I have so far used pfSense on 2 bare metal platforms and in a VM on ESXi. Now I have migrated the VM to XCP-NG using Intel e1000 (not RealTek) and disabled the TX checksumming on the interface. The VM is working fine.
I am monitoring pfSense instances via the SNMP agent on pfSense, I do not wish to use the Zabbix agent on pfSense.
So far this had worked fine. Now I am having the following problem with the migrated VM, now running on XCP-NG:
The VIFs show a speed item value of “10M” instead of “1G”.
I already deleted and rediscovered the corresponding interface items, but they still show the wrong value.
I know that I can disable the items and/or corresponding triggers in Zabbix, but I would like to avoid this because then I am losing the monitoring of the interface speed.
Any ideas why this happens and how it can be fixed?
PS: I remember I had the same problem with Zabbix monitoring TrueNAS Core on bare metal for a 10Gbps interface and the story was that the SNMP agent for FreeBSD has a bug being unable to work with higher speed values. But this cannot really be the same problem here, as 1G was not a value that would run into this bug.
I have found further instances of this problem when I ran the Zabbix Agent2 container in network host mode.
Linux VMs on XCP-ng have now way of determining the speed of the virtual network interface.
This is an issue due to the Xen pararvirtualization and may also be the reason why SNMP agent has problems determining the interface speed in FreeBSD VMs on XCP-ng.
Is there a way to solve this? I already run the XCP guest tools in the VMs, but this does not seem to solve it for the ethX ir igbX interfaces. Do I need to use different interface names (that have a better para interface) for better results?
Or what workaround makes sense?
The simplest workaround is just disabling the speed-based triggers. I would rather not lose the bandwith alerts.
Is there a way to override the discovered agent-based item with an item that uses a static value?
(In an ideal world the Linux Agent template would know about this and do this adaption on its own, the issue is known since 3 years, but nothing has happened to improve the template in that regard.)
Does no-one here use the Zabbix Agent(2) container (in network host mode!) in their XCP-ng VMs?