’ve from performance / constraints point of view as well. I do have “icinga” check and usual linux performance stats but something within Icinga json api perspective I don’t know how to co-relate each other.
Thanks
Fine grained checks of network equipment can take a lot of resources!
We do all the SNMP stuff with LibreNMS, are at ~50000 ports now and use currently 45 vCPUs for LibreNMS. I guess we soon need to move components (like the poller) to dedicated servers. Also the 700+ LibreNMS alert and health checks put a lot of strain on the Icinga masters but I added 2 more vCPUs to the masters and we’re planing to implement caching for the checks:
Sorry for the late response. Unfortunately I don’t have a pretty method, only correlating CPU time and usage of the VMs and their hosts, coupled with the host CPU architecture and affinity settings.
I ended up re-architecturing the VMs themselves to better match the hosts, as well as created 3 new zones, each with 2 satellites.