Passive check versus active check

[mbrackeva]

There is some detailed info in the docs on the types of Zabbix agent items. But what is missing is some guidelines on why you would use one or the other (unless I overlooked).
If there is such a document or discussion, I would be interested to read it.

I could think of a couple things myself, but still there should be good reasons why both types were developed in the first place.

In particular, consider the following scenarion: the Zabbix server stops (planned or unplanned). That means all passive checks from that server stop running. However, the agent still runs the active checks I presume. In that case, does the agent buffer the results and send them to the server when it is back online? If yes, will the triggers work on the oder data?

[kloczek]

There is some detailed info in the docs on the types of Zabbix agent items. But what is missing is some guidelines on why you would use one or the other (unless I overlooked).
If there is such a document or discussion, I would be interested to read it.

I could think of a couple things myself, but still there should be good reasons why both types were developed in the first place.

In particular, consider the following scenarion: the Zabbix server stops (planned or unplanned). That means all passive checks from that server stop running. However, the agent still runs the active checks I presume. In that case, does the agent buffer the results and send them to the server when it is back online? If yes, will the triggers work on the oder data?

This why part of each zabbix monitoring stack (even smallest one) should have at least one proxy over which is monitored everything.

Only monitored metrics in which sampling is involved should be zabbix server are zabbix srv internal checks.
With at least one proxy and enough buffer on agents side to store up to 1min data is possible to move proxy IP/address to new system. As long as proxy holds last few hours last data is possible to schedule even few hours downtime on server without loosing monitoring data.

Above is kind of scaffolding or spine of whole zabbix HA and with this part present only central zabbix server DB backend needs to be secured by adding slave DB instance.

With at least one slave is possible:

• use slave as instance to test zabbix major upgrade (i.e. 2.4 -> 3.0)
  In critical scenario if such process may take time which will be below time window which holds proxies is possible to make such uprade using another host with running updated zabbix server processes -> connect this process to slave -> do DB layout upgrade -> unblock (for example on FW layer) access to proxies to resync last few hours data.
• do DB backups
• optimize database (and after this promote slave as new master)
• do any performance test without affecting master prod.