[lustre-discuss] Question about Network Interface Flapping (Down/Up) on Lustre Clients in ROCE Environment

[王烁斌]

Hi~



My software environment is Lustre 2.15.8. The client uses ConnectX‑5 network adapters, and the server uses E810 network adapters. The switches are configured for RoCE, with the priority set to 3 only.


Under normal conditions, all service traffic goes through queue 3, as shown in the figure below.

 
However, after a port flap (down/up event) on a client network interface, this interface will use both queue 0 and queue 3, while other interfaces remain normal, as shown in the figure below.

 
At the same time, a strange issue occurs on the Lustre client: it incorrectly adds the flapping network interface to the peer list. Additionally, running lustre_rmmod hangs because there are still busy services in use.
 
Is this a Lustre‑related issue? If so, how should I resolve it? Please provide me with some troubleshooting ideas.


so，How should I locate the problem？


Thanks！
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20260209/10d15fb7/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 11277 bytes
Desc: not available
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20260209/10d15fb7/attachment-0003.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 13053 bytes
Desc: not available
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20260209/10d15fb7/attachment-0004.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 59969 bytes
Desc: not available
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20260209/10d15fb7/attachment-0005.png>