High zabbix queue

[ervinromel]

Hi all
i have a zabbix 5.0 installed and everything is working fine in normal conditions. when i disable multiple nodes together there is a huge queue on server and after some times it will become normal. i changed the startpoller parameter from 30 to 600 and it cause that the time i have queue become much shorter. i wonder why this happen. is it normal behavior that when you miss multiple nodes together because of network outage or everything , the busy poller process become 100%??
thanks,

[Colttt]

I strongly recommend that you reduce the poller back to ~15, if you're running so many pollers, race condition and other side effects can happen. I also recommend to use Zabbix version 6

[ervinromel]

do you have any idea of what is going on behind the scene of this?
i will reduce the pollers, but my problem exists. after increasing number of pollers my queue time became too much less.
thanks,

[Markku]

You haven't said it but my understanding from the case is that you are using passive agent items. Is that correct?

Passive items mean that Zabbix server (or Zabbix proxy) need to consume CPU and other resources on polling the item values from the hosts. In case the agent (on the host) is not reachable, it will lead to the server/proxy pollers having to deal with unreachable hosts, retries and other extra work.

On the other hand, if you would use active agent items, the hosts are themselves responsible for connecting and sending the data to the server/proxy. Thus the server/proxy does not need to consume resources on retries.

Using active agents instead of passive agents is one of the basic scaling features of Zabbix implementations.

If, for network connectivity reasons, you are forced to poll the agents, another scaling feature is to use Zabbix proxies: instead of having the server to poll the (unreachable) hosts, the tasks are delegated to Zabbix proxies, and then the proxies will deliver the collected data to the Zabbix server. Thus, you can implement the Zabbix proxies on a separate platform that has proper amount of CPU and other resources to deal with the problems with polling the unreachable hosts. And, maybe you can even implement the Zabbix proxies in places where the hosts can actively connect to, thus using active items on hosts.

Zabbix Components and Ports: https://majornetwork.net/2017/10/zab...nts-and-ports/

Markku

[ervinromel]

you know the nodes that we are shutting down them are snmp nodes only. is it normal that you shutdown multiple snmp nodes then you have huge queue on your zabbix? i do not know how to interpret this.

[Markku]

https://www.zabbix.com/documentation...stration/queue says:

Ideally, when you open this section it should all be "green" meaning no items in the queue. If all items are updated without delay, there are none waiting. However, due to lacking server performance, connection problems or problems with agents, some items may get delayed and the information is displayed in this section. For more details, see the Queue section.

Markku