Problem with unavailable SNMP device

[Markus]

It is possible to run into a rare problem when an SNMP monitored host temporarily becomes unavailable. Zabbix response when a host becomes unavailable is to regularly try the first of this host's items to check whether the host is available again. In the specific circumstances that the first item is an SNMP value which is not supported by the monitored host's MIB, Zabbix will not detect when this host becomes available again. This is because a query for a not supported MIB value will simply time out and therefore Zabbix thinks that this host is still unavailable. In turn a Zabbix agent would return 'ZBX_NOT_SUPPORTED' which at least tells the Zabbix server that the monitored host is available again.
This is probably not a problem for most Zabbix users but worth remembering if it happens. I only found out through network sniffing with tcpdump.

Markus

[Alexei]

Some time ago I was getting many complains from users that ZABBIX sets status to UNSUPPORTED in case of a single SNMP get fail (it was implemented this way initially). That's why I decided to change the logic to give ZABBIX a chance of getting the failed value.

I still think that the initial implementation was better and safer. I'm going to change it in 1.4.1.

[mucknet]

Some time ago I was getting many complains from users that ZABBIX sets status to UNSUPPORTED in case of a single SNMP get fail (it was implemented this way initially). That's why I decided to change the logic to give ZABBIX a chance of getting the failed value.

I still think that the initial implementation was better and safer. I'm going to change it in 1.4.1.

I think I recall that issue. I was very glad when you changed it.

Many SNMP OID's are dynamic, they come and go as the system changes (ie: as databases come online, JVM's created/dropped, or as vlans are created or dropped, etc). If an item is available for a while, and then disappears (In our case, I think it was while the particular DB was reloaded), it wouldn't ever try polling it again, because the SNMP agent had returned "unsupported". It required noticing that the item had stopped polling (which of course you only notice that at the time you need it), and then manually re enabling it.

If you change it back, please also add an option to try repolling unsupported items at X interval. I, for one have built a monitoring infrastructure around the way zabbix responds to SNMP, so removing a feature in a .1 release would throw me off. Maybe its best for the next major release of zabbix?

What about a seperate class for temporarily unavailable SNMP items, which would be checked periodically, since SNMP items come and go frequently?

Edit: I just saw "refresh unsupported items" in the 1.4 Manual.

Thanks Alexei!

[Alexei]

If you change it back, please also add an option to try repolling unsupported items at X interval. I, for one have built a monitoring infrastructure around the way zabbix responds to SNMP, so removing a feature in a .1 release would throw me off. Maybe its best for the next major release of zabbix?

Right, we already support this functionality! So, I think it would be pretty safe to revert back to UNSUPPORTED way of handling SNMP errors.