Best Practices for Proxies

[technikolor]

In several of the threads I've been a little surprised by how many proxies people are using. I'd like to start to build some common knowledge around:

A) What is the best way to monitor VPS per proxy?
B) What metric should be used to determine that you don't have enough proxies?
C) Do Active checks help or hurt large deployments?

I've got 300 nodes running almost nothing but active checks on a single proxy, on a box that should have no problems keeping up, but the queue on the master is always completely out of control. I'm happy to add more proxies, but I'd like some basic rules to help me determine how many and how to spread them.

[Alexei]

Just a small note. Since proxies do not send information about unsupported items to the server, the queue numbers may be a little bit misleading. Check Queue->Details to see if this the case.

[technikolor]

Absolutely, I exclude those.

Never the less, I'm dealing with situations where I'm completely killing the proxy and/or server. I've scaled back my installation from over 600 to a third of that until I get things under control.

* Monitored hosts: 239
* Items: 13535
* Triggers: 8989
* Required NVPS: 116.7
* Items queued more than 10 minutes: 13236

Just completely wedged up somewhere. Perhaps instead of 1 proxy (I was planning 1 per physical location) I should have 1 per 100 hosts, or even more?

[richlv]

i'd suggest figuring out where the bottleneck is first.

monitor iowait on zabbix server & proxies, monitor busy processes on the server (available since 1.8.5), doublecheck which items are missing data (as per alexei's suggestion).