Load Balancing?

[cronuscronus]

Hello. I am planning on rolling Zabbix out to 1500 proxies each of which query 120 items every 10 seconds.

This means the Zabbix server will be processing about 18,000 messages per second.

I was wondering if it is possible to have multiple Zabbix servers use a single database?

My goal is to use a VIP (virtual IP) and have several Zabbix servers behind the VIP to handle the load coming from the proxies.

The DB is covered. I'm using an Oracle RAC with plenty of nodes and IO bandwidth.

Is it possible to just have identical Zabbix servers all connecting to the same database host?

Thanks for the feedback.

[nelsonab]

No this is not possible without some hacking. However with a little hacking I don't see it as far from possible. Three key areas are the housekeeper and the poller and the DN code. These processes run periodically and query the database and take action as appropriate. The poller may however prove to be somewhat challenging as it queries the db periodically and then opens a connection to a server to request a specific data point. You don't want two cluster nodes asking the same server for the same data point twice.

However your biggest challenge may lie with the DN code. I haven't dug too deeply into it but I have a hunch there are some contention issues. If you have a simple Master and two child node setup when the master is updating one of the child nodes it is blocking for all data from the other child. In a way this is necessary due to the hash/checksum nature of the updates and checkpointing.

Overall there are two main ways to accomplish what you want. One is hack it yourself and hope it makes its way into mainline, or two talk to [email protected] and discuss sponsoring this change.

I would very much encourage you to pursue this as it's a feature many people have been asking for and all that's really needed is for someone to take the reins and push for it.

[cronuscronus]

10-4

Thanks for the information. I will let you know what I find.

[cronuscronus]

Nodes

I think what I'm going to do is look into nodes. I can split my locations up and have my proxies connect to different nodes, I just need to set it up statically.

I can then have the nodes forward alert data only to my master node. That way historical data storage, proxy processing etc... is "load balanced", manually, and we still have a central system to catch alerts.

Thoughts on a setup like this?