rdma issue

[15810064471]

Background: using nvme on GPU machine to build beegfs
One management node, two metadata nodes, and several storage nodes (the storage nodes support the RDMA network roce V2 (Ethernet bearer)). After deployment, the following error messages will be sent to start TCP transmission

The configuration file is as follows:

[iamjoemccormick]

Hello,

I converted this issue to a discussion for now, since typically RDMA fallbacks are due to issues outside BeeGFS itself. To investigate further we would need:

(1) Any relevant logs from clients and servers that are unable to establish connections using RDMA.

(2) Output of beegfs health net from a client that is unable to connect using RDMA.

(3) Output of beegfs node list --with-nics (optionally truncated to only include the servers and the client selected for (2)).

(4) Output from beegfs node ping.

All output would need to be collected while the system is showing fallbacks to TCP.

Note: This discussion forum is primarily community-driven. Maintainers may occasionally comment to clarify behavior or help determine whether something could be a bug, but support here is provided on a best-effort basis. If you require guaranteed assistance, please review our support policy.