[LU-8782] queue depth too large: 63 (<=16 wanted)

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.7.0
• Type: Bug
• Priority: Minor
• Reporter: Mahmoud Hanafi
• Assignee: Amir Shehata (Inactive)
• Resolution: Not a Bug
• Severity: 3

Description

We would like to get clarification and verify configuration using patch LU-3322.

We are running 2.7.2 with patch from LU-3322. When configureing two hosts with different peer_credits we get this error when the initial connection is made.

LNetError: 81:0:(o2iblnd_cb.c:2325:kiblnd_passive_connect()) Can't accept conn from xxx.xxx.xxx.xxx@o2ib, queue depth too large:  63 (<=16 wanted)

Although we get the error lnet_selftest is working.

Host1:

options ko2iblnd timeout=150 retry_count=7 peer_timeout=0 peer_credits=63 concurrent_sends=63

Host2

options ko2iblnd timeout=150 retry_count=7 peer_credits=16 concurrent_sends=63

Attaching debug output from HOST2

Comments

[Joseph Gmitter (Inactive)]

Hi Doug,

Any advice for this issue?

Thanks.
Joe

[Doug Oucharek (Inactive)]

Looking at the code for LU-3322, when a connection attempt is rejected due to a queue depth being too big, a new connection attempt should be done at the lower queue depth value.

So, when you see this error message, do the two nodes still end up with a connection and everything works fine? If so, then we should just "turn-down" this error message to just be informational.

[Mahmoud Hanafi]

Please close this case. This error is well understood.

[Minh Diep]

Thanks