I do not know if this is feasible. But it may be a good feature for scalability of Zabbix.

The zabbix-agent currently does very little processing before sending items to zabbix server. This can be improved by running preprocessing on the agent side.

Example situation:

Monitoring a host using prometheus metrics with `web.page.get` with active agent. The metrics data is 4MBytes in size. This is done by a master item and dependent items. Let's say we are monitoring a fraction of the items in the 4MBytes text document.

Currently zabbix-agent fetches the document and sends to zabbix-server. The zabbix-server extracts the dependent items and discards using preprocessing settings etc.

Wouldn't it be useful if the zabbix-agent could process the dependent items and do not send the original master item to zabbix-server if no hisory/trend period is defined? On top of it, it can take advantage of the discard/threshold functionality and further reduce the data transmission size. Data compression would be a plus...

It is currently possible to use zabbix-proxy to achieve this end result.  However it is cumbersome to use a proxy with every zabbix-agent obviously.

Could it be possible to somehow leverage the existing proxy code to make proxy run in a smaller context, only the cache local values for the same host without needing to setup proxy in zabbix server for every host? OR maybe such caching and processing features can be added to zabbix agent?

Thanks!