On Tue, May 25, 2010 at 09:34:48PM +0200, Sigurd Mytting wrote:
Interfaces are missing and all switch ports have a port number, screenshot from one of the 2960 switches:
Hi again, sorry for the late reply; we've been quite busy with NAV 3.6 lately, and since we're replacing gDD, it hasn't been given much priority lately.
Found this several times in /var/log/nav/getDeviceData/getDeviceData-stderr.log:
Want to delete port: 820, 1, 00a-sw-b3f44.vegvesen.no Want to delete port: 820, 2, 00a-sw-b3f44.vegvesen.no Want to delete port: 820, 2, 00a-sw-b3f44.vegvesen.no Want to delete port: 820, 1, 00a-sw-b3f44.vegvesen.no
(the same 2960 as in the screenshot above)
From getDeviceData.log it looks like data is collected and some of the ports dropped, did a grep on this device in todays log:
Well, my gut tells me this has something to do with a port number conflict (which is why I hate the concept of "port numbers" and why we've been trying to reduce its importance in NAV).
Seeing your gDD log and the final result, it seems it wants to assign port number 2 and 3 to two ports each: Fa0/1 and Gi0/1 both want port number 1, and Fa0/2 and Gi0/2 both want port number 2. It all depends on which MIB we're querying.
And so it seems the Gigabit ports are winning the conflict.
Using standard MIBs only, it would be Gi0/1 = 1, Gi0/2 = 2, Fa0/1 = 3, while Cisco MIBs will say Fa0/1 = 1, Fa0/2 = 2, Gi0/1 = 49, Gi0/2 = 50.
I haven't actually seen a conflict like this turn into missing ports before, but it is likely that some data structure in gDD still identifies each port by its "number". Your logs certainly indicate that it thinks there are duplicate ports, based on port numbers, and its deleting something it shouldn't
This should be reported as a bug: https://bugs.launchpad.net/nav/+filebug
It seems like all my switches have the ifAlias oidkey.
I've had another user complain about the exact same portname problem on IRC just last week. We debugged it to no avail, but it seems only a deep dive into the murky gDD code will help. Not sure whether he reported this as a bug, though...