POLLER: Poller[0] WARNING: Poller Output Table not Empty

[gandalf]

Would switching to php-snmp make our polling any faster? According to our Cacti Poller statistics, we have:
SNMP - 5820
Scripts - 125
Script Server - 835

No, it won't. Spine is using it's own snmp code and does not refer to php-snmp.
R.

[stormonts]

Ok. Can you comment on my other question, regarding the appearance of all of the items with "status_fail_date='2013-04-27 13:47:00'" in them?

I have verified that there is only one poller running, and know that this issue doesn't always occur (best guess is that it happens when there is a high network load/load on one server). As a next better step to resolve the issue, is it better to increase the SNMP timeout for all devices or to lower the maximum SNMP get request?

[gandalf]

I know that it needs to be below 60 seconds, but I can't figure out what is tripping it up. I have increased the log level to "Debug" level but am not seeing any errors/timeouts or anything like that.

I have followed the links in the signature as well, but none of the suggestions helped.

Currently, the SNMP timeout is set to 1500 for all devices. Should that number be reduced because each device is possibly waiting 1.5 seconds for responses? (or I am misunderstanding that?)

This may happen in cases, where more devices than usual need a poller retry or go into host timeout.
This way, the poller process will last longer, as Cacti makes sure, that the SNMP (script) timeout is exhausted and will retry this e.g. for 3 times. This last WAAAY longer then simply fetching valid data.
So you may watch out for unresponsive hosts.
A similar effect is seen in cases where scripts are used. In case you have some external command without using a narrow timeout, this may stop the poller process for a significant amount of time.

R.