We want to reconfigure our lustre for multi-homed network. However the 
mgs only appears to respond to clients using the original network (02ib).

Here is the kerenel module (modprobe.conf) line
 > options lnet 'networks="o2ib(ib0),tcp(eth0)"'

When mounting lustre on the new (tcp) network, we see the following 
syslog messages:
 > kernel: LustreError: 10899:0:(events.c:454:ptlrpc_uuid_to_peer()) No 
NID found for 10.2.1...@o2ib
 > kernel: LustreError: 
10899:0:(client.c:58:ptlrpc_uuid_to_connection()) cannot find peer 
10.2.1...@o2 ib!
 > kernel: LustreError: 10899:0:(ldlm_lib.c:321:client_obd_setup()) 
can't add initial connection
 > kernel: LustreError: 
8819:0:(connection.c:144:ptlrpc_put_connection()) NULL connection
 > kernel: LustreError: 10899:0:(obd_config.c:332:class_setup()) setup 
lustre0-MDT0000-mdc-ffff81022ea1 cc00 failed (-2)
 > kernel: LustreError: 
10899:0:(obd_config.c:1071:class_config_llog_handler()) Err -2 on cfg 
command:
 > kernel: Lustre:    cmd=cf003 0:lustre0-MDT0000-mdc 
1:lustre0-MDT0000_UUID  2:10.2.1...@o2ib
 > kernel: LustreError: 15c-8: mgc10.1.1...@tcp: The configuration from 
log 'lustre0-client' failed (-2).

After viewing syslogs and searching mailing lists, it appears the mdt 
needs to be reformatted to include the new network (tcp) routing to the mgs.

We have a combined mgs+mdt, My best guess, the upgrade command on the 
mds is:
tunefs.lustre --mgsnode="i...@o2ib0,i...@tcp0" dev_mdt

The original command to make the mdt did not include the mgsnode flag:
tunefs --dryrun -l dev_mdt
 >Reading CONFIGS/mountdata
 >
 >   Read previous values:
 >Target:     lustre0-MDT0000
 >Index:      0
 >Lustre FS:  lustre0
 >Mount type: ldiskfs
 >Flags:      0x5
 >              (MDT MGS )
 >Persistent mount opts: errors=remount-ro,iopen_nopriv,user_xattr
 >Parameters: mdt.group_upcall=/usr/sbin/l_getgroups

Q. Is tunefs.lustre command this non-destructive ?
tunefs.lustre --mgsnode="i...@o2ib0,i...@tcp0" dev_mdt

Q. Do I need to update config on the OSTs for our new multi-homed network ?

Thanks,
Chris Hunter


_______________________________________________
Lustre-discuss mailing list
Lustre-discuss@lists.lustre.org
http://lists.lustre.org/mailman/listinfo/lustre-discuss

Reply via email to