How to debug "output: U"

[baroo]

Hi

I have a strange problem with cacti. On some data sources I recived from the poller something like this:

Code: Select all

07/20/2005 05:35:07 PM - CMDPHP: Poller[0] Host[4] DS[10] SNMP: v2: galeria.inet.one.pl, dsname: conntrack_states, oid: .1.3.6.1.4.1.3000.1.100.1, output: U  
07/20/2005 05:35:07 PM - CMDPHP: Poller[0] Host[4] DS[10] WARNING: Result from SNMP not valid. Partial Result: 

How can I debug this query to find out what is wrong and I recive output: U?

Thanks for any advices.

[BSOD2600]

Use something like Getif or net-snmp snmpwalk to find out what the correct OID is.

[baroo]

This is the correct OID. When I use snmpwalk everything work perfect (cacti worked perfectly too before I upgrade it from 0.8.6b to 0.8.6f). The main problem is that some of query`s (totaly randomly) return output: U. When I use tcpdump I can see that remote machine return value but cacti doesn`t save them. There is something very strange: this is often the second of query (but this is not a rule). My internet connection work fine, so this is not a hardware problem.

[BSOD2600]

1) What method did you use when upgrading cacti?
2) Paste the output from a snmpget for that OID.

[TheWitness]

Increase snmp timeout for host

[baroo]

Increase snmp timeout for host

This was the real problem (stupid problem). Thanks.

[eddievenus]

I have the same problem, and I do not want to increase the timeout too high. After all some devices still do time out. So what are appropriate timeouts for this. Default being 500, I have been setting it to 1000.

How much higher should it go before it works? Obviously this is an unfair question, you could not know for sure how long to wait, but what is a good guess? What would the upper bounds be for this timeout? 2 seconds? 3? 5?

I plan to test this on my devices, but I am new to this, and if I am setting it too high I might run into problems with polling times getting too long. I want to know a max timeout value.

Thanks

[TheWitness]

If you are polling windoz hosts for disk stats, because of a faulty snmp agent on them, the A: drive is polled when you perform 5 minute polls. Therefore, you need from 5-8 second timeout unless you remove the A: drive, then it operates pretty good.

Otherwise, you have to take into the consideration of where the host is. If you are in New York, and you are polling a device in China that happens to be on a dialup network, you definately should have a 2-3 second timeout. On the other hand, if you are polling a device on a LAN from the same LAN 15 seconds might do.

Also, keep in mind that SNMP is based upon the UDP protocol, which is stateless. So you never know. If have found it to be pretty reliable on a good network LAN or WAN.

TheWitness