Sudden loss of network connectivity

[MrChao]

Overnight, my Proxmox node stopped sending network traffic.

I hadn’t made any changes, maybe there was some update, but the web admin is inaccessible, as well as SSH.

Booting the system to a live Ubuntu image shows that the port will get a IP address and network services function normally. My router isn’t blocking the traffic, and has given the NIC the reserved IP address.

However, ip a shows that, except for local host, no interface gets an IP, even when switching the cable to other ports.

Is there a way to rebuild the network?

 

[sw-omit]

That ip a you mention, is that from the live-iso still, or already "back" in Proxmox
Try running the following commands when back in Proxmox and either post a photo of it, or try to save the output to an USB and copy it here (within [CODE][/CODE] tags please) from a working PC

cat /etc/network/interfaces
ifreload -a
ip a

 

[muhammadnourman]

i have same issue like this in realtek nic r8168, i installed latest version proxmox 8.2

 

[kevin.inner]

I encountered the same issue yesterday. It looks like after the latest Debian updates somehow the networking service got disabled. In my case, the fix was to simply re-enable it.

Check the status of the networking service:

sudo systemctl status networking.service

Re-enable it:

sudo systemctl enable networking.service

 

[MrChao]

Just want to say, I love how responsive people are and how much everyone is willing to help out on this forum.

I suppose it could be helpful to mention my equipment.

The machine is a Minisforum MS-01, I am connected on one of the Intel X710 using a Cable Matters 10Gb SFP+ module–model# 104068.

I had previously tried swapping SFP+ ports and modules and was unable to get Proxmox to report an IP on either. The other 2 ports are Intel i226-V and i226-LM, I believe I tried the i226-V port, but they're unmarked so I'm not exactly sure.

I did run Memtest, which passed.

As one would expect, the NIC's MAC address is not showing up in my router's DHCP table.

@sw-omit -

The machine was only internally recognizing an IP and showing normal network traffic when booted to the Ubuntu live image.

cat /etc/network/interfaces


ifreload -a


ip a


@kevin.inner -

This is my output for the networking status:

sudo systemctl enable networking.service

 

[sw-omit]

I think I might see at least part of the issue, the names of your devices changed from
enp8s0f0
to
enp8s0f0np0
and the same for enp8s0f1 to enp8s0f1np1, most likely after an update where it now get more details about the network-devices. Recently with the 8.1 to 8.2 update I and quite a few other people had the same issue.

Change the names in the /etc/network/interfaces config, then use an ifreload -a or a server-reboot

There might still be errors, but we can fix those after, as well as that it's generally strongly advised to NOT use DHCP for the server itself, only static IPs

 

[MrChao]

I think I might see at least part of the issue, the names of your devices changed from
enp8s0f0
to
enp8s0f0np0
and the same for enp8s0f1 to enp8s0f1np1, most likely after an update where it now get more details about the network-devices. Recently with the 8.1 to 8.2 update I and quite a few other people had the same issue.

Change the names in the /etc/network/interfaces config, then use an ifreload -a or a server-reboot

There might still be errors, but we can fix those after, as well as that it's generally strongly advised to NOT use DHCP for the server itself, only static IPs

As suggested, I corrected both interface names and rebooted but still unable to get to the web admin, ssh, or ping to/from the machine.

ifreload -a finishes with no messages.

ip a shows no IP on the NIC–no other ethernet cables connected.

Understand about DHCP. My reasoning for enabling it was because my router will report the device connected much sooner if it is requesting an IP–which I had a static assignment for. I switched back to static.

I had noticed the interface ids changed. That is a pretty big change to let happen in an update. But this is a thing I have noticed about Proxmox, that generally devices don't enumerate consistently. Hopefully this update addresses some of that.

Thanks so much for the help!

 

[sw-omit]

The possible changing is the exact reason I now for each install force my own (accepted) naming context for the devices as one of the firs things to set up.

Could you try running the command ifup enp8s0f0np0 for all of the physical devices as well as vmbr0, and then run ip a again to see if any devices came "up" at least on a physical elvel?

 

[MrChao]

Here's the output from ip a after running ifup on all of the interfaces.

The only difference is inet6 coming up on enp93s0–which is the NIC that is plugged into the router.

 

[sw-omit]

For line 4 and 5 it now does have the "no carrier" tag, like it is not detecting a cable in that port, but since you said it worked on a live-iso, it doesn't seem to be JUST a hardware issue.

Since the enp93s0 does see a active connection, maybe you can, at least temporarily, configure the vmbr0 to that port, so you at least get access back (especially through SSH for easier copy/paste and the like)?

Also, could you run the command ethtool enp8s0f0np0 to verify the no-carrier status of the port/device (preferably once you hopefully got SSH back and can just copy/paste the output within code-tags here )?

 

[MrChao]

Okay, I'm back in.

I'm not sure how /etc/network/interfaces got modified so specifically but I was pretty sure that interface enp93s0 was what I had connected to the router/LAN. The settings I had applied to it were somehow applied to enp8s0f0np0. I was doubting my memory but I am now sure that enp93s0 had been listed as the first entry in /etc/network/interfaces, I had commented this position in the file as the active network port and that also did not move from that position in the file, but somehow enp8s0f0np0 had swapped positions with enp93s0 but that position retained the the original addressing. So, effectively, a manual address was being assigned to an empty NIC port.

Anyway, after reentering the original addressing for enp93s0 and running ifreload -a all is well–Web admin, ping, ssh.

...I'll be adding enumerated ID labels to the physical ports.


Thanks so much for the help!!

 

[MrChao]

Spoke too soon, I guess.

Now I can't add the Linux Bridge back to the interface. When I do I get:

iface enp93s0 - ip address can't be set on interface if bridged in vmbr0 (500)

 

[MrChao]

Got that last one settled too. I think whatever update that was installed also broke my use of a single port for connectivity and Linux Bridge.

I broke down and assigned the bridge to a second port.