Hello, we use LUSTRE with AUTOFS to circumvent (hopefully) the evicted client statahead problem and disconnect lustre client when not in use. We get the following messages just before the client is unmounted. It is reproducable and happens every unmount. The functionality of further client mounts and unmounts is not affected i.e. lustre keeps on working nice and smoothly. Is this just normal lustre behaviour before unmounting a client ? So nothing to worry about ? Or is this an indicator for something broken or ill defined ? Thanks and Regards Heiko <snip> Aug 18 09:12:36 quadcore12 Lustre: Client scia-client has started Aug 18 09:13:48 quadcore12 LustreError: 32585:0: (ldlm_request.c:986:ldlm_cli_cancel_req()) Got rc -108 from cancel RPC: canceling anyway Aug 18 09:13:48 quadcore12 LustreError: 32585:0: (ldlm_request.c:986:ldlm_cli_cancel_req()) Skipped 3 previous similar messages Aug 18 09:13:48 quadcore12 LustreError: 32585:0: (ldlm_request.c:1575:ldlm_cli_cancel_list()) ldlm_cli_cancel_list: -108 Aug 18 09:13:48 quadcore12 LustreError: 32585:0: (ldlm_request.c:1575:ldlm_cli_cancel_list()) Skipped 3 previous similar messages Aug 18 09:13:48 quadcore12 Lustre: client ffff8102229e4800 umount complete <snap> Lustre 1.6.5.1 source compilation against patched 2.6.22.19 vanilla-kernel. ./configure --disable-liblustre --with-linux=/usr/src/linux-2.6.22.19 MDS setup: mkfs.lustre --fsname=scia --mdt --mgs --failnode=mds2 /dev/drbd0 OST(s) setup: mkfs.lustre --param="failover.mode=failout" --fsname scia --ost --mkfsoptions=''-i 2097152 -E stride=16 -b 4096'' --mgsnode=mds1 at tcp0 /dev/sd{x}
On Aug 21, 2008 08:50 +0200, Heiko Schroeter wrote:> We get the following messages just before the client is unmounted. It is > reproducable and happens every unmount. > > Is this just normal lustre behaviour before unmounting a client ? So nothing > to worry about ? > > <snip> > Aug 18 09:13:48 quadcore12 LustreError: 32585:0: > (ldlm_request.c:986:ldlm_cli_cancel_req()) Got rc -108 from cancel RPC: > canceling anywayYes, this is a known harmless problem. The client is disconnected from the MGS before it has dropped the configuration lock. This is not a problem and it was decided that it wasn''t a blocker for the 1.6.5 release. Cheers, Andreas -- Andreas Dilger Sr. Staff Engineer, Lustre Group Sun Microsystems of Canada, Inc.