[Lustre-discuss] target_send_reply_msg errors

[David Dillow]

On Thu, 2010-12-02 at 17:56 -0800, Andrus, Brian Contractor wrote:
> I am seeing a TON of messages like:
>  
> LustreError: 19122:0:(ldlm_lib.c:1892:target_send_reply_msg()) @@@ processing error (-107)  req at ffff8105cbdd2c00 x1348234052426436/t0 o400-><?>@<?>:0/0 lens 192/0 e 0 to 0 dl 1291340852 ref 1 fl Interpret:H/0/0 rc -107/0
> LustreError: 19122:0:(ldlm_lib.c:1892:target_send_reply_msg()) Skipped 864 previous similar messages
> LustreError: 9363:0:(mgs_handler.c:554:mgs_handle()) lustre_mgs: operation 400 on unconnected MGS
> LustreError: 9363:0:(mgs_handler.c:554:mgs_handle()) Skipped 352 previous similar messages
>  
> Any ideas what that may be and how to remedy it?

Check that 'lctl get_param timeout' gives you the same value on the
servers and the clients.

We've seen this quite often -- when using a separate MGS, the timeout
doesn't get set automatically from the parameters like it does for the
MDT and OSTs. We added it to our filesystem start scripts and things are
much happier.


-- 
Dave Dillow
National Center for Computational Science
Oak Ridge National Laboratory
(865) 241-6602 office