Random 10 minute network timeouts

[validtimeofday]

Hi everyone,

I recently deployed Proxmox on my home server and have provisioned LXCs for Plex, Samba and Apache. All was working well for a day or two until Plex suddenly cut out while in use, I did a hard reset at the server and it continued working again for about 20 minutes until it cut out again and I gave up. I tried again another day with a 2 hour movie, where it played fine, but then another day it happened again after 50 minutes, and then instantly again after a hard reset.

After the first cut out I scaled back the resources assuming that the CPU was getting hit hard, despite the graphs looking fine, but the issue has persisted.

Looking at the global log it looks like the network interface times out and then takes a few minutes to get going again. The same hardware, network switch and drives where all running fine on a Ubuntu server before this deployment so I feel as if this isn't a hardware issue, but I won't rule it out after seeing a hardware call trace. Looking at the logs, has anyone seen this behavior before?

 

[shrdlicka]

Hi,

can you check what driver is used for your network interface:

ethtool -i <iface name>

 

[validtimeofday]

$ ethtool -i enp3s0f0
driver: tg3
version: 5.15.30-2-pve
firmware-version: 57766a-v1.15
expansion-rom-version:
bus-info: 0000:03:00.0
supports-statistics: yes
supports-test: yes
supports-eeprom-access: yes
supports-register-dump: yes
supports-priv-flags: no

 

[shrdlicka]

Looks like there is a but report [1] for the kernel. What I read the ethtool bugfix didn't help most people. But you can try it:

ethtool -K <iface name> sg off
ethtool -K <iface name> tso off
ethtool -K <iface name> gso off

[1] https://bugzilla.kernel.org/show_bug.cgi?id=12877#c32

 

[bobmc]

I believe the tg3 driver is buggy on Linux, I've seen comments that disabling IOMMU on the host (if it's enabled) helps

 

[validtimeofday]

I've since installed vanilla Debian on the same hardware and I can sadly replicate it. Looks like a hardware issue. Thank you though, I appreciate the support regardless!