[LU-14113] lnet no credits should be a warning message

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.12.5
• Type: Improvement
• Priority: Minor
• Reporter: Mahmoud Hanafi
• Assignee: Cyril Bordage
• Resolution: Fixed

Description

Current you must have +lnet debugging enabled and dump logs to see if there is an issue with running out of credits. Because running out lnet credits can cause failures at many levels we recommend that the "no credits" should be changed to a warning that is printed to the console.

 

--- a/lnet/klnds/o2iblnd/o2iblnd_cb.c
+++ b/lnet/klnds/o2iblnd/o2iblnd_cb.c
@@ -896,7 +896,7 @@ __must_hold(&conn->ibc_lock)
        }
 
         if (credit != 0 && conn->ibc_credits == 0) {   /* no credits */
-                CDEBUG(D_NET, "%s: no credits\n",
+                CWARN( "%s: no credits\n",
                        libcfs_nid2str(peer_ni->ibp_nid));
                 return -EAGAIN;
         }

Comments

[Peter Jones]

Cyril

Could you please assist?

Thanks

Peter

[Andreas Dilger]

Mahmoud, have you tested this patch on your system? I would be worried that it could be very verbose on a heavily-loaded system. However, if you have this patch in production on your network and it is not generating a ton of noise, then it seems unlikely to cause problems elsewhere.

[Mahmoud Hanafi]

You are correct. This patch is very verbose. We'll need some other of determining when credit starvation results in dropped or delayed messages. 

[Mahmoud Hanafi]

We can close this