[Lustre-discuss] Lustre Falling over

[Wojciech Turek]

Hi,

Yes I came with same conclusion. I have tried to apply patches from  
this bug but patching process has FAILED.
[lbuild at mds01 lustre-1.6.3]$ patch -p0 < ../../../patch_13540
patching file lustre/ldlm/ldlm_request.c
Hunk #2 succeeded at 772 (offset 2 lines).
Hunk #4 succeeded at 866 (offset 2 lines).
Hunk #6 succeeded at 989 (offset 46 lines).
Hunk #7 FAILED at 1061.
Hunk #8 FAILED at 1134.
Hunk #9 succeeded at 1169 (offset 36 lines).
Hunk #10 succeeded at 1228 (offset 46 lines).
2 out of 10 hunks FAILED -- saving rejects to file lustre/ldlm/ 
ldlm_request.c.rej


[lbuild at mds01 lustre-1.6.3]$ cat lustre/ldlm/ldlm_request.c.rej
***************
*** 1005,1012 ****
                             int count, int max, int flags)
   {
           cfs_time_t cur = cfs_time_current();
-         struct ldlm_lock *lock, *next;
-         int rc, added = 0, left;
           ENTRY;

           spin_lock(&ns->ns_unused_lock);
--- 1061,1068 ----
                             int count, int max, int flags)
   {
           cfs_time_t cur = cfs_time_current();
+         struct ldlm_lock *lock;
+         int added = 0;
           ENTRY;

           spin_lock(&ns->ns_unused_lock);
***************
*** 1078,1107 ****
           spin_unlock(&ns->ns_unused_lock);

           /* Handle only @added inserted locks. */
-         left = added;
-         list_for_each_entry_safe(lock, next, cancels, l_bl_ast) {
-                 if (left-- == 0)
-                         break;
-
-                 rc = ldlm_cli_cancel_local(lock);
-                 if (rc == LDLM_FL_BL_AST) {
-                         CFS_LIST_HEAD(head);
-
-                         LDLM_DEBUG(lock, "Cancel lock separately");
-                         list_del_init(&lock->l_bl_ast);
-                         list_add(&lock->l_bl_ast, &head);
-                         ldlm_cli_cancel_req(lock->l_conn_export,  
&head, 1);
-                         rc = LDLM_FL_LOCAL_ONLY;
-                 }
-                 if (rc == LDLM_FL_LOCAL_ONLY) {
-                         /* CANCEL RPC should not be sent to server. */
-                         list_del_init(&lock->l_bl_ast);
-                         LDLM_LOCK_PUT(lock);
-                         added--;
-                 }
-
-         }
-         RETURN(added);
   }

   /* when called with LDLM_ASYNC the blocking callback will be handled
--- 1134,1140 ----
           spin_unlock(&ns->ns_unused_lock);

           /* Handle only @added inserted locks. */
+         RETURN (ldlm_cancel_list(cancels, added));
   }

   /* when called with LDLM_ASYNC the blocking callback will be handled
[lbuild at mds01 lustre-1.6.3]$

Any idea why? Can I get help with that?

Regards,

Wojciech Turek
On 11 Jan 2008, at 18:29, Brian J. Murrell wrote:
> On Fri, 2008-01-11 at 07:18 -0500, Charles Taylor wrote:
>> Were are running a patched (mballoc and others) 1.6.3 lustre on
>> x86_64 platform under Centos4.5.   Kernel is ...
>
> This looks like bug 13917.
>
> b.
>
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss at clusterfs.com
> https://mail.clusterfs.com/mailman/listinfo/lustre-discuss