Low level discovery items that are automatically generated seem to ignore SNMP timeouts as specified in /usr/local/etc/zabbix_server.conf

Building SNMP items "the old fashioned way" (meaning, building SNMP items one by one by hand with their associated SNMP OID) seems to work fine, which seems to indicate that possibly the new LLD poller code ignores the timeout value.

This leads to NETWORK_ERROR and "Timeout while connecting to <IP>:161" errors in the web frontend and in the debug logs, and so eventually the host itself gets disabled due to these errors. Graphs become "spotty" as the host "flaps" in and out, from monitored to unmonitored. snmpwalks against the device (with '-t 3 specific in the snmpwalk command) work perfectly fine during all tests.

I tested against a Cisco 4948 switch and a Cisco 2811 router. All tests were done using SNMPv3.