ifAliases not being retrieved anymore

[ae0]

Hi all,

I'm running two installs of 0.8.7d (old, I know) monitoring 10 Juniper nodes: an MX80, an MX240, and eight MX960's. We just started having a problem where the ifAliases are no longer being retrieved from two of the 960's on both cacti installs.

Things I've checked/tried:
- adjusting all the timeouts and intervals
- one of the affected 960's has about 1900 interfaces (units, specifically) being monitored, and the other has around 600. We have nodes with around 1500 that are still working correctly.
- ran an snmpwalk on the nodes via command line, which worked correctly
- I went through the unit descriptions, to check for any invalid characters. The descriptions are not absurdly long either.
- Verified that I can see the ifAliases from another SNMP monitoring system (non-cacti)
- Google searches haven't turned up anything

When I run the verbose query, I get this:

Code: Select all

<snip>
+ Found item [ifName='xe-0/0/0.1046'] index: 282252 [from value]
+ Found item [ifName='xe-0/0/0.1174'] index: 282279 [from value]
+ Found item [ifName='ge-5/3/4.251'] index: 282312 [from value]
[b]+ Located input field 'ifAlias' [walk]
+ Executing SNMP walk for data @ '.1.3.6.1.2.1.31.1.1.1.18'[/b]
+ Located input field 'ifType' [walk]
+ Executing SNMP walk for data @ '.1.3.6.1.2.1.2.2.1.3'
+ Found item [ifType='ethernetCsmacd(6)'] index: 1 [from value]
+ Found item [ifType='other(1)'] index: 5 [from value]
+ Found item [ifType='softwareLoopback(24)'] index: 6 [from value]
<snip>

Any ideas what this might be? Are there any limitations or special characters that would cause that part of the query to fail? Upgrading isn't really an option for me at this point, sorry!

Thanks for any help!

Casey

[cigamit]

When doing a manual walk of the OID, how long does it take to respond? And what is your snmp timeout for those hosts set at? I am wondering with that many objects if it is just taking too longer to return.

Other than that, it would be extremely hard to diagnose as that version of Cacti was released over 6 years ago (10 versions behind). Its hard to say if this was a bug that was found and fixed in that time or not.

[ae0]

When I do it via CLI, it runs in 4.3 seconds. From when I click "Verbose query" to the time the page has fully loaded, it's 25 seconds.

The node settings are SNMP Timeout: 500 and Maximum OID's 10. These are the same on all the nodes. It's running cmd.php for the poller, not spine. It's set for 20 max concurrent processes. When I do the verbose query, all the items after the ifAlias are still fetched correctly (ifType, ifSpeed, etc).

Unfortunately, doing an upgrade isn't really an option for me at this point

When doing a manual walk of the OID, how long does it take to respond? And what is your snmp timeout for those hosts set at? I am wondering with that many objects if it is just taking too longer to return.

Other than that, it would be extremely hard to diagnose as that version of Cacti was released over 6 years ago (10 versions behind). Its hard to say if this was a bug that was found and fixed in that time or not.

[ae0]

I just did some packet captures against one of the nodes. I can see that both instances of cacti do manage to retrieve around the first 1500 ifAliases. Then curiously, both instances get hung up at the same ifIndex number. They send one retry after about 500ms, then it appears that 500ms later, it gives up and moves on to the ifTypes.

I then did the same capture to the other node. Again, they get the first 600 ifAliases, then they both hang on the same ifIndex number before moving on.